Wednesday, February 07, 2007

Intel's 2007 Processors: Mist In The Morning Sun

After more than two months of speculation about Intel's 2007 processor offerings it seems that the mist is finally clearing. However, what is left behind is much less than was suggested. It is now clear that the speculation greatly overestimated what Intel would be capable of delivering.

Most of the talk about Intel's 2007 processor offerings grew out of two sources: the VR-Zone roadmaps and glowing reports about Intel's 45nm process. According to VR-Zone, Intel was to release the 45nm quad core Yorkfield chips in Q3 07 along with the 45nm dual core Wolfdale. As seen in the chart, Yorkfield would clock from 3.46ghz to as high as 3.73Ghz. Likewise, Wolfdale would clock from 3.5Ghz to as high as 4.0Ghz. Clock speeds that high would have surpassed anything AMD could have offered and given Intel a tremendous lead. Any scepticism about these clock numbers was quickly brushed aside as people waxed almost poetic about Intel's 45nm process. Supposedly it would have allowed huge increases in clock speed while also drawing less power. The typically stated cornerstone of Intel's 45nm process was high-K, or at least, this was the only item that Intel had that AMD did not. Adding to this, AMD's 65nm Brisbane chips were not readily available in the channel in Q4 06. When they did become available in the channel they were at lower clocks than the existing 90nm chips. So, the chatter immediately moved to talk about problems with AMD's 65nm process and even speculation that Barcelona wouldn't arrive until 2008. Not surprisingly, when Intel recently announced that it had actual working chips with high-K, the press reports were overflowing with praise. The second half of 2007 looked to many like it would be a landslide success for Intel with performance even further ahead of Barcelona than C2D was ahead of K8.

However, in the latest roadmaps, all of that is gone; Intel's amazing 45nm speed has completely vanished. The highest clocking Yorkfield is now only 2.4Ghz and Wolfdale tops out at 3.0Ghz. Worse still, even with these reduced speeds, 45nm has been pushed back to Q1 08. And, adding insult to injury, even Anandtech now agrees that AMD's Barcelona quad core will begin production in Q3 07 at 2.5Ghz. This makes sense for a lot of reaons however it seems that history and common sense are often casulties of processor enthusiasm. For example, when Core 2 Duo arrived, it probably seemed that Intel's Midas Touch had returned. The problem with this view however was that C2D's improvement was not unique. AMD had a similar increase in IPC with K8 and it now appears that Barcelona will have a similar increase in IPC over K8. SSE was the only area where C2D saw phenomenal increases in speed and this was done by doubling the width of all the necessary hardware. However, Barcelona's FP/SSE hardware will receive a similar stretch in width. If Barcelona is any slower than C2D at the same clock, it won't be by much. Rather than just being a simple upgrade, it has now been indicated that K8 was completely resurveyed and many improvements were made all over the architecture including branch speculation, out of order loading, stack operations, bus access scheduling, prefetch, and instruction decoding. Barcelona is therefore obviously a new core in the same way that K8 was. Presumably, Barcelona would then be the K9 or K10 core but AMD has not indicated this yet.

So, 2007 is now shaping up to be pretty much even in terms of performance between Intel and AMD. And, 2008 shouldn't be much different since IBM/AMD also have high-K and will use it on 45nm just as Intel will. The talk about what Intel would offer in 2007 looked like giants in the mist. And, now that the mist is clearing we can see that Intel's processors don't stand any taller than AMD's. It looks like we are going to have some very nice processors from both AMD and Intel in late 2007 and early 2008. It looks like both will have good Integer and SSE performance with very similar IPC and power consumption. This should be great for customers of both companies. Nevertheless, this still seems to be unacceptable to a lot of enthusiasts. Perhaps this is why the talk now seems to be about AMD's supposed poor finances and speculation about bankruptcy. I'm sorry but AMD is not going to go bankrupt while introducing a new architecture, new mobile chipset, and being able to compete better in the commercial segment with its desktop, integrated chipsets.

An interesting final note is how popular this blog is with Intel employees. For example, AMD employees are no more likely to read my blog than employees of Pratt & Whitney or General Electric. However, I get 20 times as much traffic from Intel employees. In fact, Intel employees are the single largest group of readers. I've seen this blog dismissed a number times as blatantly pro-AMD FUD by various people (such as the Tommies at ForumZ). However, I do have to wonder then, if this blog contains so little valid information, why readers from Intel keep coming back.

224 comments:

1 – 200 of 224   Newer›   Newest»
Scientia from AMDZone said...

Traffic from Intel employees is the single largest group at about 7% of the total blog traffic.

Scientia from AMDZone said...

ho ho
Actually there is an extreme edition of the Yorkfield too but there is no clockspeed specified.


Yes, I'm thinking the X could be released at 2.6Ghz. I don't see this as a big factor though. If AMD has 2.5Ghz in Q3 07 and Intel has 2.6Ghz in Q1 08 then perhaps AMD can bump up 100Mhz by then. Even if they don't, the difference in speed is only 4%. So, still basically even.

Also, does anyone remember at what speed did those CPU's run that Intel demonstrated some time ago, about a year before launch? I can't remember if they were at 2.13 or 2.4GHz.

Which cpu's are you referring to?

I've read that IBM will use High-K with 45nm but no concrete information about AMD. Could you link to some sources?

Well, this is AMD's process too; Intel and AMD have 100% reciprocity on process tech as 45nm. However, if you don't believe me:

xbitlabs

"International Business Machines, Advanced Micro Devices as well as their partners Sony and Toshiba have unveiled details concerning their jointly developed 45nm process technology. The companies, which will begin production using 45nm nodes in 2008, said they would also use high-k dielectrics, like Intel Corp. does with its 45nm fabrication process."

It might be because Intel has lots more workers than AMD.

No. Intel doesn't have 20X as many as AMD.

Ho Ho said...

Scientia
"If AMD has 2.5Ghz in Q3 07 and Intel has 2.6Ghz in Q1 08 then perhaps AMD can bump up 100Mhz by then"

Technically, Intel has had 2.66GHz for a couple of months already.

Scientia
"Which cpu's are you referring to?"


anandtech
It was 2.13GHz.

The five working systems were configured as the following:
Unknown frequency 45nm dual-core mobile processor in a notebook with Microsoft Vista running Microsoft Office 2003 applications

2.13GHz 45nm dual-core desktop processor running high definition video content (1080P) under Microsoft Vista.

1.86GHz 45nm quad-core desktop processor running Ubisoft's Rainbow Six Las Vegas game under Microsoft Vista

Two, 2.13GHz 45nm dual-core processors running Glaze Workstation application under Microsoft Windows 2000 Advanced Server

Two, 2.13GHz 45nm quad-core processors encoding a video in Adobe Premier under Microsoft Vista



Scientia
"However, if you don't believe me"

xtremetech:
"Grose said AMD was still considering whether to use a new technique developed by research partner IBM in upcoming chips.
The technique, also achieved separately by Intel, helps ensure chips can continue to shrink by using a new material that stops electricity leakage. Intel said it will use the process in its 45-nanometer chips coming out later this year.
"It really is going to depend on what makes most sense and on our product roadmap. It could be late in the 45-nanometer timeframe or the 32 node application," Grose said."


What to believe?

enumae said...

Hey Scientia, I am not sure how accurate the roadmap for desktop chips is.

I was looking at the site today and they have the Mobile roadmap posted, and well Intel has a 2.8GHz slated for Q1 08.

Maybe there is something being overlooked by PCWatch.

Anyways just wanted to post the info.

Aguia said...

I don’t know about that enumae,

If you see it right is still 2.4Ghz.

The 2.8Ghz part you are talking about is a extreme edition that maybe have its power consuming raised that will no fit in regular notebooks pcs, more for the DTR segment.

The 2.8Ghz is also in Q4/2007 but its still 65nm.
And there is the 2.6Ghz version at 65nm in Q4/2007 too.

Aguia said...

As seen in the chart, Yorkfield would clock from 3.46ghz to as high as 3.73Ghz. Likewise, Wolfdale would clock from 3.5Ghz to as high as 4.0Ghz

But Intel said they didn’t want high power consuming processors. What I’m trying to say is that isn’t impossible to do, they just wont do it because of that.

Like I said to ho ho in previous posts,
that would raise the TDP of the quad core parts to unbelievable levels 160W and up.
I'm not sure but a hypothetical quad core at 4.0Ghz would go for 200W TDP for sure even at 45nm.

Ho Ho said...

aguia
"Like I said to ho ho in previous posts, that would raise the TDP of the quad core parts to unbelievable levels 160W and up."

I would think 45nm with high-K would drop power usage/leakage a lot. After all, when Intel first started talking about 45nm it said it had solved power leakage problem. Solved, not just reduced. Of cource they didn't eliminate it entirely but reducing it tenfold is close enough :)

aguia
"I'm not sure but a hypothetical quad core at 4.0Ghz would go for 200W TDP for sure even at 45nm."

I highly doubt even current highly OC'd quadcores could take that much power. Most motherboards aren't capable of that and the CPUs run at >5GHz. Before you say it, I know it runs on LN2. But bear in mind that is "old" 65nm and ~90% OC.

Also the TDP numbers you see are not usually the ones the CPU really has. E.g I don't think quadcore Xeons from 1.6-2.33GHz all use 80W power under full load. I expect it is the same with higher clocked CPU's too.

E.g, here is a quadcore @3.2GHz and two slightly OC'd 8800GTX'es and powerhungry NV i680 chipset. All that was powered by a 500W PSU. The entire set sucked at most 429W. What do you think, how much of that was from the CPU? FYI, NV has said G80 should peak at around 140-170W depending on manufacturer and cooler (high-speed fan can take quite some power to run).

Yes, it is inq but I think they can read the numbers on a box. It shouldn't be so difficult, even for them.

Aguia said...

ho ho,
I found this:
xbitlabs

The QX6700 at 2666Mhz does 109.4W
The Q6600 at 2400Mhz does 87W
For each 266Mhz added to each dual core goes more 22.4W

I didn’t found OC power consuming numbers.

Of course this is not linear, and again, hypothetical power consuming numbers:

2932 Mhz … 131.8W
3198 Mhz … 154.2W
3454 Mhz … 176.6W
3730 Mhz … 199.0W
3996 Mhz … 221.4W

The TDP value would have to be higher than those values, so the reason why I think Intel doesn’t do higher clocked processors is that.

Besides not they are not having real treat from AMD right now, so it doesn’t bother releasing high performing parts.
And what we have see in their road maps is that they are going to release lower performing parts/models based on Conroe (E4xxx and E2xxx).

Unknown said...

I like how I've made aguia's last point multiple times before about Barcelona's initial clocks, and was told that it had to be a problem with the process.

Ho Ho said...

aguia
"The QX6700 at 2666Mhz does 109.4W"

... with TDP of 130W

aguia
"3198 Mhz … 154.2W"

2*140+154~=420W
Do you think that motherboard, HDD, RAM and other accessories take at most 19W?

aguia
"Besides not they are not having real treat from AMD right now, so it doesn’t bother releasing high performing parts."

I agree.

Also please note that e6700 took around 62W and two e6700s that make qx6700 took less than twice that, ~109W. That is around 1.77x power usage increase, not 2x as might be expected.

Also 3GHz C2D Xeons have TDP of 80W, much less than x6800 at a bit lower clock speed.

Scientia from AMDZone said...

The significant points in the article were that the initial estimates of Intel's 45nm performance and expected delivery were wrong. It looks like no serious volume until Q1 08. Likewise, the initial estimates of clock and power draw were wrong. The initial speeds will not be that high.

The scale of hype for this is staggering when you consider that people were predicting 3.2Ghz at launch of C2D on 65nm. Now, we are talking about nothing higher than 3.0Ghz even 18 months later. This should be enough to clear away any lingering misconceptions that Intel is sitting on higher clock versions of C2D that they can release at will.

Scientia from AMDZone said...

Now, as far as AMD's 45nm process goes the point is not whether AMD will use high-K right away; the point is whether high-K is as superior as had been suggested. If high-K really did offer such tremendous benefits then AMD would be working hard to use it as quickly as possible.

Intel will use high-K in Q1 08. IBM specifically said that they were not 12 months behind Intel which would be Q1 09. It then should be possible for AMD to get high-K out by end of 2008 if necessary which is not much later than their initial release in Q3 08. AMD's ambivalence suggests that high-K is not offering that much benefit.

Ho Ho said...

Scientia
"This should be enough to clear away any lingering misconceptions that Intel is sitting on higher clock versions of C2D that they can release at will"

Do you still honestly believe that Intel can't release (much) higher clocked C2D/Q's if it wishes? As has been said numerous times before, it can, just that there is no need to.

One theory why initial clock speeds of 45nm CPU's is that low is that again, there is no need for faster CPU's since there is no hard competitors. Going to 45nm would simply help cut costs.

Ho Ho said...

erlindo
"Intel can release higher clocked C2Ds and decrease price on current models just for sake of making AMD upset (but it isn't)."

Of cource loosing money by lowering prices so soon is not the reason why they are not doing it.

erlindo
"An X2 3600+ at $92 dollars is unbeatable at that price."

I agree, at low-end Intel can't beat AMD's bang for buck ratio.

Then again, at 2.93GHz, x6800 is unbeatable in performance, so is qx6700, at least for the people who can use all four cores. People who buy such CPU's don't care much about price.

Why should Intel release a faster CPU when it already makes a whole lot of money on the ones it sells now? To make its customers happy?

Unknown said...

He's talking about the 90nm product, which can be found (last I saw) at that price range, and is definitely not a bad deal.

Hoho, this is what I was saying about Barcelona. I'm not saying you rejected this idea, but pretty much everyone else seemed to when I said it.

The significant difference here is that barcelona is not only 1Q away from the release of a major threat to its performance dominance. Core2 and core2quad are, however. If Intel really could release higher clocked chips at whim, then it would be releasing them to the server segment right now or within the next 3 months while lowering the prices of its current processors to make AMD hurt as much as possible before the release of this product while lowering the price of its current product so that it will also lower the perceived value of barcelona's performance advantage. However, Intel does not seem to be doing any of this at all. As such, there is absolutely no evidence of Intel's ability to release these products with higher clockspeeds without lowering ASP's significantly.

My thought is that Intel sells chips that are tested to be capable of reaching higher clockspeeds than they are binned for to the channel in order to inflate the image of core2. It would be interesting to check how well a core2 pulled from a Dell would overclock vs. one from the channel.

enumae said...

This is off topic, but on AMDZone there is an article talking about Opteron price cuts with a link to DigiTimes.

Some of the price cuts were substantial, about 20 - 30%, and now there will also be price cuts for Athlons.

I have a few questions for those interested.

1. Why would AMD lower prices in a segment that has already shown to have declining ASP's, and which will have a direct effect on AMD's margins?

2. Is AMD trying to win back market share in the Server segment?

3. Is AMD trying clear inventory?

4. Does AMD have enough partners or marketshare to compensate for the lower ASP's in the Server segment?

That beings us to the new price cuts for Athlon Desktop parts.

1. For a company that did not have a good Q4 2006 (even without the aquisition charges for ATI), what is AMD doing?

2. Is there a chance that Intel is regaining market share?

3. If Intel is regaining market share, could these price cuts be an attempt to maintain market share?

4. If Intel is not regaining market share, did they possibly accelerate the 65nm transition?

5. Is AMD trying clear inventory?

6. With these price cuts Server and Desktop how will this effect AMD's margins for Q1?

Thanks in advance to anyone who takes the time to answer these questions. :)

Scientia from AMDZone said...

enumae

No, those are just cascade reductions like we usually see. AMD released 6000+ and that basically just bumps everything down one price level.

Unknown said...

Well, with my usual "these people actually know what they're doing and you shouldn't think they're idiots" attitude, I'd say AMD probably has something up their sleeves. Chances are, they've known that they would need to have prices at this level as far back as q2 06, if not earlier. Either that or AMD is just going the stoic route until Barcelona is released, hoping that their market penetration will make the effect of increased margins much larger and much more profitable.

It seems as though ATI is taking the exact same route as AMD. They actually gained volume share in the desktop market, and lost volume share in the mobile market. Overall, they lowered margins, thus creating their losses. Lets see if this turns around upon the release of r600 for mobile on 65nm as well as rd600 chipsets.

enumae said...

Scientia
No, those are just cascade reductions like we usually see. AMD released 6000+ and that basically just bumps everything down one price level.

Thanks for the response.

Like I had said in the past, I have not been following these things long enough to completely understand certain actions.

Could you elaborate on a pevious pricewar in which this was previously done?

Greg
Well, with my usual "these people actually know what they're doing and you shouldn't think they're idiots" attitude, I'd say AMD probably has something up their sleeves.

I am not sure why you assume that I am thinking they are idiots.

I am simply questioning what their strategy is.

If they are having trouble with there ASP's, there is a reason why they would sacrifice there ASP even more?

Scientia from AMDZone said...

Well, as I keep trying to explain to Real, AMD's costs drop as the ratio of production shifts to 300mm and 65nm. It really doesn't matter if AMD doesn't make a profit in Q1. Things should begin picking up again in Q2. As I said in my article, AMD is not going to go bankrupt in 2007 with a new architecture, new mobile chipset, and integrated desktop chipset.

Scientia from AMDZone said...

A cascade reduction is very simple. Let's say you have:

$305 - 5400+
$244 - 5000+
$195 - 4600+
$156 - 4200+
$125 - 3800+

Now, you introduce 5800+. Naturally, it takes the top price and pushes everything down so then you have:

$305 - 5800+
$244 - 5400+, -20%
$195 - 5000+, -20%
$156 - 4600+, -20%
$125 - 4200+, -20%
$100 - 3800+, -20%

The naiive assessment would be that because most of the models show a 20% drop that this would cause AMD's revenues to drop 20%. In reality, there has been no change in ASP.

These price reductions are routine and have nothing to do with a price war. During a genuine price war, all of the prices drop therefore there are corresponding drops in ASP.

enumae said...

Thank you for explaining, but you have lead me to more questions...

The total for X2's from the link I provided would equal $3,691 for one of each of those 15 processors.

That would equal an ASP of $246.06 (3691/15).

The new prices total $3,368 for one of each of those 17 processors.

That would equal an ASP of $198.12
(3368/17).

Wouldn't that in fact change there ASP by 19.5%?

Now if these are made on 65nm vs 90nm, then the lowered ASP could be offset by the new manufacturing node.

65nm parts it totals $646, with an ASP of $215. The new prices will total $628 with an ASP of $157.

65nm ASP change is about 27%.

90nm parts total $3045, with an ASP of $253. The new prices will total $2740 with an ASP of $210.

90nm ASP change is about 17%.

I do not mean, nor am I attempting to be stuborn, but the numbers tell a different story than what your saying.

Unknown said...

I wasn't saying that you thought they were stupid, I was just pointing out my attitude that for some reason seems to be an uncommon one, with the exception of you, me, and scientia.

Scientia from AMDZone said...

enumae
Wouldn't that in fact change there ASP by 19.5%?


Well, I'll give you some credit for effort but your method is way off. Both totals are garbage values.

The only way that you could do the calculation would be if you knew what percentage of each was previously sold with the previous price, and then the new percentage of each with the new price. Then you could get a real value.

However, even a casual look at the values show a high degree of correspondence. For example:

6000 -> 5600
$465 ~= $505, -8%

5600 -> 5200
$326 ~= $295, -10%

5400 -> 5000
$267 ~= $285, -6%

5000 -> 4600
$217 ~= $215, +1%

4600 -> 4200
$195 ~= $173, +13%

4200 -> 3800
$159 ~= $138, +15%

So, you can see that although the values drop at the high end, they actually increase at the bottom of the range. So, no big drop in ASP on average.

Ho Ho said...

greg
"Lets see if this turns around upon the release of r600 for mobile on 65nm as well as rd600 chipsets"

R600 will start having some real effect from the middle of Q2, if not at start of Q3. Even then I am not sure if it would be anything too big. I would assume NV to relase its products refreshes around that time.

azary omega
"they know that 32nm is far-far away"

Around 2.5-3.5 years away assuming they haven't radically changed their roadmap.

azary omega
"Also, what do they need to make an 8-core CPU? Will two native quadcores gonna do it?"

I can't see a reason why couldn't they continue with their dual-die approach. I'm quite sure that we will have first dual-die octacores in around two years on 45nm. Of cource FSB will get a bit crowded but it is possible to make it work at much faster clock speed to ease the load.

Die-area wise these shouldn't be too big. My guess is around 400-450mm^2, assuming it won't have too much L2.

Scientia from AMDZone said...

Paul
I have no idea why you are having trouble understanding this. This is the chart:

Intel Roadmap

The 2008 Q1 column shows the initial clocks for Yorkfield at 2.4Ghz and Wolfdale at 3.0Ghz. These are the first 45nm chips listed. Kentsfield does show a higher clock at 2.66Ghz but Conroe only shows the same 3.0Ghz clock as Wolfdale.

The point is that there are no big initial increases in clock on 45nm nor are there any higher clocks on 65nm than what is currently offered.

I gave up on your earlier dross and removed it from the comments. See if you can post something more relevant.

Ho Ho said...

scientia, do you have any comments on the xtremetech article about high-k and AMD?

Aguia said...

Scientia in fact Paul is right. I was only aware of that after seeing the legend.

Yorkfield is Orange/pink.
Kentsfield is Yellow/pink.

Aguia said...

paul c,

Good explanation but that doesn’t remove the fact that there will not be higher clocking parts.
In fact lets see if Intel can release the new 45nm parts with higher clock than the old 65nm parts.

Everyone whine about AMD 65nm parts clock lower than the 90nm but that already happen in the transaction of 130nm to 90nm.
Intel P4 to 130nm » 90nm » 65nm how much clock speed increase occurred?
PentiumM/Core to 130nm » 90nm » 65nm how much clock speed increase occurred?
Core 2 Duo 65nm » 45nm how much clock speed will occur? I’m not expecting miracles.

enumae said...

Thanks Scientia.

To be clear, we can not calculate there current or future ASP due to the fact that we do not have production percentages, correct?

Aguia said...

Intel P4:
180 nm – 2.0 Ghz
130 nm – 3.46Ghz
90 nm – 3.8 Ghz
65 nm – 3.73Ghz

Pentium M/Core (mobile):
130 nm – 1.7Ghz
90 nm – 2.26 Ghz
65 nm – 2.33Ghz (still updating)

AMD Athlon:
180 nm – 1.733 Ghz
130 nm – 2.4Ghz
90 nm – 3.0 Ghz
65 nm – 2.6Ghz (still updating)

Aguia said...

And about the AMD ASP, you (Paul) assume that:
-AMD will not release new products.
-The performance of the new products will not be high enough to charge a premium for them.

savantu said...

And about the AMD ASP, you (Paul) assume that:
-AMD will not release new products.
-The performance of the new products will not be high enough to charge a premium for them.


What I'm saying is that AMD has never had most of its product line under $300.From a marketing POV it isn't very smart.Basically they're squizzed in the sub $300 region with Core at the upper half and Pentium at the lower one.

Look at their ASPs : $76 vs. $130.This should tell you something.

http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_609,00.html?redir=CPPR01

As for new products , it is obvious they cannot ask a premium for them.Gone are they days when a 4800+ sold for $600...

Aguia said...

"As for new products"

I was referring to the based on the new K8L core not the current 4xxx+ 5xxx+ and 6xxx+.

Those will be faster than their current offering, so priced above (300$ and up).
We are talking about 5 months maybe less for the introduction of the dual core parts, but who knows.

enumae said...

I found this interesting and it relates to clock speed head room for Core 2 Duo Quad cores.

Gateway is aparently providing a warranty ($170 - 4 years) on their FX530, which has an Intel QX6700 overclocked to 3.2GHz using a modified BTX cooling unit.

If in fact Intel does not have some headroom, Gateway will be spending well over the $170 to replace Intels processor.

So there is obviously headroom if a company like Gateway is willing to risk this.

*Sorry hit enter and had not added the link*

Unknown said...

Lol, Paul. It's pretty obvious you don't know how to read the cpu chart. Yes, each CPU has its own specific color patch for its respective period of existence, but if we used your method of analysis we'd have seen penryn parts last quarter (I'm pretty sure that's not the case). If you bothered to look closely, you'd notice that yorkfield and wolfdale's squares are slightly darker due to being mottled with a slight orange tint. This tint only occurs in q1 08 onward, not in q3 or q4 07.

hoho, Nvidia is not releasing their new chipset refresh until q3 and will release their lower end 8 series graphics parts at the same time ATI releases the r600. As such, in q3, when things really start to heat up is when we'll see who is doing better. It could be a dramatic turn around if Nvidia does not release its lower end and upper end refreshes at 65nm. I have not heard anything to this effect, and as such, Nvidia may lose the performance crown for long enough that it really hurts them.

Unknown said...

I've decided I should be helpful instead of critical, so here is another roadmap released by the same guys that is quite a bit clearer. As you'll notice, 45nm is specifically blocked for q1 2008, so there's no mistaking when that will exist. Also, penryn is specifically past the 2008 dividing line, so again, no mistaking that.

One thing I find interesting is that Intel is releasing both low and high end penryn cores, all at the same time. I'm pretty sure they've never done this before, and don't no about the exact execution ramifications that would have. Anyway, that just seems really fishy to me, either to the credibility of these graphs, or to Intel's surety of its own performance.

Unknown said...

Lol, forgot the link.
http://www.rubyworks.net/quarkimages/inteldie.jpg

savantu said...

Lol, Paul. It's pretty obvious you don't know how to read the cpu chart. Yes, each CPU has its own specific color patch for its respective period of existence, but if we used your method of analysis we'd have seen penryn parts last quarter (I'm pretty sure that's not the case). If you bothered to look closely, you'd notice that yorkfield and wolfdale's squares are slightly darker due to being mottled with a slight orange tint. This tint only occurs in q1 08 onward, not in q3 or q4 07.

You have the legend down , Wolfdale is a different colour from the 65nm E6850.

Besides , believe what you want.If you think Yorkfield will launch at 2.4GHz a year from now when it was demoed on A0 silicon at 2.13GHz a month ago so be it...

savantu said...

In fact lets see if Intel can release the new 45nm parts with higher clock than the old 65nm parts.

Everyone whine about AMD 65nm parts clock lower than the 90nm but that already happen in the transaction of 130nm to 90nm.

Extrapolations based on AMD's results is useless for Intel.


Intel P4 to 130nm » 90nm » 65nm how much clock speed increase occurred?
PentiumM/Core to 130nm » 90nm » 65nm how much clock speed increase occurred?
Core 2 Duo 65nm » 45nm how much clock speed will occur? I’m not expecting miracles.


P4 130->90nm gained 340MHz and it was stopped on TDP issues ( it had to be <115w )
From 90nm to 65nm it gained a lot , from 3.8GHz SC it went to 3.73GHz DC in the same thermal envelope.
Recently , a 631 was OC to over 8GHz. => Intel held back the 65nm P4 from political reasons , it had nothing to do with the technology.

A 3.6GHz P4 65nm burns 65w of power.With a 115/130w TDP it could have reached at least 4.5GHz.But with Conroe next corner it was useless.

How much headrom Conroe has ? A lot.Go to xtremesystems.org and see.

Intel's 45nm process works , and it works well.

Unknown said...

If it works well, why only demo it at 2.13ghz when you could demo it at higher TDP without telling anyone about it. It wouldn't be inflate prospects enough to be illegal because it's not like you couldn't reach those numbers if you want to.

I'll also point out, again, that there may be discrepancies in how high chips sent to dell can overclock compared to chips sent to the channel. This, more than anything, would effect what clockspeeds Intel would be able to release chips at.

There's a reason AMD doesn't change cores and processes at the same time. Maybe Intel was too confident.

savantu said...

If it works well, why only demo it at 2.13ghz when you could demo it at higher TDP without telling anyone about it. It wouldn't be inflate prospects enough to be illegal because it's not like you couldn't reach those numbers if you want to.

Huh?! It's A0 silicon for God's sake ! That are the very 1st chips.

K8 A0 silicon ran at 800MHz and was launched at 2GHz.

I'll also point out, again, that there may be discrepancies in how high chips sent to dell can overclock compared to chips sent to the channel. This, more than anything, would effect what clockspeeds Intel would be able to release chips at.

There's a reason AMD doesn't change cores and processes at the same time. Maybe Intel was too confident.


Of yes , Intel releases OC friendly chips in retail.Who would have thought of that? Do you see conspiracies everywhere ?

Intel isn't changing cores , that's the whole point between the tick tock strategy.Penryn is a tweaked shrink , not a new core.

Roborat, Ph.D said...

... why readers from Intel keep coming back.

maybe because people have a tendency to watch other peoples misery. 2 years ago when AMD was hammering (pun intended) Intel, most pro-intel fanatics stayed away from forums or kept quiet and didn't like reading benchmarks and opinions about CPUs.

Christian H. said...

The lack of you comprehension skill is very relevant.

I thought you were different from Sharikou.Looks like you suffer from the same problems.

If you try to poke Intel in the eye , at least do it in a smart way.But for that you'd at least need to know how to read a roadmap.



I am seeing what Sci is seeing.

2.4GHz Yorkfield
3GHz Wolfdale

As far as clockspeed increases, it's one thing to have a chp that be OCed to a certain speed and anoter to ship it at that speed. That means higher multpliers and higher TDP.

I don't think we'll get too far above 3.5GHz for a few years on AMD64/EMT64.

I mean C2Q is drawing 130W at 2.66GHz. Crankng that up to 3.4GHz would take you above 150W.

Scientia from AMDZone said...

aguia
Scientia in fact Paul is right. I was only aware of that after seeing the legend.


Okay, if you understand what Paul is talking about then maybe you can explain it to me. Frankly, I couldn't sift anything relevant out of his rude comments.

Yorkfield is Orange/pink.
Kentsfield is Yellow/pink.


Yes, all of the segments from Performance down to Mainstream 2 are a different color and show that the 65nm chips overlap with the 45nm production. Therefore, the older 65nm chips will continue to be made. Apparently in the Extreme category 65nm is completely replaced with 45nm.

Now, I suppose you could be suggesting that the numbers are for the older 65nm chips and no clock speeds are given for 45nm. If this is the point then Paul certainly could have made this himself quite a few posts back. Instead, what he posted was both rude and written so sloppily that I could not tell what point he was trying to make.

Now, let's say that this is true and the numbers listed are only for 45nm. Okay. This still would not change the fact that 45nm has been pushed back to Q1 08 and that no 3.2Ghz Conroes will be released on 65nm. So, again, this is still a lot less than was suggested. I recall the talk about 3.2Ghz Conroes at launch. None even 18 months later suggests that there is no headroom on 65nm.

Sorry about removing your post; I removed all of Paul's earlier posts and all direct replies to them (including mine and others). I put up with Red's comments for a long time and let him trash thread after thread and still ended up banning him. I'm not going to keep having threads trashed.

So, Paul, if the above was the point you were trying to make then I suggest that you: stop being rude and learn how to write something in a clear fashion. I have no problem with having corrections pointed out and if you are optimistic that 45nm will release at higher clocks than 65nm then that is fine, but you are not going to be rude here.

Scientia from AMDZone said...

enumae
Gateway is aparently providing a warranty ($170 - 4 years) on their FX530, which has an Intel QX6700 overclocked to 3.2GHz using a modified BTX cooling unit.


This is almost identical to what Sun did with Opteron. Sun had a special version that clocked 200Mhz higher. It had different cooling and was warranteed by Sun. So, this doesn't really indicate headroom.

As far as the ASPs go, right you can't calculated them just from the prices, you also have to know how many of each are being sold. We know for a fact that desktop ASPs came down in Q3 and server ASPs came down in Q4. They could potentially drop again in this quarter. I guess we'll find out in April. I would imagine that around that time someone will probably get their hands on an ES Barcelona as well.

Scientia from AMDZone said...

Savantu is right that early engineering samples are not indicative of release clocks. Opteron, Itanium, and Conroe all had early samples at 800Mhz but the release clocks were much higher.

However, I would disagree with Savantu about headroom. There is a long history of Intel's chips having very good OC potential but these chips are never released at those clocks. I have tried before to explain that OC'ing is similar to souping up the engine on your car. You can do it if you are knowledgeable but that doesn't mean that the factory will start offering that high of HP on the stock engine.

Probably one of the most extreme examples of that I can think of was the 2.3L engine on the Ford Pinto and Mustang II. The stock engine developed 73 HP. Yet, this engine was capable of being souped up to 200HP with the stock crank, rods, and pistons. In contrast, when GM wanted to use a Fiero as an Indy pace car, the first thing they did was replace the engine with one of their racing engines. The stock Fiero engine was only capable of, at most, 140HP. It wasn't just Ford though, the 2.2L Dodge engine used in the Dodge Omni was also capable of 200HP. The Ford engine was never offered at these power levels while the Dodge engine did see some limited production at 175HP - 220HP.

There is a big difference between someone knowledgeable tweaking the chip voltage and settings and monitoring cooling, and having the same chip put into a stock case with stock fans and being run in someone's house in the summer with no air conditioning. OC'ing is not headroom. Knowledgeable OC'ers will always get more performance than stock.

Unknown said...

Savantu, I'm not saying it's a conspiracy theory, and in fact, would think both Intel and AMD would be stupid not to implement that sort of system if they could. Why waste highly OC'able cores to the general public when they wont bother or be able to use that type of control. Seriously, you should quit being so hostile.

Also, I didn't realize what AO silicon was, but thanks for explaining that instead of just acting like I'm stupid(sarcasm).

Ho Ho said...

Scientia
"There is a big difference between someone knowledgeable tweaking the chip voltage and settings and monitoring cooling, and having the same chip put into a stock case with stock fans and being run in someone's house in the summer with no air conditioning"

I'm getting quite repetible already but why can't that be done at factory? Most chips of one product family run at different TDP and changed voltages. E.g take e6700 vs x6800. Also please note that 3GHz dualcore Xeons have TDP of 80W. What stops Intel from releasing 120W 3.xx GHz dualcores? Especially if it has 16M L3 MP Netbursts with 150W TDP.

Also I'm still waiting for a comment on the AMD and high-k thingie.

Scientia from AMDZone said...

ho hot
I'm getting quite repetible already but why can't that be done at factory? Most chips of one product family run at different TDP and changed voltages.


You don't understand. Clock tolerance is not the same as thermal tolerance. Chips can be tested for stable clocking but not for thermal tolerance on bulk silicon. The only way to establish thermal tolerance is to run the chip until it experiences thermal cascade failure (which can damage the chip). So, instead of testing to find thermal tolerance they test to find measured power draw (which is equal to TDP) and then rate with margins.

This is common in engineering BTW, you can't test every beam to failure so you test some and establish a normal failure limit and then set your maximum at 67% of that. They basically do the same thing with bulk silicon chips. AMD chips are different because SOI does not experience thermal cascade failure.

Also I'm still waiting for a comment on the AMD and high-k thingie.

Okay, if 45nm high-K were such a huge jump in performance then AMD would be pulling out all the stops to get it into production. Their ambivalence suggests that the difference is less profound.

Aguia said...

ho ho,
let me correct you:

Intel TDP

Acording to Intel:
Quad-Core Intel Xeon processor X5355 2.66GHz 1333MHz 8MB 120W

With your own calcs:
80W x 2 = 160W not 120W!

I have already told you that multiple times.

Aguia said...

Scientia,

Thats exacly what you said.

Yes Paul was being a little rude...

Ho Ho said...

Scientia
"You don't understand. Clock tolerance is not the same as thermal tolerance"

Could you please provide some links where I could read more about the subject? Especially good would be if it had some connections with (Intel) CPU's.

Scientia
"Okay, if 45nm high-K were such a huge jump in performance then AMD would be pulling out all the stops to get it into production. Their ambivalence suggests that the difference is less profound."

Perhaps but then why does IBM think it is so good? I have two guesses:
1) AMD current technology wouldn't gain much benefit from it
2) AMD doesn't have the time and/or recources to make it work with 45nm


Aguia, could you
show me some C2D based quadcore CPU where TDP increased 2x when going from dualcore to quadcore? I could only see 2.33GHz LV5148 vs E5345.

Closest non-LV I can see is from 5150 or 3070 to X5355 where TDP increased from 65W to 120W, or around 1.85x at same clocks. At 2.4GHz going to dualcore meant increase of 1.62x.

So what exactly was your point? If it was to show that TDP doesn't increase 2x when going to quadcore then we have been trying to say the same thing :)

Aguia said...

savantu,

"Extrapolations based on AMD's results is useless for Intel."

Not really. You just cant compare Core 2 Duo (Desktop and Server) on 65nm VS 90nm because that processor never existed in that process. But Core 2 Duo 65nm VS 45nm we can.

I used the PentiumM/Core Duo (Mobile) because with those I can compare. The reality is that there was just 70Mhz clock increase from 90nm to 65nm. I'm not talking of performance, just clock speeds.

"A 3.6GHz P4 65nm burns 65w of power.With a 115/130w TDP it could have reached at least 4.5GHz.But with Conroe next corner it was useless."

Exacly. And with Conroe there are 2 real cores not just one. But dont forget Conroe is based on a mobile processor design, thats why it consumes much less.

"How much headrom Conroe has ? A lot.Go to xtremesystems.org and see."

But the PentiumM/P4 also had lots of headrom and you never saw faster version because Intel couldnt do it, or want to do it. One or another is not important if the product will no be released. Overclockers care (1% or less), but the rest dont.

"Intel's 45nm process works , and it works well."

Well AMD 65nm process works, and it works well, whats your point? We never discussed if it works or not.

abinstein said...

What's your opinion regarding the rumor that Intel's going to pull 45nm ahead to 1H 07? If that turns out to be true, would it be a big impact on AMD? Does that mean Intel's actually more ahead of AMD than we previously thought in terms of processing technologies? Thanks.

Wise lnvestor said...

Well... I got something that's not a rumor: nividia is having a serious driver crisis.

Seems series 6,7,8 video cards and nforce mother boards all have major issue under Vista. In addition under Xp also.

Unknown said...

Abinstein, that rumor is only being thrown around in forums of little repute. Intel's fab scheme means that just because they can display a working chip, does not mean they can produce it in volume, as they have to wait for their 45nm fabs to be finished building before they can begin production. This is likely where the delay comes from.

Another thing I noticed is that Intel's press release said that "production" will "begin" in the second half of 07, not 3q and not availability. As such, it makes perfect sense that the processors would first be available in q1 08, and I now no longer believe this is a delay, but simply overhyping by overly optimistic fans that ended up getting to everybody (and I mean this on both sides of the issue).

Ho Ho said...

wise lnvestor
"Seems series 6,7,8 video cards and nforce mother boards all have major issue under Vista. In addition under Xp also."

Everything is working perfectly here in Linux. Just yesterday I was playin Serious Sam 2, a rather nice game.

Intel fanboi said...

Scientia,
You are a respectable person, but obviously biased for AMD. You pick out the data that supports your case and ignore the data that contradicts. What about vPro? What about the fact that the roadmap you link to shows 3Ghz as mainstream sub $300 parts in Q307? What about Bearlake? Sonoma refresh? These are all happening in 2007. These will all have impacts on the competitive landscape.

I can think of a few good reasons why someone would be an AMD fanboi:

1. They work for AMD.

2. They like the idea/story of an underdog.

3. They have a chip on their shoulder.

If you are #1, then great, keep writing. If you are #2, you should reconsider the difference between a corporation and a person. If you are #3, stop writing cause Shakirou has the insane fanboi catergory locked up.

Intel Fanboi said...

Response to azary omega.

Intel just announced more on an 80 core device today. Two weeks ago it was high-K devices. This year will be Robson Tech, next year will be Wi-Max. Seems like Intel is doing a heck of a lot of innovation.

You should switch from being an AMD fanboi to an Intel fanboi. You won't feel so angry.

Unknown said...

Even without looking at stocks, innovation, and product, Intel is still a bad company. They have a group of employees that has formed an advocacy group due to the amount of pollution Intel generated in its fabs and the poor workers comp for fab workers. They use up tons of resources with their drop and go fab scheme, where instead of updating their fabs, they build entire new ones, and just let the antiquated ones sit.

They have also forced companies out of the market because of "perceived" possibility of competitiveness. They threaten smaller companies than them and user their market dominance to threaten their customers into refusing products from competitors or more "perceived" possible competition.

Via is a fairly recent example of this, and AMD is the only other surviving example of all of this. Remember HP's big downturn? Remember how that corresponded with their offering AMD products? Remember how they still sold those products very well, but simply couldn't keep their prices low enough? Yet another example.

Look, I know Intel makes a good product quite often, but do we want to support a company whose management is so morally bankrupt? I don't.

Intel Fanboi said...

Azary Omega,
Come on guy, these are all great things Intel is doing. And the list I gave, that you tried to dismiss, is just the tip of the iceburg. They have new GPUs in development, a new system on a chip, NAND and NOR flash, fiber optic on silicon, vPro, Centrino Pro, etc. And to top it all off they will have an IMC and CSI in 2008/2009 with Nehalem. You can dismiss these too if you wish, or you can give them credit for the incredible sucess they are having right now tech wise. We should be civil here, as Scientia requested, but it seems you suffer from Intel Fanboi's #3 hypothesis.

abinstein said...

Intel made good PC processors, good compilers, and good gigabit Ethernet chips. It certainly has good fabs. Other than above, other stuff like wireless, communication, system/platform architecture/design, it didn't do well (yet), and is probably downsizing them (except wireless).

So I'm really not sure how great a company Intel is. In terms of PC processor dominance, it is of course at the top of the world. However, anyone with a clear mind knows that Intel got there by some luck and lots of nasty tactics. For about the same size, IBM has much more innovations and contributions to the industry than Intel over the last 20 years.

Aguia said...

"Come on guy, these are all great things Intel is doing."

Yes they are. But doesn’t Sun, IBM, Nvidia, Ati, and many others also do?

"They have new GPUs in development, a new system on a chip, NAND and NOR flash, fiber optic on silicon, vPro, Centrino Pro, etc."

Most of these things we don’t know because they aren’t available yet, but if they are great let them come and see if they can innovate the market.
The vPro Centrino Pro, you know that’s marketing right?
If a great thing for the industry is marketing or a glued logo on the computer, thanks but I don’t want it.
Most of Intel great things is doing, is to get the others out of the market.
For example the Centrino logo put companies that had good products in the mobile area to rest.

"And to top it all off they will have an IMC and CSI in 2008/2009 with Nehalem."

But that as been available since 2001 from AMD. The only point I see here is that they are behind of AMD in that department in at least 6 years.

AMD also have high end GPUs now, how many years will take Intel to come up with a similar solution?

Intel Fanboi said...

To Aguia,

The vPro and Centrino Pro offer new technologies and services. To call it purely marketing is to call Centrino purely marketing. I completely agree that the original Centrino consisted of pieces that anyone could have put together. Intel's contribution was to make those pieces the standard that the consumer expects from a laptop. Within two years, you will not buy a business desktop or laptop unless it says Pro on it or has the equivalent services/technology. AMT, virtualization, security, gig ethernet, etc will all be standard features on business platforms.

Unknown said...

John, vPro and Centrino (both Pro and non) only serve to make the markets they serve less competitive. Wireless chipsets are now only handled by Intel, and as such, they get by with selling worse chipsets for more money because they're "certified". The customer, thus, gets screwed. Also, the technologies that go into vPro and Centrino Pro are not new, they're simply being certified by Intel to show that the platform is stable, as if that couldn't be accomplished with broadcom and other parts.

After centrino first came out, I immediately made sure to avoid all of their products, due to their slow and quirky wireless cards and the massive amount of bloatware that is nearly essential to the operation of those platforms. My dad had the misfortune of getting stuck with one of their "certified" centrino systems that limited network connections to 1MB/s. To add insult to injury, it would also constantly disconnect, due to a now patched flaw in windows. The moral of the story: the sticker neither proved quality, nor insured stability and ease of use.

Also, those standards that will "someday" be standard on business platforms have existed for a very long time, and would already be standard if Intel had reason to update its Centrino platform (a.k.a. competition).

abinstein said...

Greg:"My dad had the misfortune of getting stuck with one of their "certified" centrino systems that limited network connections to 1MB/s. To add insult to injury, it would also constantly disconnect, due to a now patched flaw in windows."

When I tell similar things that I saw on Centrino to some Intel fans, they simply shrug it off and demand "performance numbers" that would back up my stories. Ironically, the first person I know who recognizes Centrino's poor performance is an Intel employee who has to use and benchmark it day over day.

It's obvious that Centrino has been very successful (as a marketing campaign), that people believe what Intel told them on TV ads but nothing else. Intel is really a great marketing company, partly due to its huge marketing budget. Other than this, as I said, it really has just few innovative contribution to the IT industry.

Christian H. said...

Via is a fairly recent example of this, and AMD is the only other surviving example of all of this. Remember HP's big downturn? Remember how that corresponded with their offering AMD products? Remember how they still sold those products very well, but simply couldn't keep their prices low enough? Yet another example.

Look, I know Intel makes a good product quite often, but do we want to support a company whose management is so morally bankrupt? I don't.




AMEN TO THAT! I also won't buy Intel because of their predatory actions and isolationist stance.

AMD Live! is outselling Viiv probaly because of that, where with AMD you can use a choice of components but not Intel.


You get what they give you period.

Unknown said...

I always thought it was freakishly hilarious that Viiv Outsold live when it had more obscure (to the average user) components and was from a smaller less well known processor company.

Maybe I'd like Intel on the whole if it became a marketing company, and left the technology to everyone else.

Anyone else see that Intel is going to use its 80 cpu design as a GPU? Sounds pretty pathetic, seeing as it will only do 1 Tflop, and be out as late as 09.

SweepingChange said...

greg
I always thought it was freakishly hilarious that Viiv Outsold live when it had more obscure (to the average user) components and was from a smaller less well known processor company.

Live outsold Viiv, simply because AMD has a larger US retail presence than Intel. And US retail figures dont include Dell. Viiv started as a digital home experience kind of brand, that is trying to build some kind of convergence. AMD Live was started as a "me-too" response to Viiv. Even today, AMD Live and Intel Viiv mean nothing more than just specifications for a PCs that can be used as good HTPCs. The covergence isnt there. Their relative sucesses is proportionately no greater than just their (Intel/AMD)'s equivalent US retail market share. If you want to look at some form of convergence, look at Apple TV and Xbox 360.

greg
Maybe I'd like Intel on the whole if it became a marketing company, and left the technology to everyone else.
Viiv didnt have much of a marketing program. If you looked at all TV ads, you would have seen Centrino and Core2 Duo ads, but no Viiv ads, simply because Intel isnt sure it has all the things right about the platform to promote it. Now look at them as a corporation, who felt it wasnt right to fool the public with what you dont have quite right.

Anyone else see that Intel is going to use its 80 cpu design as a GPU? Sounds pretty pathetic, seeing as it will only do 1 Tflop, and be out as late as 09
No one, does. Simply because Intel themselves have said they dont intend to bring out the CPU in another 5-10years. That was a test-chip, to solve a potential interconnect routing problem prevalent in future many-core architectures.

greg
Even without looking at stocks, innovation, and product, Intel is still a bad company. They have a group of employees that has formed an advocacy group due to the amount of pollution Intel generated in its fabs and the poor workers comp for fab workers. They use up tons of resources with their drop and go fab scheme, where instead of updating their fabs, they build entire new ones, and just let the antiquated ones sit.

Look, I know Intel makes a good product quite often, but do we want to support a company whose management is so morally bankrupt? I don't.


Sir, please promise me that in the next 10 years, that in your path to success in your job or what ever, you wont step on a few toes to get there, and I will call Intel the worst company ever.

Its outright war out there. A company the size of Intel cannot succeed if they believe in completely open standards. And they cannot let anybody in the door. You are crying foul because you think they stifled competition.

Many people in the world are jealous of Intel and Microsoft's position, because they think these companies became successful by anti-competitive and monopolistic practices. They think that these companies are the evil empires. But unfortunately the sad truth is that empires create stability in the region. Its much better to have 2 CPU manufactures to choose from than 10. Think on a small scale, of the chaos, for not only user but also developers.

Microsoft is considered a monopoly, as it is through Windows that it has stifled others like Netscape and many more. Yet the founders Bill and Melinda Gates are one of the largest philanthropists in the world. Intel is considered a monopolist yet it was made huge financial contributions to Katrina, the Asian Tsunami, and is a huge supporter of US education and graduate programs. Besides this it wasnt a vitim of stock options investigation like so many companies. Apple, you think. Now here is a company whose CEO is being investigated plus the guy doesnt do no charity work or even cares about the "population". Intel still has a lot of integrity as a company, it still employs around 100,000 who around half work in those polluting fabs you mentioned and are still enjoying their careers.

Sirs, you should seriously get over it. Intel is Intel because to get where they are now, they had to protect their business and stifle others. In today's world that is the only way to success. Its not a healthy thought, not everyone dares to think that way, thats why we see only one or two successful entities, and everybody else is mass majority, talking about "how we should all align and have open standards blablabla." Smarter choice... some more bs.

Sir, today as a common man, you believe that Intel is an evil entity and you cant stand for it, so you wont buy its products. But I am sure tommorrow as you become a top shot, and you see several small frys vying and baying for your blood, you will begin to appreciate Intel for its achievements and maybe even buy a Centrino.

Intel's immoral management is nothing but folks like you and I, who are weighed by a family that needs to eat and a stock market that does not forgive. My saying here is not to praise Intel, but dont ding them for what any corporation would have done to protect their market, their turf, their bread. The same immorality will soon be seen in AMD, when they start to realize that "open-standards" and "choice" dont give them much revenue growth, and they have to increasingly cross allies, double-cross friends and begin owning more of the pieces of the product that they were helped by others in building.

Intel Fanboi said...

Greg said:
vPro and Centrino (both Pro and non) only serve to make the markets they serve less competitive. Wireless chipsets are now only handled by Intel, and as such, they get by with selling worse chipsets for more money because they're "certified". The customer, thus, gets screwed. Also, the technologies that go into vPro and Centrino Pro are not new, they're simply being certified by Intel to show that the platform is stable, as if that couldn't be accomplished with broadcom and other parts.

Intel is not the only company/partnership that can put together a platform like Centrino or vPro and get it to the market. Obviously Intel planned these years in advance, so give them credit for being way ahead of the market AND their competition.

After centrino first came out, I immediately made sure to avoid all of their products, due to their slow and quirky wireless cards and the massive amount of bloatware that is nearly essential to the operation of those platforms. My dad had the misfortune of getting stuck with one of their "certified" centrino systems that limited network connections to 1MB/s. To add insult to injury, it would also constantly disconnect, due to a now patched flaw in windows. The moral of the story: the sticker neither proved quality, nor insured stability and ease of use.

I am sorry your Dad had a poor experience. But anecdotal stories aren't the best reference points.

Also, those standards that will "someday" be standard on business platforms have existed for a very long time, and would already be standard if Intel had reason to update its Centrino platform (a.k.a. competition).

First you say Intel is stifling innovation. Then in the above paragraph you say that Intel isn't bringing out innovation because they don't have to due to lack of competition. So which is it? Is intel stifling competition or inviting competition by not innovating? You can't have it BOTH ways.

Intel is about to update its Centrino platform. Robson tech will be a highlight. WiMax will be next year. Where is the competition on this? No one is stopping AMD or Broadcom from doing these sorts of things.
Also you claim that these features of vPro have existed for a long time. Can you show me the product out there that has AMT, virtualization, Gig Ethernet, dual core, and security all in one package please? Please be specific about how you define AMT.

hyc said...

Frankly, in terms of high tech, too much stability is a bad thing, because then it is no longer "high tech."

The chaos of having 10 different competing CPU ISAs is really no big deal if all of them are openly documented and anybody can support them without paying exorbitant licenses. You can have "chaos" and a multiplicity of innovations as long as all of those innovations are openly documented.

Ho Ho said...

aguia
"AMD also have high end GPUs now, how many years will take Intel to come up with a similar solution?"

I'd say 3-4 years at most. Something competitive should be here in at most two years.

Only thing is I don't think Intel tries to do yet another rasterizer but it'll try to innovate and go the way of ray tracers instead. At least that is what I've concluded after reading lots of things about how Intel has talked about real time ray tracing.

greg
"Sounds pretty pathetic, seeing as it will only do 1 Tflop, and be out as late as 09."

You've misunderstood. The 1TFLOP CPU has been in usable state for almost a year now.

They use it to research many-core architectures. The thing that comes out in 09 will probably be based on similar architecture but will most likely be a lot faster. CPU's can get quite a bit faster in two years, especially with 1-2 manufacturing process changes (IIRC, intel was supposed to start producing 32nm CPU's in 09).

hyc
"The chaos of having 10 different competing CPU ISAs is really no big deal if all of them are openly documented and anybody can support them without paying exorbitant licenses. You can have "chaos" and a multiplicity of innovations as long as all of those innovations are openly documented."

Well, there are tens of well known Linux distributors and much more little known ones. Problem with them is newbs don't have a clue what to choose

hyc said...

ho ho said
Well, there are tens of well known Linux distributors and much more little known ones. Problem with them is newbs don't have a clue what to choose


As far as CPUs go, it doesn't matter. If the choice of CPU is really important to your workload, then you do your homework and compare. In most cases, for a newbie the choice is irrelevant because any processor will handle the tasks they need to accomplish.

Likewise for choice of Linux distro - pick the one in the shiniest package, or the one with the thickest printed manual, or whatever. For a newbie it's all the same. The underlying differences only become important as your usage gets more sophisticated.

Unknown said...

Success is not market domination, and you're right, you can't easily succeed in business without stepping on a few toes. That doesn't mean you land an uppercut in someone's stomach. Intel could've competed by simply offering better deals, and being more competitive product wise. Instead, Intel blocked other companies from entering or having an opportunity to be competitive in their field. Then, when those companies were gone, they sat back and enjoyed the benefits. At this point, they could not know if they were protecting their business, but they knew they were stifling competition, and thus innovation. Removing competition and surviving are two completely different things, and you should know that.

While the anecdote was unnecessary, there was still a point behind it.

Like I just pointed out above, you CAN have it both ways. Maybe not at the same time or same way, but Intel definitely stifles both.

No PLATFORM had any of those features before, but they've been out individually for a while. AMD has been incapable of providing such benefits in a platform because of *cough* certain limitations in their financial status and market presence.

hyc, very true, and in fact, it could be said of many markets that too much stability is a bad thing.

I can't find back my source that says the CPU design will also be used for graphics, but Intel's graphics products will be based on the design (in the discrete market at least). I have been mis-informed about the performance of certain graphics cards, though. 1Tflop would be on par with other gfx cards in the 09 era. Also, I'm not saying that the 80 core processor will be released then, I'm saying their graphics solution will be.

Hoho, the same can essentially be said of the automotive industry. There are tons of crappy, and just odd cars out there, that no one really knows how to get the full potential out of. That doesn't mean there shouldn't be that many companies or that much competition.

Let's move off the topic of Intel's morality as a company. Too many people are emotionally involved with this specific argument, and will probably argue till their heads a-splode if we don't stop now.

Scientia from AMDZone said...

Here's good description of SOI Substrates.
Page 4:

the chips have far greater heat tolerance, easily functioning up to 160°C compared with bulk silicon’s 125°C inside the chip. This means that high power handling ICs can be created without heat sinks, further reducing both size and costs. Plus, they can operate correctly in very hot environments (under the hood of a car, for example).

Scientia from AMDZone said...

intel fanboi
but obviously biased for AMD. You pick out the data that supports your case and ignore the data that contradicts.


If you know of a roadmap that shows something different then give me a link and I may have to edit the main article.

What about vPro?

What about it? My understanding is that vPro is just Intel branding of the same technology you can get without the brand name. And, this has nothing to do with hardware.

I can think of a few good reasons why someone would be an AMD fanboi:

I can think of several reasons why your comments were a waste of time. For example, You didn't bother linking to any roadmaps that contradicted the ones I listed or refute the original assertion that there would be 3.2Ghz Conroe's on 65nm. Then you bring up items like vPro and Bearlake which have nothing to do with the article.

Scientia from AMDZone said...

wallachian
Intel and Microsoft's position, because they think these companies became successful by anti-competitive and monopolistic practices.


The Zilog Z8000 and the Motorola 68000 were both superior to the Intel 8088. Supposedly, one of the reasons IBM chose the 8088 was because Intel lied to them and claimed it would run CP/M.

Then IBM went to Microsoft for an OS but MS had absolutely nothing to offer since they had never written an OS and had nothing in development. So, IBM went to Kindall. However, when IBM ran into problems with Kindall, MS scrambled and bought QDOS off of Seattle Computer Products for $50,000.

Neither company got the IBM PC contract because of their own merits. And, since that initial contract both companies have engaged in monopolistic practices at various times.

But unfortunately the sad truth is that empires create stability in the region. Its much better to have 2 CPU manufactures

You can't give Intel any credit for this. The only reason AMD makes x86 processors is because IBM insisted on a second source.

Microsoft is considered a monopoly, as it is through Windows that it has stifled others like Netscape and many more.

MS killed Digital Research's GEM OS which was superior in every way to Windows. Perhaps I should mention MS's little trick of having Windows issue a phoney system error message if it detected DRDOS instead of MSDOS.

Intel is considered a monopolist yet it was made huge financial contributions to Katrina

Huge contributions?

Intel $1 Million

Amerada Hess, American Express, American International, BellSouth, BP Amoco, Boeing, Capital One, Eli Lilly, General Electric, Lowes, Merck - $1 Million each

Hewlett-Packard, McDonalds - $2 Million ea
Amgen $2.5 Million
Citigroup, Dow Chemical, JP Morgan Chase - $3 Million ea
Chevron, Coca-Cola, Johnson & Johnson - $5 Million ea
Exxon Mobile - $7 Million

Intel accounts for 2% of the contributions listed here.

Scientia from AMDZone said...

john
Also you claim that these features of vPro have existed for a long time. Can you show me the product out there that has


AMT,
ASF 2.0 - 2003

virtualization,

Hardware - AMD Pacifica; Software - available for X86 from two vendors for several years.

Gig Ethernet,

Broadcom and others

dual core

AMD had dual core before Intel

and security

AMD, same deal, ASF leverages virtualization and the no execute bit along with some software to provide security. In some cases AMD can do this much better with dedicated HTX encryption cards.

all in one package please?

An AMD system from several vendors.

Please be specific about how you define AMT.

How about what Intel said:

Rather than competing with the ASF standard, Intel says its Active Management Technology initiative, which enables IT enterprise management of client machines, even when they have not been booted, is complementary to ASF.

The ASF Specification defines the Remote Management Control Protocol (RMCP), which allows a network administrator to respond to an alert remotely in many different ways: powering on the system, powering off the system, or forcing a reboot (either to the default boot device or to an alternate boot device such as a Pre-Boot Execution Environment [PXE] server).

“Security is a critical issue in today’s enterprise,” says Andrea Westerinen of Cisco, vice president of technology for DMTF. “ASF 2.0 defines a four-phase session protocol, which includes discovery,
authentication, command transfer and termination. With its new authentication capabilities, ASF 2.0 delivers the remote management administrators need, with the security that organizations require.”

Scientia from AMDZone said...

ho ho
The thing that comes out in 09 will probably be based on similar architecture but will most likely be a lot faster. CPU's can get quite a bit faster in two years, especially with 1-2 manufacturing process changes (IIRC, intel was supposed to start producing 32nm CPU's in 09).


No. The description of the device for 2009 uses the same lightweight x86 processors. The real punch comes from additional vector units. It is supposed to be configured with a wide ring bus.

Ho Ho said...

Scientia
"AMD had dual core before Intel"
IBM had dualcore way before AMD ;)

Scientia
"No. The description of the device for 2009 uses the same lightweight x86 processors"

I know that. When I said similar I meant it would be similar by having loads of cores and fast interconnects. Indivudual core architecture will surely be something totally different compared to what they have now.

An interesting thing might be a hybrid Cell-like thing where you have x86 cores and lots of "dumb" wide SIMD cores. The latter ones would be much simplier than the x86 ones and provide lots more computing power. Sure, most people might not like to program for two relatively different architectures but it would offer a lot more performance than just regular x86 cores on the same die area.

Unknown said...

ho ho, the problem with that is not only that people would find it hard to program for 2 entire architectures, but they'd also find it extremely hard to program for the specific characteristics of the processor, like those who program for the cell do.

enumae said...

Analyst Sees AMD in Need of Capital

This is what I have been trying to make clear, Intel is attacking AMD's wallet, and it seems that it could be working.

Ho Ho said...

greg
"ho ho, the problem with that is not only that people would find it hard to program for 2 entire architectures, but they'd also find it extremely hard to program for the specific characteristics of the processor, like those who program for the cell do."

Well, in a way programmers already have to code for a different "architecture" when they want to use SIMD instructions. No compiler can do it automagically and efficiently. I know it is not as simple when you have to run your entire thread in a (slightly) different core but it would be much more efficient.

They could also do one really efficient x86 core and those others might be more optimized for SIMD stuff but still run general x86 code, though not as efficiently (worse branch prediction/software-only prediction, no OoO execution). That should allow to create more cores with same transistor budget and it should also give more performance, assuming your programs are decently multithreaded.

Of cource that would also mean that somehow OS should understand how to distribute threads between cores. One way would be that programmer tells OS that this thread would work nicely on such not-as-optimized core.

Scientia from AMDZone said...

ho ho

You can forget the notion that what is to be released in 2009 will help in terms of general programming; it won't. As I've already said, these are lightweight processors; the real punch is in additional vector units. The processors themselves are are relatively slow and weak. IBM's cell is already showing how poor an assymmetric design can be. Cell is nearly useless for most general purpose programming (which is why IBM's latest supercomputer is a hybrid using Opteron for general programming). Curiously, Cray has said the same thing about Opteron, that it is a better general purpose processor than their more specialized vector units.

IBM's Blue Gene L, which is built entirely out of lightweight processors, is very limited in what programming tasks it can handle. For many applications, Blue Gene is slower than machines with 1/4 of its theoretical performance.

enumae

A rationale would be nice. Why exactly would AMD need an infusion of cash in six months after it has already launched a new mobile chipset plus Barcelona? If the analyst said that AMD might need an infusion of cash next quarter I would take him more seriously since revenue is typically lowest in Q2 and Barcelona won't be out in volume yet. But his six month statement shows that he doesn't really know what he is talking about.

Roborat, Ph.D said...

Scientia said: "Why exactly would AMD need an infusion of cash in six months after it has already launched a new mobile chipset plus Barcelona?"

Because AMD's making huge losses. Lots of money out, few money going in. I don't understand how you think AMD's limited Barcelone release would make a difference and when you never thought Core2Duo in its simultaneous multiple segment releases wouldn't for Intel.

Again, you've under estimated Core2Duo and now you have so much hope on Barcelona.

AMD said it themselves, they'll only turn things around in 2008. That is WHY they need the money MID-2007.

enumae said...

erlindo - Scientia

This is an image from Google Finance comparing Intel and AMD since the launch of the Core architecture.

As I have stated, I understand what I am seeing and have read, but I am not capable of explaining my point, this image shows what is currently happening.

Investors know there is a pricewar and who is winning.

enumae said...

This is a little older, but shows what I have been trying to point out.

Link

george said...

AMD has the best deals at the low end and Intel has the best deals at the high end. this explains the difference in Revenue share and volume share selling 1 $1000 chip is approx. = 10 $99.99 chips. that means that AMD sells more cheap chips and Intel Sells more Expensive Chips.

Christian H. said...

Because AMD's making huge losses. Lots of money out, few money going in. I don't understand how you think AMD's limited Barcelone release would make a difference and when you never thought Core2Duo in its simultaneous multiple segment releases wouldn't for Intel.

Again, you've under estimated Core2Duo and now you have so much hope on Barcelona.

AMD said it themselves, they'll only turn things around in 2008. That is WHY they need the money MID-2007.



You guys never cease to amaze me. AMD had losses in Q4 because of a $550M charge for the ATi purchase.

It's statements like this that make me understand how wasteful ad inefficient US business is

Intel is still ripping people off with NetBurst and all I hear is how bad AMD is doing.

No wonder I forget to delete expletives.

Christian H. said...

AMD has the best deals at the low end and Intel has the best deals at the high end. this explains the difference in Revenue share and volume share selling 1 $1000 chip is approx. = 10 $99.99 chips. that means that AMD sells more cheap chips and Intel Sells more Expensive Chips.



What? What do you consider a good deal or the best deal? AMD has a guaranteed upgrade path for ALL chips high and low end. Intel will not.

Intel is still selling NetBurst for ridiculs prices which is tantamont to laughing at the people who buy them.


AMD has a stable set of CPUs with only around 10 desktop models.

Intel is "blinding people with science" and you "enhthusiasts" (more likely vicariously, self-conscious ninnies) are just helping them.

AMD will be fine. Nearly 1/3 of the desktop market means you won't need a Ch11 lawyer.


Go AMD!


ALL HAIL THE DUOPOLY!!!!!!

Fujiyama said...

I think that this quarter will tell what is the truth about AMD portfolio.
Latest processor price adjustments states that 92$ X2 3600+ is a good deal for AMD. Is it?
ASP doesn't matter - this is a price war. What matters now is when AMD is able to maintain or extend the market share. Every 0.01% lost will be counted as Intel win. Every 1% up should change AMD position to prove healthy business. This quarter results are critical. The income and the marketshare - units, units and once again units.
How about a small profit? :)

enumae said...

Maybe this will show that Intel's performance and price is attacking AMD's wallet...

AMD Athlon 64 FX-62 $389.99 OEM

AMD's top of the line AM2 processor has now been reduced to an almost mainstream price.

coldpower27 said...

I am going to have to disagree with you and say your misreading the road map for your own pessimistic projections on Intel's future.

Think logically scientia, when Intel changes the core of a processor, the name for that SKU will change.

Pentium 4 1.6, and Pentium 4 1.6A

Pentium 4 630 Prescott-2M, Pentium 4 631 Cedar Mill.

What that roadmap is saying is that there will be a Wolfdale derivative and Yorkfield derivative at those price points, but as to that actual clock rates and names, that is something that is undefined.

As well just because the 65nm versions are at those clockrates is no gurantee the 45nm will be the same, since Yorkfield is MCM and is cheaper to produce then Kentsfield because each die is smaller, then Intel simply could leave the 65nm versions prices as they are and only reduce prices on the 45nm versions.

Intel has done this before, the 65nm Cedar Mill's fell in price far earlier then the 90nm Prescott-2M ever did.

It simply states there will be Wolfdale and Yorkfield derivatives at those price points, we can't derive much more then that at this point.

I say from the looks of things the roadmap is simply incomplete. Look at Q3 2007 for instance with regards to the E6850, I don't believe Intel would suddenly discontinue the 1066FSB versions of the processor, so I would expect the E6800 to be there as well as the older E6700 and E6600's.

Anything from Q3 2007 is simply speculation at best, since it's too far too predict.

coldpower27 said...

enumae

I see where your coming from.

Athlon 64x2 5600+
http://www.mwave.com/mwave/skusearch.hmx?SCriteria=BA23445

But considering you can get basically the same processor with a newer F3 stepping at reduced TDP to 89W, who would want the older hotter version for far more costly. You aren't going to get away with 700USD when the 5600+ is selling as low as 325USD.

enumae said...

Scientia

Maybe this is the information needed to explain why AMD will need money later in the year.

Link

Christian H. said...

Maybe this will show that Intel's performance and price is attacking AMD's wallet...

AMD Athlon 64 FX-62 $389.99 OEM

AMD's top of the line AM2 processor has now been reduced to an almost mainstream price.




Maybe this will show how gullible you are.


The NetBurst ripoff.


Why aren't you shouting at the top of your lungs?

enumae said...

TheKhalif
Maybe this will show how gullible you are...The NetBurst ripoff...

Almost every comment you post is either Anti Intel or baiting somone to argue with you, my question is why?

----------------------------------

Please explain and answer the following questions.

----------------------------------

Why am I gullible?

Beacuse I read the news and can piece things together?

Because I do not let emotion rule my decision making or ability to see the other side of a debate?

I have recognized that I have not made my point clear, but I do believe that Intel effects AMD's margins and will be proven right or wrong when AMD releases its Q1 report.

If I am wrong I will gladly admit it, I am not here to be anti AMD, I enjoy debate and truth.

Do my opinions, points of views or interpretations relating to AMD effect you by noting that AMD might be in a financial situation that would be dirrectly related to Intel's pricing and agressive ramp towards dual and quadcore processors?

Does this upset you, and if so why?

----------------------------------

Why aren't you shouting at the top of your lungs?

Why would I?

I don't have loyalty to Intel.

If someone were to buy it, well, I would not feel sorry for an uninformed shopper, who, if they were to take a look online, would be able to see that Core 2 Duo is Intels new and better processor.

People all over the world make uninformed decisions everyday.

enumae said...

Scientia
This is almost identical to what Sun did with Opteron. Sun had a special version that clocked 200Mhz higher. It had different cooling and was warranteed by Sun. So, this doesn't really indicate headroom.

This is going back a little ways, and I appologize for it, but it made sense until I had reread it again today.

Your reference is a dual core and only a 200MHz increase in clock speed.

I do not know what type of cooling was used for Suns, but the link I had shown was using a standard BTX cooler, it is also a quad core, and the clock increase is about 600MHz, so we really can not relate the two.

So again, it would seem that Intel has some headroom for its quadcore parts.

Unknown said...

Coldpower, you are definitely reading the chart wrong. The chart has nothing about product EOL dates, and only shows when they focus on a specific product segment (a.k.a., when they are launched, and how long they are at their original pricing). If a part isn't on there, it's not going to be launched. Intel knew what it could and couldn't launch in q107-q307, by at least q106, so it's not likely they'd suddenly find out 1 quarter before their release that they can run them at a higher clockspeed. They'd either have to have really stupid engineers (which is pretty much an oxymoron) or a really bad marketing department (who thought it would be better to release at a lower tdp than necessary).

Enumae, you're talking about this like AMD had no idea it would be suffering losses in q406 or that it had not been planning for any losses then. Do you think the second largest processor manufacturer in the world wouldn't know how to plan their own business? I don't know how AMD is dealing with their losses, but I'm sure they knew they would happen, and had a way to deal with them.

As to the overclocking discussion. I'd like to point out that anandtech "conveniently" left out any thermal monitoring. They did have power usage, and we could extrapolate from that, but I really don't feel like it at the moment. You do notice that Gateway had to increase the wattage substantially though, and thus heat output as well. I'm also still of the opinion that overclock friendly chips are sent to the channel, and to discriminating retailers.

Also, the person in control of the price war is the one who reduces prices first. If I'm not mistaken, it has always been AMD that has been lowering their prices, despite increased demand. I don't know why they're doing this, and I don't know what exactly Intel's response will be, but it's obvious taht AMD feels it can lower the value of these chips and survive, despite Intel's pricing remaining completely stable.

coldpower27 said...

Greg
Coldpower, you are definitely reading the chart wrong. The chart has nothing about product EOL dates, and only shows when they focus on a specific product segment (a.k.a., when they are launched, and how long they are at their original pricing). If a part isn't on there, it's not going to be launched. Intel knew what it could and couldn't launch in q107-q307, by at least q106, so it's not likely they'd suddenly find out 1 quarter before their release that they can run them at a higher clockspeed. They'd either have to have really stupid engineers (which is pretty much an oxymoron) or a really bad marketing department (who thought it would be better to release at a lower tdp than necessary).

Like I have already re-iterated there is no mention of what clockspeed are going to be released for the Penryn derivatives on the 45nm node. So assuming that 2.4GHZ Quad Core is the highest for the mainstream, is all that is an assumption at this point. Anything that is Q3 2007 or beyond on this particular roadmap is very vague unless it's on know existing SKU's.

The chart also shows for some SKU's farther along then just their original pricing, the E4300 is shown farther along then just it's 163USD price point.

There is currently no data on the actual clockspeeds of the Penryn derivatives (in this roadmap) coming in at Q1 2008, all it is saying is that they are coming in at that time and occupying those particular pricing segments.

Keeping the Q6x00 series at the same price wouldn't be EOL for them, the Pentium D 8xx were kept in price for quite some time as well in the past officially.

The chart follows the pricing accurately of SKU's up to Q2 2007 but after that there is alot of speculation going on.

I am also of the opinion that the Core 2 Extreme Yorkfield will launch in December 2007, to make sure to say that Intel is shipping product based on 45nm technology by late 2007. This would jive well with how the first occurance of Presler XE is in Q1 2006 (in this roadmap), even if it was launched in Dec. 27th 2005, which is still Q4 2005.

There is simply too many inconsistencies in this roadmap for me to take it as any more then speculative.

coldpower27 said...

TheKhalif said...
Maybe this will show that Intel's performance and price is attacking AMD's wallet...

AMD Athlon 64 FX-62 $389.99 OEM

AMD's top of the line AM2 processor has now been reduced to an almost mainstream price.

Maybe this will show how gullible you are.

Why aren't you shouting at the top of your lungs?


Considering that Intel is selling the E6600 at the level what exactly is there to complain about?

Intel has introduced a lower priced Pentium D 9x5 line to fill in the lower 113 to 163USD price points.

Intel's processors don't tend to automatically reduce in price when they are still on the price chart and have an actual price.

enumae said...

Greg
Enumae, you're talking about this like AMD had no idea it would be suffering losses in q406 or that it had not been planning for any losses then.

I know they expected losses, but even AMD in the Q4 call said they were larger than expected.

Do you think the second largest processor manufacturer in the world wouldn't know how to plan their own business?

No, I believe that the factor overlooked by alot of people is chipsets. Intel was 80% of ATI's chipset business, and now it will start to really show up.

I don't know how AMD is dealing with their losses, but I'm sure they knew they would happen, and had a way to deal with them.

If they are greater than expected they might be in financial trouble though, wouldn't you agree?

... You do notice that Gateway had to increase the wattage substantially though, and thus heat output as well...

Substantially?

It does say they had to "jack up" the voltages, but they do not say how much.

Also, the person in control of the price war is the one who reduces prices first.

Only if the prducts were equal.

AMD has had to lower prices to remain competitive, just look at a previous post of mine showing the price of the FX62.

If I'm not mistaken, it has always been AMD that has been lowering their prices, despite increased demand.

Maybe the increase in demand is Dell, and not the channel.

I don't know why they're doing this, and I don't know what exactly Intel's response will be, but it's obvious taht AMD feels it can lower the value of these chips and survive, despite Intel's pricing remaining completely stable.

Again, AMD is responding to Intel, and from what claims have been made, AMD is trying to clear the build up in the channel by lowering prices.

Unknown said...

coldpower, penryn will only exist on the 45nm node, there will be no derivatives on a different node. Thus, what they show on that page is THE clockspeed that penryn will run at, unless they clock it lower. If they were planning on clocking it higher, they would have said so, because it would be bad publicity to site lower expectations. Also, while the lower end parts are vague (most likely because they have a smaller team of engineers working on them, and execution is not as important on this level) the high end parts are very specific as to what will run at what. And while the high end yorktown core does not list a specific clockspeed, seeing as the highest they plan (yes, we can assume this is what they're planning, because that's exactly what the map says) the high end part wont be running much faster (because that would go completely against sane market tactics, and be completely against Intel's current history of launches). I'm pretty sure now, that you're looking at a different chart than anybody else, because the chart I'm looking at that was linked directly on this comments board says exactly what clockspeed those cores would be produced at.

hoho, chipsets are an extremely low margin business, and while not having Intel will hurt them, it wont be as important as making sure their graphics card business does well.

While AMD says they were greater than expected, they don't say how much larger, and they don't say anything like, we will be turning this around immediately, or that it is unnacceptable performance on their part, and that they will do better immediately. If you want my opinion, I think the only reason AMD said they were larger losses than expected was because AMD didn't want to piss off the analysts too much (because they've taken offense to expected losses before).

Anandtech lists exactly how much power the gateway system uses, and the increase was substantial.

The increase in demand, is dell, hp, lenovo, founder, IBM, gateway, and the list goes on. Maybe the channel doesn't have increased demand, but it is still there. And I don't know where you're getting this idea of large part inventories in the channel. Last I saw, AMD couldn't keep their products on the shelves.

While their initial price decreases were due to Intel's performance lead, they kept dropping their prices, after many believed they'd hit pricing parity with Intel. Obviously their high end was not in parity, but everything else was. Since then, they've continued to drop prices.

Unknown said...

*edit

when I was talking to coldpower about the clockspeed of the high end penryn, I meant to say

And while the high end yorktown core does not list a specific clockspeed, seeing as the highest clockspeed they plan is 2.4 ghz(yes, we can assume this is what they're planning, because that's exactly what the map says) on the high end mainstream part the enthusiast part wont be running much faster (because not doing so would go completely against sane market tactics, and be completely against Intel's current history of launches).

enumae said...

Greg
hoho, chipsets are an extremely low margin business, and while not having Intel will hurt them, it wont be as important as making sure their graphics card business does well.

I believe you meant me, right?

It may be low margin, but it is still income which they no longer have. I know these numbers are incomplete (I can not find ATI reports for the gap between there Q3 2006 and the AMD aquisition), but the amount of money is very substantial...

ATI/AMD Chipsets and graphics...
Q2 2006 = $539.9 Million
Q3 2006 = $499.7 Million

If we assume that the $278 Million is 2 months of business and just add one more month we get $417 Million for Q4 2006.

Thats a loss of about 17% vs Q3 2006 and a loss of about 23% vs Q2 2006.

I think we can expect to see a continuation of this thru Q1 2007.

--------------------------------

Anandtech lists exactly how much power the gateway system uses, and the increase was substantial.

This is your previous comment... You do notice that Gateway had to increase the wattage substantially though, and thus heat output as well.

Your comment made me believe you are talking about processor voltages which were not shown.

The difference in power usage... The Gateway FX530 consumes quite a bit more power than our stock Kentsfield configuration, but keep in mind that it has two X1950 XTX graphics cards compared to a single 8800 GTX., thats quoted from the article.

--------------------------------

Maybe the channel doesn't have increased demand, but it is still there. And I don't know where you're getting this idea of large part inventories in the channel. Last I saw, AMD couldn't keep their products on the shelves.

Here is a link to an article on [H]ard OCP, click the link about DigiTimes, also note the projected price cuts.

While their initial price decreases were due to Intel's performance lead, they kept dropping their prices, after many believed they'd hit pricing parity with Intel. Obviously their high end was not in parity, but everything else was. Since then, they've continued to drop prices.

Again look at the link for DigiTimes, whether or not Netburst is competitive with K8, that is where AMD has to match prices, not Core 2 Duo, and it would seem that they are following Intels lead relating to price cuts.

One thing to watch closely will be on April 22 when Intel has the price cuts on Core 2 Duo's and what effect it will have on AMD the following weeks.

Unknown said...

Lol, ya, me and hoho have been having it out about whether or not it would be cool to own a quantum computer. Guess I just carried that over here.

I didn't realize Intel had been cutting prices on netburst. That does change a lot of what I've been thinking.

I'll have to grok this for a bit.

Joshua said...

Azary and Enumae! Come back to Rubyworks and post!

Anyhow, for some reason, Scientia, AMD workers is .35 percent of total blog traf? double check and follow up please.

enumae said...

What is going to happen in Servers?

I was thinking about AMD and the constant comments regarding drop in quad core capabilities.

I am wondering how many people have actually noticed or factored in that the drop in compatability is only present Scoket F and how short the time frame for adoption has been.

The reason for concern is that Socket F was released on August 15th, or about a month and a half after the release of Woodcrest (June 26th).

With Intel already having the ability for drop in quad core (not all motherboards), the continuation of a pricewar and the recapture of Server market share, why would people be buying 2P or 4P dual cores systems when you could have 2P quad cores for a much better price?

Here are some prices of parts I can find at Newegg for both Intel and AMD while trying to simulate comparable platforms with what they offer. I will need your help here, I do not work with servers.

- AMD 2P -
Opteron 2220SE $700 (2)
ASUS KFN4-D16/SAS $500.00
Kingston 4GB(2 x 2GB) $588 (8)

Total = $6,604

- AMD 4P -
Opteron 885 $1529 (4)
TYAN S4881G2NR $1587
SAMSUNG ORIGINAL 512MB $221 (16)

Total = $11,239

- Intel 2P -
Xeon 5355 $1215 (2)
Intel S5000XVNSAS $600
Kingston 4GB(2 x 2GB) FB-DIMM $644 (8)

Total = $8,182

I am strictly looking at this financially.

Intel's dual quad core cost about 30% more than AMD's 2P dual core, and about 34% less than AMD's 4P dual core.

- Power consumption -

AMD 2P = New 68W*2
AMD 4P = New 68W*4
Intel 2P = 120W*2

New = Recently released low power Opterons.

For applications that perform well over multiple threads, and also looking at total power consumption (not includeing FB-DIMM) is there a reason to go for the 2P or 4P AMD system vs Intel 2P quad core if (if I understand correctly) the majority of servers do not rely or utilize Floating Point where AMD has the obvious advantage?

Now that that is covered, what about the current change in market share, and the effects on AMD's margins?

What about AMD, as of now, not having a response to Cloverton in 2P?

How much more adoption of Intel's 2P quad core can AMD withstand to loose while their Server ASP's are declining and sales are flat?

The whole point of this post is to point out that the adoption of AMD's Opteron on socket F might be slower than people expect or believe.

How the drop in compatability comments are (in my opinion) over hyped after looking at the numbers and Intel having a similar ability.

Intels 2P quad core will effect the acceptance of socket F, and will then lead to a slower adoption of Barcelona than expected by some.

Your thoughts/opinions would be apprceiated.

Thanks.

enumae said...

Azary Omega
Why are you trying to spin intel here?

Actually if you look at the systems, I am grabing the most expensive parts for both Intel and AMD.

When I was proofiing it I had taken that comment out that said these were the most expensive parts, aparently I shouldn't have.

If I was spinning it for Intel, I could have lowered its price as well, don't you think?

Also, have you ever built a server?

No, hence my comment "I do not work with servers."

Do you know that because of a significant importance of the memory bandwidth Clovertown's perform on about same MHz/performance level as K8 in server environment?

Again like I had stated "if (if I understand correctly) the majority of servers do not rely or utilize Floating Point where AMD has the obvious advantage?", floating point is where AMD has an advantage due to bandwidth, not Integer.

What's your take on the other questions?

Scientia from AMDZone said...

roborat
I don't understand how you think AMD's limited Barcelone release would make a difference


The reason you don't understand is because you ignore what I actually say. It isn't just Barcelona. AMD has a new desktop chipset and will be releasing a new mobile chipset. These should help as well. Also, AMD's costs continue to drop as production ramps on 300mm and shifts to 65nm. When you ignore 75% of what I say I understand why you would be confused.

Scientia from AMDZone said...

enumae

That graph shows stock value. There isn't much I can get out of that. I've seen stock analysts make predictions for the most ridiculous of reasons and sometimes investors buy or sell for equally ridiculous reasons. I can't really get anything out of a subjective stock value.

Scientia from AMDZone said...

enumae

Okay, I read your second link which does explain why the blog author things there could be a problem. However, absolutely nothing firm is said. Several times it is stated that AMD's capex is on track. Also, I noticed that the people who suggested a cut in capex had to speak "off the record". I place about the same faith in off the record comments as I do in anonymous posts. There is nothing firm in the blog that says that AMD will have a problem. Also, the blog author only mentions funding from Germany; he does not mention money from selling 200mm tooling from FAB 30.

Scientia from AMDZone said...

Okay, some of you are still talking about Intel's mythical headroom. I could try to explain again about the fact that bulk silicon needs more margin than SOI. However, I have the past four years of Intel's own releases to back me up. If anyone here is seriously suggesting that Intel just hasn't felt like releasing faster clocks for the past four years then you are definitely chasing shadows in the mist.

The rumors were that Intel was going to release 3.2Ghz at launch of C2D last year. Now we see that there will be no 3.2Ghz chips even 18 months later. We also see that 45nm will not appear in Q3 07 as other rumors suggested. I do not know what the maximum clocks might be in 2008 but that has nothing to do with this article which is only about 2007.

Joshua

I'm not sure I understand your question about AMD employee traffic.

enumae

Some have suggested that people may be putting off AMD purchases until Barcelona is available. I have no idea if this is true or not. There is probably no way to tell this until we have Q3 numbers.

I'm sorry but your notion that AMD is going to keep losing ATI business is silly. AMD gets a big benefit by having AMD branded chipsets and by getting a new mobile chipset. I suppose it is possible that ATI could continue to decline but I don't think that is likely. It is much more likely that ATI sales will pick up due to desktop and mobile integrated chipsets. I've also seen it suggested that ATI was gaining against nVidia in terms of discreet graphics just before Q4. So, we'll see.

enumae said...

Scientia
AMD gets a big benefit by having AMD branded chipsets and by getting a new mobile chipset. I suppose it is possible that ATI could continue to decline but I don't think that is likely.

Ok, if ATI/AMD sold 80% of there chipsets for Intel, and we assume that AMD got the remaining 20%, how does ATI/AMD account for the 80%?

Is AMD big enough to offset the 80% loss?

If it can, it will be going against what it has been claiming about there continued use and development with Nvidia chipsets.

AMD will not be able to do both, and the release of its own chipsets will be a good thing, but the volume is not enough to offset the losses, at least not right away.

So I still believe that the income generated by ATI/AMD chipsets and graphics division will be down, and I gonna spit ball here and quess about another 5-10%, continuing to show the effects of the loss of Intel.

The release of the R600 will probably not be in any volume until next quarter, and the same goes for the R690G chipset.

I've also seen it suggested that ATI was gaining against nVidia in terms of discreet graphics just before Q4.

I am gonna go out on a limb and say I don't believe that happened, not with all the fuss made about the 8800 and price reductions as a result, but like you said, we will see in about a month and a half.

Scientia from AMDZone said...

enumae
Ok, if ATI/AMD sold 80% of there chipsets for Intel


The only place I've seen that 80% figure is the INQ. Do you have another source? I still have no idea where you get the "sky is falling" mentality about ATI.

AMD managed to retake 2.1% of the market in the fourth quarter.

AMD nevertheless enjoyed an increase in its share of the discrete desktop graphics market from 43% to 46.2%. Nvidia's share of that market shrunk from 57% to 53.8%.

AMD's share of the discrete mobile segment shrunk from 47% to 40.9%.


So, are you claiming that ATI has already taken big losses or that these losses haven't occured yet? I would assume that the new mobile chipset would help with the drop in discrete mobile.

enumae said...

Scientia
...Do you have another source?

Aparently that is where I had read it aswell, I thought it may have been on The Reg., DailyTe. or some other place, and well, since I find The Inq. to be more of a blog with the newest headlines, the 80% is void.

I still have no idea where you get the "sky is falling" mentality about ATI.

I am not implying that the sky is falling, my post are just a continuation of my original claims that Intel is hurting AMD financialy.

Whether it be from no longer using chipsets or a price war, Intel has a direct effect on AMD's finances.

------------

Sorry, I had misread the Jon Peddie Research report.

AMD gained in Desktop, but lost in Mobile.

So, are you claiming that ATI has already taken big losses or that these losses haven't occured yet?

I feel that ATI has taken some losses, but not all of them.

Again we will find out alot more when AMD releases it Q1 2007 report.

I would assume that the new mobile chipset would help with the drop in discrete mobile.

I was under the impression that a mobile chipset's main purpose was to Integrate the GPU while reducing power and not have to rely on a Discrete GPU, so I do not see how the mobile chipset would increase Discrete sales.

Wasn't this the reason AMD aquired ATI, was to be able to offer a platform to the Enterprise space, and wouldn't this be the most likely place for Integrated GPU's being that they are cheaper?

Again I appologize for the faulty information from The Inq..

Scientia from AMDZone said...

Sorry, my last post was a bit mixed up. Integrated mobile is far more important than discrete mobile. So, I see AMD's new mobile chipset as important in taking mobile share. I do not consider discrete mobile to be as important.

Both discrete and integrated desktop are important but integrated will gain them more commercial business.

Roborat, Ph.D said...

Scientia from AMDZone said...

The reason you don't understand is because you ignore what I actually say. It isn't just Barcelona. AMD has a new desktop chipset and will be releasing a new mobile chipset. These should help as well. Also, AMD's costs continue to drop as production ramps on 300mm and shifts to 65nm. When you ignore 75% of what I say I understand why you would be confused.


You need to read your post and tell me if I ignored 75% of what you said. You spoke of K10 95% of the way, mentioned chipsets with one sentence and the others, not a mention.
AMD has a new chipset, so what? so does everyone else.
300mm/65nm ramp, you fail to recognize that AMD has yet to go through the partial Fab shutdown and the massive $1.3B depreciation cost this year for running 2 Fabs.
Cost continue to drop? If you've ever worked in the Fab you'd know that cost always continues to drop. CEO's just like to say it as if its a new thing. From what i'm seeing, Intel has more aggressive plans for cost reduction at $1.5B savings this year on top of the normal "cost continue to drop" due to technology progression.

Bottomline, AMD has nothing significant to offer this year that will turn things around. So far all we have from them are powerpoint presentations and bogus benchmark PDFs.

Roborat, Ph.D said...

Scientia from AMDZone said...
Integrated mobile is far more important than discrete mobile.

You're putting too much financial hope on products that ATI was losing massive money on.
Without the discrete graphics price fixing arrangements these fabless-low margin business are going to be a pain for AMD's financials like we already saw in their last CC.
Graphics and chipsets aren't a good match for expensive R&D, Capex and state of the art Fabs that AMD owns.
You're pretty much on your own to think that this is good business for AMD. ATI will be as useful as Intel's chipsets division in the long run.
ATI good for designing future products? Now we're talking sense.

Scientia from AMDZone said...

roborat
AMD has a new chipset, so what? so does everyone else.


Everyone else being who? AMD is the only other major manufacturer of X86 processors.

300mm/65nm ramp, you fail to recognize that AMD has yet to go through the partial Fab shutdown and the massive $1.3B depreciation cost this year for running 2 Fabs.

You are aware that depreciation isn't a cost, right? Depreciation is only used for asset and tax accountability; it doesn't cost AMD any actual money. The conversion won't start until the new bump and test addition opens.

From what i'm seeing, Intel has more aggressive plans for cost reduction at $1.5B savings this year on top of the normal "cost continue to drop" due to technology progression.

Which will be matched by AMD at the same ratio.

Bottomline, AMD has nothing significant to offer this year that will turn things around.

That has got to be the clearest example of ignorance on your part that I have yet witnessed. So, a new architecture is not significant? Amazing.

So far all we have from them are powerpoint presentations and bogus benchmark PDFs.

I haven't seen any benchmarks, bogus or otherwise. Where did you see benchmarks?

Scientia from AMDZone said...

roborat
You're putting too much financial hope on products that ATI was losing massive money on.


Actually, no, I'm not. Chipset products are only one piece.

Without the discrete graphics price fixing arrangements

What are you talking about?

these fabless-low margin business are going to be a pain for AMD's financials like we already saw in their last CC.

and wouldn't this be doubly true for nVidia? I don't see the logic.

Graphics and chipsets aren't a good match for expensive R&D, Capex and state of the art Fabs that AMD owns.

Then you had better immediately run to Intel and tell them to stop producing chipsets on their 300mm FABs because it isn't a good match.

You're pretty much on your own to think that this is good business for AMD.

Right, just me and the AMD Board of Directors.

ATI will be as useful as Intel's chipsets division in the long run.

And, also in the near term.

ATI good for designing future products? Now we're talking sense.

That's 2008; AMD will benefit in 2007.

Unknown said...

Roborat is talking about the DOJ's little huff with Nvidia and ATI about them price fixing, which really doesn't make any sense. Nvidia price fixing with ATI would be the least Jen-Hsun Huang thing anyone could do. Besides, have you looked at graphics card prices? I don't think they've really changed at all, and yet the graphics cards keep getting better, and start to have a more expanded price point. All signs of a market that isn't suffering from price fixing.

enumae said...

An article which is implying that AMD will be looking to raise operating cash.

Link

"Ruiz also said AMD will continue to pursue options for raising operating cash."

Scientia from AMDZone said...

Ruiz doesn't give any details. I have no idea how much cash he is talking about and if this would include the $250 million from selling FAB 30 tooling.

enumae said...

Scientia

I understand you have a different opinion than I do on AMD's current financial situation, but here is some information that should help make my view clear that AMD is in need of money.

Assuming that the loss of Intel's business for chipsets is complete and that AMD's ASP don't continue to decline despite the recent price cuts on Server and Desktop...

AMD/ATI 2006 = $7.6 Billion

Now if we assume that AMD/ATI will make about $8.0 Billion for 2007.

AMD/ATI 2007 Capex on fabrication and infastructure $2.5 Billion.

AMD/ATI 2006 cost of revenue was about $4.2 Billion.

AMD/ATI 2006 Operating expenses $3.2 Billion.

So for 2007 we have a total of about $9.9 Billion, and an expected $500 Million in negative cash flow (as stated in the prior link).

Now if we factor in a decrease on Cost of revenue of 33% ($4.2 Billion * 0.66%) due to 300mm wafers and 65nm, we are now at $2.7 Billion, and this puts us right at the $8.5 Billion, or $500 million in negative cash flow.

This doesn't fator in...

1. ATI chipset business for 2007 will be without Intel for the whole year, not just a 2 quarter decline.

2. The fact that Intel has about 3 - 4 more months to capitalize on the Server, Desktop, Mobile segments without any real competition from AMD in regards to new products, as well as the oportunity to regain market share.

3. Intel pursuing the pricewar further.

I understand AMD/ATI has a new chipset, processor and graphics card scheduled for 2007, but it doesn't look good for AMD/ATI entering or leaving 2007 financially.

I am not saying AMD is going to bankrupt or any nonsense, but it does look bad financially, there is a possibility of AMD not making a profit in 2007.

From what I am looking at, it would seem pretty clear that AMD does indeed need to raise money.

Unknown said...

Enumae, we wont see a "decline" in any sense with ATI's chipset business. They have no chipsets that they haven't already sold that are Intel compatible that Intel will allow anyone to install in their systems. As such, they've already lost all of their Intel business, and have nothing more to lose on that front.

Maybe AMD would lose money in 07, but like you said, they aren't going bankrupt. Even if they did, we really don't have anything to worry about, because it's not like they can essentially cease to exist. Someone or something will have to step in to take over AMD or help them out a lot if something like that, due to the fact that without AMD, Intel is an illegal monopoly.

enumae said...

Greg
As such, they've already lost all of their Intel business, and have nothing more to lose on that front.

ATI did carry an inventory of Intel chipsets and I believe the inventory average is about 45 days (some help here from someone who knows would be appreciated), and with that inventory they could have made it through about a month and a half of the two months with AMD in Q4 2006.

This is another reason why I believe AMD/ATI chipset revenue will continue to decline compared to Q3 and Q4, their financial numbers have not shown the depletion of inventory.

With that in mind, we will not know until the Q1 2007 earnings are reported since the inventory will be depleated.

Joshua said...

Intelers always love talking up storms while doing nothing, wonder how they get away with that.

Roborat, Ph.D said...

Greg said “Maybe AMD would lose money in 07, but like you said, they aren't going bankrupt. Even if they did, we really don't have anything to worry about, because it's not like they can essentially cease to exist.
Have you heard of a semiconductor company that turned things around after it went bankrupt? Do you honestly think that AMD will execute and compete better while in chapter 11 with operational oversight from financial institutions and nearly zero investments for future products?
At the moment, only a fool would think that AMD will go bankrupt. It just needs to stop competing with Intel to become profitable just like in the early 90’s. I don’t understand why AMD even started thinking that they can do better.
If you think I’m crazy, look at AMD’s books. They started making losses when they started selling Athlons and said it was better.

Scientia from AMDZone said...

roborat

I've given up on the idea that Pointer will ever make any real contribution here. Pointer is a fraud who talks big but doesn't seem to have any real technical knowledge or understanding to back it up. Now, I'm wondering about you.

For example, in spite of the dozens of times that I have countered and corrected Sharikou's posts and comments you still insist on trying to lump the two of us together. Why? Are you that blinded by bias that you can't understand that there is a gulf between Sharikou's ideas and mine?

The most comments you've ever received for anything you've written is 4. Out of those four: 1 was from you, 2 were from me, and the last one was from some anonymous poster telling me not to post there. In simple terms your blog is a joke; is there anyone at all who refers to your blog for information? Have you ever been linked?

I know you like to post on Sharikou180's blog. I assume this is simply because he tends to be very pro-Intel. However, I don't believe any of his articles have ever been linked and he only averages about 11 comments for article. Since he posts twice as frequently as I do this would be equivalent to 22 comments for me.

My average is 100 comments per article and I've been linked to many times. If I were simply trying to promote AMD then why would I have a link to an Intel employee blog on the first page? Just how fast and how far do you want your credibiity to fall?

Scientia from AMDZone said...

enumae

I started to compose an answer to your comments but it started getting long. Which means that I may to have to revisit a comparison with 2003 and do an article.

Unknown said...

Roborat, you do realize how ludicrous what you said is? AMD just can't stop competing with Intel. By existing and selling product, they compete with Intel. By making money on the same product they compete with Intel. By upholding American law, and not colluding or cooperating with Intel, they compete with Intel. They can't just quit competing with Intel unless they cease to exist. AMD didn't quit competing with Intel in the 90s, Intel just had so much control of the market in the 90s, that only people who were consistent and informed consumers would buy AMD. AMD had simply shrunk so much as a company that Intel didn't care.

However, ignoring everything above (if that's even possible) if AMD quit trying to compete with Intel in the way you suggest, they'd be exactly where they are now in terms of product because of development cycles, would only have 1 fab for the forseeable future, would have no platforms to compete on, no ability to move on to their modular core design in 08, and no consistent and stable customer base. Do you think it's intelligent, nay, possible to compete with core2 and all of Intel's future products without any of these? Do you think it would've been healthier for AMD to have competed with core for the last few months without any of this coming afterward to at least make up for their hard times right now? Do you think the channel would have even bothered if AMD hadn't cur prices, come out with newer but hotter products, and maintained their same product line, but simply transitioned it to 65nm. Because this is what you (and many analysts) are implying they should have done, and would not be a good idea at all.

enumae said...

Scientia
...I may to have to revisit a comparison with 2003 and do an article.

As I have stated in previous post, I have not followed the segment very long, and since I find your post to be insightful, I look forward to your view.

Scientia from AMDZone said...

enumae
With that in mind, we will not know until the Q1 2007 earnings are reported since the inventory will be depleated.


Are you suggesting that ATI has stopped supporting Intel platforms? I'm sure you know that discrete graphics cards work with either AMD or Intel. But, lets see what AMD currently supports in terms of Intel Integrated chipsets and Intel Mobil chipsets.

So, why would the inventory be depleted if AMD is continuing to make Intel chipsets?

enumae said...

Scientia
Are you suggesting that ATI has stopped supporting Intel platforms?

No.

...But, lets see what AMD currently supports in terms of Intel Integrated chipsets and Intel Mobil chipsets.

I understand that AMD/ATI will continue support for Intel platforms, but I also believe that will be short lived, just as Intel does not produce a chipset for AMD.

Why would Intel have any intrest in helping AMD/ATI by allowing them to develop a competing chipset?

Do you believe that Intel is going to send AMD an enginerring sample of a new processor so that they may work on a new chipset?

So, why would the inventory be depleted if AMD is continuing to make Intel chipsets?

I was under the impression that ATI was making chipsets for Intel, in the sense that Intel could not produce enough chipsets, this is the business I believe AMD/ATI will be losing.

Roborat, Ph.D said...

Scientia wrote: … you can't understand that there is a gulf between Sharikou's ideas and mine?
Apart from the idea that “AMD will have a good 2007”, I never mentioned that you and Sharikou ever agreed on anything else.

The most comments you've ever received for anything you've written is 4... In simple terms your blog is a joke; is there anyone at all who refers to your blog for information? Have you ever been linked?
Wow! Now that’s really mean and hurtful.

I know you like to post on Sharikou180's blog... However, I don't believe any of his articles have ever been linked and he only averages about 11 comments for article. Since he posts twice as frequently as I do this would be equivalent to 22 comments for me.
I’m sure that’s very hurtful to Sharikou180 too. I pray that he never reads this.

My average is 100 comments per article and I've been linked to many times. If I were simply trying to promote AMD then why would I have a link to an Intel employee blog on the first page? ...
I'm really happy for you.

enumae said...

I have a bad link, the last one should go to X-Bit Labs.

Unknown said...

Roborat, you constantly lump scientia and sharikou together on 180's blog, there is no reason to lie about it.

To reference an amazing quote from one of the owners of Google "If you don't like what you see in the mirror, replacing the mirror wont fix it." I think this perfectly reflects on your response to scientia's criticism.

I don't think scientia should have bothered comparing himself to 180. While I value 180's articles, they are short, and not very in-depth. However, scientia, you might show a bit more humility (though I suppose you've earned some bragging rights).

Scientia from AMDZone said...

I'm not bragging. I could stand here all day and proclaim that I have the greatest blog on the planet but my opinion would be worth almost nothing. I mention the comments and links because that is an indication that is not based on my opinion.

Let me put this a different way. The comments here are good because other people choose to come here and post. Obviously the comments would be worthless if I were the only one posting.

Roborat, I'm saying the same thing now that I said when I posted on your blog. Your blog is set up as a parody of Sharikou's and that is a mistake. Your blog should be your own and your own ideas, not just something that is anti-sharikou or anti-AMD. And, my respect for Sharikou180 was certainly dented when he started doing spoof posting. He has now done three posts about spoof posting; he might be amused by this but there is no substance. Again, these are not his own ideas; he is just reacting to others.

I think it is just as bad to be blindly pro-AMD as it is to be blindly pro-Intel. I'm not looking for posters who just want to pump up AMD. What I respect is an opinion that has some basis in fact and a point of view that can change with the real world.

In all honesty, I have very little right now to criticize about Intel. They seem to be on the right track. I think it would be good if they could increase compatibility on their chipsets but that is not a huge problem.

Obviously, AMD's cash position is not as good as it could be but again just how much of a problem is that? If it doesn't hurt AMD's ability to design new processors and doesn't hurt their capex then does it matter?

Scientia from AMDZone said...

enumae

Some industry observers noted that Intel could pressure its partners among mainboard and computer makers to make them stop using Intel-compatible chipsets from ATI. According to an editor of TweakTown web-site, Intel has put pressure on mainboard makers so that they cancelled release of motherboards based on ATI’s next-generation Intel-compatible code-named RD600 chipset.

Sounds like the definition of a monopoly to me.

Intel Fanboi said...

Well it looks like there has been an official comment from Intel about the speed bins that will be produced this year. Interpret the comment how you want to: http://www.theinquirer.net/default.aspx?article=37772

enumae said...

Scientia
Sounds like the definition of a monopoly to me.

Who's talking about a monopoly?

I was talking about the loss of Intel business, and how it will effect AMD/ATI business in the current quarter.

It is widely understood that Intel has some, shall we say, agressive business tactics.

So while your point is noted and agreed too, I do not see a connection to my previous points and questions.

enumae said...

It seems Intel may have some headroom.

Scientia from AMDZone said...

Intel Fanboi said...
Well it looks like there has been an official comment from Intel about the speed bins that will be produced this year.


What the INQ actually says is:

He said that Intel will be able to produce better bins and bus rates during this year.

The faster bus matters mostly on quad core but this probably means getting the 666Mhz FSB's up to 800Mhz. I doubt very seriously that Intel will be able to match AMD's 2132Mhz but Intel's greater cache size will help somewhat. Better bins is too vague. This could mean faster clocks or it could mean better yields. We'll see if Intel actually releases a 3.2ghz Conroe.

He also said that Intel was likely to disclose its CSI plans at the Beijing April Developer Forum.

Now this is something to look forward to. This will indeed be official information.

Roborat, Ph.D said...

Scientia said from previous article: "2007 shouldn't be bad for either company"

Lehman says otherwise.
Although the microprocessor (MPU) industry can look forward to a strong performance in the coming months, the struggles for Advanced Micro Devices Inc. (AMD) are not about to let up anytime soon, one firm believes. ... Longtime market leader Intel is now firmly back on top, Lehman said...
http://www.edn.com/article/CA6418172.html?ref=nbednnenews&industryid=2815

Scientia said:
Worse still, even with these reduced speeds, 45nm has been pushed back to Q1 08


Now Intel pulls in Penryn to 2H2007. So that basically throws your "Giants in the Mist" article out the window.
http://www.dailytech.com/Intel+Pulls+45nm+Xeon+Launch+Into+2007/article6185c.htm

It appears I don't need to look very far to find how many of your already vague and unquatifiable claims are completely off.

One can already deduce that the more expensive, lower yielding Barcelona will not be able to charge a high enough ASP competing with Penryn. So it appears that AMD's undemoed 2007 savior will be too little too late.

Scientia from AMDZone said...

First, let me correct the speeds. For Intel it should be 1333Mhz and 1600Mhz versus AMD's current 1600Mhz and Barcelona's 2132Mhz.

roborat

Actually no. To throw my Mist In The Morning Sun article out the window Intel would need to release a 3.2Ghz Conroe on 65nm and would need to deliver 45nm in volume on FAB 32 in Q4 07. I am not denying that Intel could start production on D1D in Q4.

I'm not sure exactly what to say about the Lehman's article. They have apparently downgraded AMD before and they are only one company. They also seem to have the false assumption that 45nm will be available in the first half of the year.

enumae said...

ashenman
Both are monolithic dies and thus share the same yield disadvantages.

Actually they are not, Penryn Xeons will not be monolithic.

As is, I see the immediate advantage being to AMD.

If there is an advantage here regarding yields I believe it would belong to Intel due to the fact that they will continue the use of MCP, while AMD will be using a monlithic design.

Christian H. said...

Scientia from AMDZone said...
Ruiz doesn't give any details. I have no idea how much cash he is talking about and if this would include the $250 million from selling FAB 30 tooling.



From this interview from the 19th Hector is saying that rumor was BS. AMD doesn't need venture capital.


CRN

bk said...

dr yield
What about Pennryn? Based on aspect ration in web images and reported die size of 110mm^2, I expect it to be ~8.7x12.6mm for 557 gross dpw. Using the same assumptions, .3/cm^2 defect density yields at 72.6% and 404 net dpw. That's 6 yield points and almost 270dpw more than Barcelona at half the defect density.

First off you are assuming Intel's 45nm process will have comparable yields to AMD's 65nm. I doubt that will be the case. Second, you are ignoring the fact that Intel needs two pennryns to equal one barcelona.

I would also be curious as to whether there is any yield loss when Intel pastes the two dual cores together.

Roborat, Ph.D said...

thanks enumae and doc yield...

Any AMAT engineer will also disclose to you that most non-intel Fabs like AMD are currently struggling to get meet 45% yield with 65nm. And were only talking about single/dualcore parts.
The Penryn pull-in suggest very healthy yields for intel's 45nm node. Mind you, Intel is a cost disciplined company that's very strict with technology certification which requires yield's/cost-per-die targets as a gating requirement.

SOI wafers are also significantly more expensive that bulk silicon.

There is some truth in Hector's statement that AMD will not make significant volume for Barcelona in 2007. Intel was always been correct in predicting that monolithic Quad Core only becomes feasible (at least for intel's cost targets)at the 45nm node.

Unknown said...

Yay roborat quoting unnamed inside sources that he has no way to back up. Good job proving scientia right.

dr. yield, thank you very much for your information. Unlike roborat, you are capable of citing resources for you info, and I definitely respect that.

bk does raise a good point about requiring two penryn cores per quad core. However, they do have a very strong yield advantage with MCP, which doesn't have any real effect on yields, except making it so that one core is not dependent on the yield of another. Intel also has the advantage of being able to place higher clocking cores together, instead of hoping all four cores on one die end up clocking high enough.

BTW, I'm also ashenman, but accidently used the wrong sign in for my last post (I used to have two accounts, because this computer used to have two users).

Scientia from AMDZone said...

roborat

I'm going to give you the benefit of the doubt and assume that you believe what you post is true.

Any AMAT engineer will also disclose to you that most non-intel Fabs like AMD are currently struggling to get meet 45% yield with 65nm.

Your statement would be correct if you replace the word "like" with the word "except".

SOI wafers are also significantly more expensive that bulk silicon.

This is true but it is also true that this adds very little to the actual cost of the chip.

There is some truth in Hector's statement that AMD will not make significant volume for Barcelona in 2007.

This is quite true; AMD will target the server market with quad core.

Intel was always been correct in predicting that monolithic Quad Core only becomes feasible (at least for intel's cost targets)at the 45nm node.

And yet, this is obviously wrong. If AMD could make a profit from dual core on 90nm on a 200mm FAB then a competitor would have to be seriously incompetent not to be able to make a profit with quad core on 65nm on a 300mm FAB.

Scientia from AMDZone said...

dr. yield
Intel yield is 72.6% vs. AMD's 66.4%. Numbers used were hypothetical, but probably not far from reality.


I agree; these should be pretty good ballpark numbers. Interestingly, when I did a quick die fitting for Barcelona on my calculator I got 204 dies per wafer.

However, you have to add 70mm^2 for the quad FSB Northbridge. This reduces the Intel placement to 199 / wafer. However, because the yield is better for the smaller northbridge you end up with about 207.

you get 202 4core Pennryns vs. 136 Barcelonas.

After allowing for the northbridge the yield is 150 Penryns per wafer.

Net result, at a 2X defect density, the smaller die size and multi-chip approach yields an extra 49% saleable CPUs (66).

extra 10% saleable CPU's (14).

Assume a 2% higher yield loss at packaging than the monolithic approach, and you still get 62 more CPUs.

you get 11 more CPU's.

At an ASP of $500 for high-end parts (both AMD/Intel), that translates to an exra $31,000 of revenue per wafer.

an extra $5,500 of revenue per wafer.

Roborat, Ph.D said...

Greg said...
Yay roborat quoting unnamed inside sources that he has no way to back up.

You need to check how ridiculous your request is. The proof of AMD’s larger die costing more is an industry known principle that governs all semiconductor manufacturing. It’s like asking me links to Moore’s Law to prove it’s correct.

As for yields, the link below is just a general view on 65nm yields.
http://www.fabtech.org/content/view/2462/

> SOI wafers are also significantly more expensive that bulk silicon.
Scientia said: This is true but it is also true that this adds very little to the actual cost of the chip.
I’m sure AMD & Intel would love this statement to be true. Can you please tell us which variable cost that’s part of the chip is the most expensive? I’m sure the current shortage in worldwide supply of silicon wafers is making your statement more truthful. I’m pretty sure you have no idea how much both pays for wafers per quarter and comparing it to their current quarterly profits as a percentage, I can tell you its very significant at least for AMD.

This is quite true; AMD will target the server market with quad core (in 2007).
Yet another one of your vague statements that can be misleading. The server market is pretty big. What percentage of that server market are you specifically talking about? According to Hector it appears like all that AMD plans to do with K10 this year is demo and take orders for next year. If you listen to the CEO double talk, it’s pretty safe to say it’s less than 1% of AMD’s server output.


> Intel was always been correct in predicting that monolithic Quad Core only becomes feasible (at least for intel's cost targets)at the 45nm node.
And yet, this is obviously wrong. If AMD could make a profit from dual core on 90nm on a 200mm FAB then a competitor would have to be seriously incompetent not to be able to make a profit with quad core on 65nm on a 300mm FAB.
Comparing Intel’s & AMD’s cost structure is apples to PC. The difference in economies of scale between Intel and AMD is complex enough to make an accurate analysis. I’m sure Intel can generate a profit from monolithic Quad Core with 65nm but I bet Intel can generate more profit otherwise (baring of course performance issues). Unless Intel sees a switch over in cost benefits, according to their standards, it isn’t an option. It doesn’t make business sense.
As for AMD, it’s obvious now that with their limited capacity, ramping Barcelona in 65nm will not be in their best interest profit wise. I’m not sure if they just realized that late or they made a conscious decision to implement and show capability without ramping for performance and design wins purposes. One thing’s for sure and is confirmed by Hector that they’ll have to wait for 45nm before significant a ramp. And this goes back to my original statement that Intel was correct in its prediction.

Scientia from AMDZone said...

roborat
As for yields, the link below is just a general view on 65nm yields.
http://www.fabtech.org/content/view/2462/


Yes, and that link does not apply to Intel, AMD, or IBM. It does apply to foundries and DRAM manufacturers. Comparing Intel, AMD, and IBM to foundries or DRAM manufacturers is beyond silly. This is even mentioned in the same article:

However, within Kispert's comments we should bear in mind the likes of Intel and AMD, as his comments referred to 65nm in general. I have seen many nice charts from Intel, among others, over the last few years, noting that the yield learning curves are actually improving, node on node, allowing for faster volume ramps.

How did you miss this?

Can you please tell us which variable cost that’s part of the chip is the most expensive?

I'm certain that the largest cost is the cost of the cleanroom and tooling. After this cost the next should be testing and packaging.

Semico analyst Joanne Itow calculates that when all manufacturing expenses are considered, including packaging and testing, SOI adds only about 5 percent to per-chip costs. "[SOI's cost] really gets diluted when you look at all the other costs that are loaded onto those chips," she says.

Give me a link that shows otherwise.

I’m sure the current shortage in worldwide supply of silicon wafers

Maybe you can post a link that talks about the shortage. For example this is from September 2006:

Gartner now confirms the whole issue was just another Wall Street tout. There never was a shortage and, with new supply quickly coming on-stream, there won't be a shortage for the foreseeable future.

I hope that you can post something to show that this whole thing isn't just some ridiculous fanpiece.

Yet another one of your vague statements that can be misleading. The server market is pretty big. What percentage of that server market are you specifically talking about?

The mid to upper server segment, roughly 2/3rds of the server chip volume.

According to Hector it appears like all that AMD plans to do with K10 this year is demo and take orders for next year.

Were you half asleep when you read this supposed statement by Ruiz? AMD will ship millions of K10's in Q3 alone.

If you listen to the CEO double talk, it’s pretty safe to say it’s less than 1% of AMD’s server output.

This statement is so far off I'm now having serious doubts about your contact with reality.

Comparing Intel’s & AMD’s cost structure is apples to PC.

You know, I find it strange how you have such profound knowledge that you can confidently state that K10 will be less than 1% of servers and now you are claiming that you don't know enough to compare Intel with AMD. The truth is that if Intel did not use MCM their cost per die would be higher than AMD's.

As for AMD, it’s obvious now that with their limited capacity, ramping Barcelona in 65nm will not be in their best interest profit wise. I’m not sure if they just realized that late or they made a conscious decision to implement and show capability without ramping for performance and design wins purposes.

I'm sorry but you don't seem to understand the technical aspects. The slower ramp is due to a newly designed transistor and new die mask, not due to economics. Do I have to remind you that Intel encountered the exact same thing with C2D?

One thing’s for sure and is confirmed by Hector that they’ll have to wait for 45nm before significant a ramp.

No, AMD did not say this. You are confusing quad core on the desktop with server. Desktop requires more volume which won't be available until 2008. However, this volume comes up before 45nm is released.

Scientia from AMDZone said...

robo

You are also failing to take FAB account into account. Remember that FAB 30 does not begin 300mm production until 2008. AMD cannot produce large volumes of quad core until FAB 38 comes online without cutting into other production.

enumae said...

Scientia
However, you have to add 70mm^2 for the quad FSB Northbridge. This reduces the Intel placement to 199 / wafer. However, because the yield is better for the smaller northbridge you end up with about 207.


Why are you bringing up the northbridge?

Doesn't AMD need one too?

When has Intel produced a northbridge at the same node as the processor?

I was under the impression it was at least one node behind, 65nm processor with a 90nm northbridge.

enumae said...

The 40% advantage that AMD's Barcelona would have was based on a clock to clock comparison, right?

Here are the intitial clock speeds for Barcelona.

Unknown said...

Hector stated 40-70% performance advantage compared to Intel's current server offerings, he did not specify clock to clock or anything like that, but just said it was an overall performance gain on a wide variety of tasks. As such, I'm pretty sure this means its based on current price points that Intel offers, and that if Intel didn't change their prices to match their performance when Barcelona is released, AMD's processors would outperform them that much comparatively. It seems like the only reasonable conclusion that could be made, if you only look at what he has been saying, and you ignore what we know about the processors (as we likely have a relatively incomplete picture).

Scientia from AMDZone said...

enumae
Why are you bringing up the northbridge?


Because Intel makes these too.

Doesn't AMD need one too?

You've already included AMD's northbridge in the estimate; it's on the cpu die.

I was under the impression it was at least one node behind, 65nm processor with a 90nm northbridge.

You can figure it at the next larger process size if you like however the numbers will be worse because the higher yield does not make up for having a die twice as large.

enumae said...

Sorry, I got north and south confused, in regards to what they do.

My point is the yields discussed were only in regards to CPU's, and that you bringing up chipsets has no relevance to the points laid out by Dr. Yield, PhD, MBA.

It is understood that Intel has no IMC, but in regards to yields while using MCP Intel should be better off per wafer in a direct comparison to AMD per wafer.

sharikouisallwaysright said...

Something interesting to all the doomsayers:
http://www.investorvillage.com/smbd.asp?mb=476&mn=36770&pt=msg&mid=1504465

AMD-Chipsets plus 40% comnpared to q4/05.

enumae said...

sharikouisallwaysright
Something interesting to all the doomsayers...AMD-Chipsets plus 40% comnpared to q4/05.

AMD/ATI chipsets for AMD processors increased 81% vs Q4 2005.

sharikouisallwaysright said...

Yes, i was not clear, q4=05 is only around 60% of q4/06.
So from q4/05 its 80% up.
You are right!

Scientia from AMDZone said...

enumae

My point is the yields discussed were only in regards to CPU's, and that you bringing up chipsets has no relevance


No. We are discussing cost and the northbridge adds to cost. I assume you aren't suggesting that Intel can build a northbridge for free. The only honest way to compare is Intel cpu + northbridge vs. AMD cpu w/IMC. Every northbridge built by Intel takes up space on a wafer and costs real money to manufacture. You can't ignore it and have an honest comparison.

The data I gave is for a dual FSB northbridge. The effect for a single FSB northbridge is a bit less than halfway between the original numbers and the correction I gave. So, single socket cost is not much of a factor. However, the proposed quad FSB northbridge is nearly twice that size. I guarantee it will bounce Intel's costs significantly.

In other words, you can only ignore northbridge cost if you are only talking about single socket motherboards.

Scientia from AMDZone said...

pointer
Though, it doesn't change the fact that you are AMD fanboy claiming not to be one and not able to intelligently discuss when people are not agreeing with what you said. You would turn to name labeling, post deleting, making fun of what the poster said or claiming the poster has no necessary background, etc.


You are done. No more posting here.

enumae said...

Scientia
No. We are discussing cost and the northbridge adds to cost.

Actually, this is what was originally stated and was wht was being discussed, ashenman said...

Roborat, could you point out your source for the lower yield on barcelona than on penryn? Both are monolithic dies and thus share the same yield disadvantages. Also, Barcelona comes out about 6 months before the xeon penryns, and can, and probably will be, rehashed at 45nm a few months after penryn is released. As is, I see the immediate advantage being to AMD.

Again, ths dicsussion was about wafer to wafer, not platforms.

Here is what Dr. Yield, PhD, MBA had said...

Net result, at a 2X defect density, the smaller die size and multi-chip approach yields an extra 49% saleable CPUs (66). Assume a 2% higher yield loss at packaging than the monolithic approach, and you still get 62 more CPUs. At an ASP of $500 for high-end parts (both AMD/Intel), that translates to an exra $31,000 of revenue per wafer.

While it is understood that Intel needs a Northbridge and that it cost money, it was not a part of the discussion, the discussion was Penryn vs Barcelona wafers, and only those, you brought up Northbridges and then the discussion turned towards platform.

Fujiyama said...

I was wandering what is the advantage of monolithic chip.
- worse yields
- larger project
- worse power dissipation
- no flexibility in design

I always hoped that AMD or Intel is going to design modular chip - single core/cores with HT links, separate cache and it would be possible to assembly of buy different package.
So everyone can buy a dual-core chip and upgrade CPU buying another two cores. What is a problem to do that?

Unknown said...

The packaging adds resistance and increases power usage. By increasing the number of packages of cores (increasing the number of physical processors) you increase power usage significantly.

Also, monolithic dies have better power usage, because the die was designed for the specific power requirements of all four cores. Also, redundant elements in the cores can be eliminated, and thus consumer less power. Monolithic dies also pwerform better, now that we're looking at shared cache, regardless of vendor. With shared cache, a single app that is multi-threaded can use a single cache source, and is thus more efficient and faster.

The only downside to a monolithic core is yields and flexibility, and even then, dual dies have certain flexibility issues as well.

Ho Ho said...

fujiyama
"I always hoped that AMD or Intel is going to design modular chip - single core/cores with HT links, separate cache and it would be possible to assembly of buy different package."

There are 2P/4P workstation or server motherboards you can buy that are pretty much what you asked. I remember THG testign dualcore Opteron and singlecore Opteron on same motherboard. It worked but was a little unstable in some cases. I bet it is possible to get it working fine.

Roborat, Ph.D said...

I said: As for AMD, it’s obvious now that with their limited capacity, ramping Barcelona in 65nm will not be in their best interest profit wise.
-----------------------
Scientia said: I'm sorry but you don't seem to understand the technical aspects. The slower ramp is due to a newly designed transistor and new die mask, not due to economics. Do I have to remind you that Intel encountered the exact same thing with C2D?


Saying that capacity ramp is limited by "transistor design and new die mask" is just ridiculous and only shows you have abosolute zero knowledge of how a Fab performs.

The decision to ramp a new technology and by how much is guided by several factors (i.e., transition planning with mininal disruption to current production, tool set availability) but mostly it is due to "MARKET DEMAND". Intel's C2D's ramp was just like any other ramp and being market dominant, it is driven by customer demand. What good would it do Intel if they can supply say 1,000,000 C2D when there are only 500,000 C2D motherboards out there? (hint: inventory)
Poor yields and new mask sets have never been capacity ramp limiters and its a bit late at high volume manufacturing to be fixing critical issues and early design parameters like "mask sets".

Scientia from AMDZone said...

roborat

Intel already had 65nm production and processes in place when C2D was introduced. There was no delay caused by 65nm development. In spite of this, Intel was only at 30% C2D production six months later. The ramp speed was limited only by the new die.

Yet, you are trying to suggest that AMD will introduce a completely new transistor on 65nm (which is a very big change) and you feel that they should be able to ramp faster than Intel. AMD's ramp of K10 should be nearly as good Intel's ramp of C2D in spite of the difficulty of a new transistor. I'm baffled why you seem to think that this ramp speed is too slow.

BTW, your suggestion that C2D ramping was due to a lack of demand is so ridiculous I don't even know how to answer that.

Scientia from AMDZone said...

Fujiyama
monolithic chip.
- worse yields


If everything were equal this would be true.

- larger project

This isn't quite true. With shared bus the project size is smaller for MCM. However, for IMC the project size is larger for MCM.

- worse power dissipation

I don't where you got this one from but you have it backwards. The power disipation for an identical MCM will always be higher.

- no flexibility in design

I suppose this is theoretically true but Intel has shown no additional flexibility with MCM.

I always hoped that AMD or Intel is going to design modular chip - single core/cores with HT links, separate cache and it would be possible to assembly of buy different package.
So everyone can buy a dual-core chip and upgrade CPU buying another two cores. What is a problem to do that?


MSI did this with the MSI K8N Master2-FAR motherboard. Only one cpu was connected to memory. The second cpu only has an HT connection to the first cpu.

Scientia from AMDZone said...

enumae

It is pointless to claim a cost advantage for Intel based on yield and then to ignore the cost disadvantage for the northbridge. Basically you can say:

Single socket - MCM gives Intel lower cost than AMD.

Dual socket - MCM gives Intel only a slightly lower cost than AMD.

Quad socket - In spite of MCM the cost of the northbridge is prohibitive for almost all 3rd party companies. Intel's cost is higher than AMD's in spite of MCM.

Roborat, Ph.D said...

Intel already had 65nm production and processes in place when C2D was introduced. There was no delay caused by 65nm development. In spite of this, Intel was only at 30% C2D production six months later. The ramp speed was limited only by the new die.

Again, what do you mean by limited by the die? More than anything the product itself has nothing to do with how fast a semiconductor company ramps volume. Product complexity affects design validation and time to market. But volume ramp is a function of manufacturing capability to convert. Why do you keep repeating this misconception? Different resources: Die = R&D, ramp = manufacturing.

Yet, you are trying to suggest that AMD will introduce a completely new transistor on 65nm (which is a very big change) and you feel that they should be able to ramp faster than Intel. AMD's ramp of K10 should be nearly as good Intel's ramp of C2D in spite of the difficulty of a new transistor. I'm baffled why you seem to think that this ramp speed is too slow.
If you look at my original post, I said according to AMD they will not ramp K10 in 2007. My belief was this due to the fact that AMD is capacity limited and ramping Barcelona isn’t the best option for them. As far as I am aware AMD has no similar plans of ramping to 30% K10 6 months after launch.

BTW, your suggestion that C2D ramping was due to a lack of demand is so ridiculous I don't even know how to answer that.
Well that’s because maybe you think that Intel sell to end users. Maybe you don’t realize that Intel and AMD sell to system builders who have inventories. Maybe you don’t realize how long the supply chain is and what volume commitments each of those supply chains have and how much time it takes before a specific line of PCs can change direction.
Between you and Intel, I’m pretty confident that Intel knows how much share of the market was going to be Core2Duo early adopters. And if Intel ramps to 30% after 6 months, I’m willing to bet the market demand for C2D was close to that number.
But what’s truly is ridiculous is that people not realizing that if Intel and AMD switched products completely without product ramps, $Billions worth of inventories goes to waste.

Roborat, Ph.D said...

Scientia:
It is pointless to claim a cost advantage for Intel based on yield and then to ignore the cost disadvantage for the northbridge…


Unless AMD can charge extra for its IMC or if Intel gives away its chipsets for free then I think this argument is moot.

The MCM’s approach saves Intel money by avoiding the yield loss associated with 2X defect density and the ability to sell 2 dice in one package at slightly higher price that it were sold individually. Business wise I think it’s a smart approach and timing wise it was perfect (quad-core marketing). I’m sure this approach wasn’t as appealing to AMD with their strategy of grabbing as much market share as possible. I think we should leave it at that.

enumae said...

Scientia
...It is pointless to claim a cost advantage for Intel based on yield and then to ignore the cost disadvantage for the northbridge...

Hey, I am not trying to step on your toes or be disrespectful and I understand your point, but, like it or not my point is that the debate was about Intel Penryn and AMD Barcelona yields.

Thats all, not about platforms or other cost associated with choosing AMD or Intel that you brought up, just CPU wafers and the advantage that Intel's MCM approach has over AMD's Barcelona.

It is understood that AMD has a potential cost advantage for having the northbridge on the chip, and no one is ignoring or debating you about that.

Again, to be clear, you brought up a variable (chipsets) that was not relevant to the original discussion, yields per wafer and cost savings associated with them.

Scientia from AMDZone said...

roborat

I started counting how many incorrect statements you had made in these comments and then gave up after I got to 10.

MCM does indeed reduce Intel's costs by increasing yield. It does not have any effect on defect density (one your latest incorrect statements).

However, you incorrectly attributed AMD's lack of MCM to market strategy when it is in fact due to architecture. That you would try to attribute it to market strategy shows that you don't understand the difference between shared bus and IMC.

That you think depreciation is an expense shows that you don't understand accounting.

That you can't even get the name of the article you are commenting on correct (when it is printed at the top of the same page) shows that you don't read very carefully. I think this is shown again when the article that you linked to to support your statements actually disagreed with you. And, this is even further shown when you somehow misread Ruiz's statements to mean that almost no K10's would be produced in 2007. In fact, you were so confident in your misunderstanding that you claimed K10 would be less than 1% of AMD's server output in 2007.

Your statements about chip ramping are completely false. C2D was a smaller die size than the Prescott based dual cores that Intel was already making. If your theory about ramping was correct then Intel could have dropped all dual core P4's the next day and produced C2D's at lower cost. Yes, I know that you try to rationalize this by talking about inventory. However, even allowing for inventory Intel would have been able to be at 60-70% C2D by years end instead of 30%. The fact is that new dies cannot be ramped overnight as you seem to believe.

It is true that the quad core version of K10 would consume too much die space in 2007 to be a large fraction of production. However, this argument is not true of the dual core version which will be the primary version on the desktop. Your inventory argument is invalid for AMD since the newer chips still work on existing chipsets (unlike Intel's). The truth is that any new die has to be tweaked in production to get good yields and good clock speeds and this takes time.

As far as I can tell, the likelihood of any statment by you actually being true is less than 20%. I guess we can leave it at that.

Scientia from AMDZone said...

enumae

I agree that MCM increases Intel's yield and does give them an advantage on a die to die comparison. However, roborat's original statement was:

One can already deduce that the more expensive, lower yielding Barcelona will not be able to charge a high enough ASP competing with Penryn.

I don't believe this statement is correct. Do you agree or disagree with roborat here?

So it appears that AMD's undemoed 2007 savior will be too little too late.

This statement is ludicrous. I certainly hope you don't agree with this.

enumae said...

Scientia
"I don't believe this statement is correct. Do you agree or disagree with roborat here?

In my opinion the question can not be answered right now. We do not have any hard numbers and there are alot of other variables, but here is what I will speculate.

If AMD has good power and performance (achieving what has been claimed) then I will not agree with his statement.

As of now, from what has been said, and looking at what changes AMD has made, Barcelona should be a great processor with great power management.

The big question is how will Penryn compare with Barcelona performance wise?

If Penryn increases performance while reducing power, and is able to be competitive (within about 5-10%) with Barcelona then I would agree with his statement.

Alot of AMD's future relies on a good transition to 65nm and the performance, launch and acceptance of Barcelona.

In regards to the present... we will know alot more in just over a month.

Scientia
This statement is ludicrous. I certainly hope you don't agree with this.

I do not.

Roborat, Ph.D said...

MCM does indeed reduce Intel's costs by increasing yield. It does not have any effect on defect density (one your latest incorrect statements).

Who said anything about MCMs which is an assembly process having an effect on defect density which is a fab issue? please don't put words in my mouth.

That you think depreciation is an expense shows that you don't understand accounting.

From WIKIPEDIA: The use of depreciation affects the financial statements and in some countries the taxes of companies and individuals. The recording of depreciation will cause an expense to be recognized, thereby lowering stated profits on the income statement...

Enough said.

that you claimed K10 would be less than 1% of AMD's server output in 2007.
Unless you have actual AMD volume plans for K10 in 2007 you have no right to claim i am incorrect.

The truth is that any new die has to be tweaked in production to get good yields and good clock speeds and this takes time.

So? And what does that have to do with product ramp? You can have the best yields in the world but without the planning of purchasing the tools 6-9 months in advance and installing them will not give you a product ramp. On the other hand you can have the crappiest yields but still ramp the hell out of it if you have enough tools. Yields have nothing to do with a planned ramp. It may affect the actual output and cost, but not the RAMP.
Honestly, you really should stop arguing about this because you really sound like you have no idea what you're talking about here.

bk said...

roborat
On the other hand you can have the crappiest yields but still ramp the hell out of it if you have enough tools. Yields have nothing to do with a planned ramp. It may affect the actual output and cost, but not the RAMP.

So you don't think the cost to produce the product will effect the ramp. Lets say a company is getting 10% yield on a new process it is ramping. If it produces at this yield, it will loose money vs. producing on its current established process. Do you honestly believe a company will continue to ramp production at a yield that causes them to loose money? My bet would be that the company would slow/delay the ramp until the yields improve.

Roborat, Ph.D said...

bk said:
So you don't think the cost to produce the product will effect the ramp. Let’s say a company is getting 10% yield on a new process it is ramping. If it produces at this yield, it will loose money vs. producing on its current established process. Do you honestly believe a company will continue to ramp production at a yield that causes them to loose money? My bet would be that the company would slow/delay the ramp until the yields improve.

But that is what the contention is about. You never release a new process to production unless you achieve the yields that allow profitability. We’re talking about A YEAR and a HALF in advance of capacity planning here. Tools needs to be ordered and build and by average it takes 6-9 months before one can be running in your Fab.
Take for example R&D did screw up and the actual production yields turned out to be crap. And you now have tools in the Fab and some more tools coming for your planned ramp. The question now is do you slow down your RAMP or do you continue to RAMP while fixing your process? Trust me. Slowing down your ramp to leave your Fab and your depreciating tools underutilized while causing you to not meeting your customer commitments will only make things worse. I have never heard of a scenario where a Fab had to stop a planned ramp because of bad yields. Not in this day and age where design & development (D&D) and new product introduction methodologies are well established and $Millions are in stake. Scientia’s statement that slow ramps are due to D&D problems like “newly designed transistor and new die mask” is completely absurd.

Scientia from AMDZone said...

Roborat
Who said anything about MCMs which is an assembly process having an effect on defect density which is a fab issue?


Fair enough. I probably just misinterpreted your 2X defect density remark.

The recording of depreciation will cause an expense to be recognized, thereby lowering stated profits on the income statement...

You shouldn't rely on wiki. Depreciation is reported as a loss of assets which has to be accounted for on financial statements. However, depreciation is tax deductible and therefore reduces expenses rather than increasing them.

Unless you have actual AMD volume plans for K10 in 2007 you have no right to claim i am incorrect.

Roborat, you are without doubt 100% incorrect. There is no possible way that K10 will only account for less than 1% of server production in 2007. This would be less than 200K processors when K10 will be at least 7.5 million. You are wrong by a factor of at least 40.

So? And what does that have to do with product ramp? You can have the best yields in the world but without the planning of purchasing the tools 6-9 months in advance and installing them will not give you a product ramp.

Are you serious? You are trying to beef up your lame arguments by suggesting that I am talking about making chips without tooling? You are desperate. AMD will have 75% of the 65nm tooling in place in FAB 36 by mid year.

On the other hand you can have the crappiest yields but still ramp the hell out of it if you have enough tools.

Really? Perhaps you can name even one company that did this. Ramping a new die with poor yields would not be very cost effective unless this chip sold for a very high price.

Yields have nothing to do with a planned ramp. It may affect the actual output and cost, but not the RAMP.

Which clearly shows that you don't have a clue about chip production. The truth is that AMD could have released K8 in Q4 02. The reason they did not do this was that yields were poor at the time and the best clock speeds they could have achieved were only about 1.4Ghz. Since the die size was twice as big as the K7 die this would not have been cost effective. So, AMD delayed the release until Q2 03 and then ramped fairly slowly through Q1 04. Yields were the specific reason for this.

Honestly, you really should stop arguing about this because you really sound like you have no idea what you're talking about here.

Robo, I have less respect for your opinion with each new post you do.

Roborat, Ph.D said...

Scientia said:
Roborat, you are without doubt 100% incorrect. There is no possible way that K10 will only account for less than 1% of server production in 2007. This would be less than 200K processors when K10 will be at least 7.5 million. You are wrong by a factor of at least 40.

This is what Hector said about the K10 ramp during the CRN interview: “The biggest impact it will have is that we'll see a large number of customers and partners align themselves behind the technology. We expect that ramp to follow along the same lines as when Opteron began to get adopted . So I expect it to follow the same patterns. Over 2007, it will have a significant impact on what I call design wins”.
Now can please tell me how long before OPTERON was truly adopted by the industry? I’m counting years not months. 7.5 million in 2007? I don’t think so. You’re even contradicting Hector himself.

Are you serious? You are trying to beef up your lame arguments by suggesting that I am talking about making chips without tooling? You are desperate. AMD will have 75% of the 65nm tooling in place in FAB 36 by mid year.

You’re changing the argument.

Which clearly shows that you don't have a clue about chip production. The truth is that AMD could have released K8 in Q4 02. The reason they did not do this was that yields were poor at the time and the best clock speeds they could have achieved were only about 1.4Ghz. Since the die size was twice as big as the K7 die this would not have been cost effective. So, AMD delayed the release until Q2 03 and then ramped fairly slowly through Q1 04. Yields were the specific reason for this.
Thank you for giving an example of AMD fixing yield problems before the ramp. Anybody who’d actually worked in a Fab knows that the speed of which a Fab can ramp depends on tool availability and eng resources. Yield slows down design & development and product release but not the product ramp.

enumae said...

Scientia
...K10 will be at least 7.5 million.

Is this dual core or quad core?

Or is it both?

Christian H. said...

Roborat, you are without doubt 100% incorrect. There is no possible way that K10 will only account for less than 1% of server production in 2007. This would be less than 200K processors when K10 will be at least 7.5 million. You are wrong by a factor of at least 40.


I just couldn't take it anymore. There seems to be a lot of people here with intimate knowledge of manufacturing etc. but don't seem to have any common sense.

Hector's statments should be clear to unbiased eyes.

The reason Barcelona will be such a small part of 07 revenue is simple: there is no 65nm K8 Opteron(I am surprised they didn't shrink the HE or SE).This means that most server chips will still be dual Opteron.

Barcelona is not ALL of the chips, just the 2P and 4P server chips.

They still need room for the volume SKUs like Kuma/Rana and to some extent Agena.

Since 2P/4P is a drop in the bucket and AMD is after volume they will ramp the desktop chip first and Dell and Sun will get MOST of the Barcelona SKUs.

Budapest will also need to have Fab space as it will be the new wksta chip.

Because they only have Fab36 doing 65nm until Chartered comes onine around May, they have to reserve the most space for Brisbne and Lima.


When Fab38 gets 1/3 (AMD said ony 1/3 would go offne at a time), then Opteron will solely come from Fab 38 and all desktops will come from Fab 36.

I would guess that Chartered will take over Brisbane/Lima and Fab 36 will do the Barcelona derivatives.

Fab 36 is reportedly at 15K WSPM and 283m^2 is a lot bigger than 126mm^2.

Then AMD still has to get 45nm equipment installed somewhere. My guess is 1/3 of Fab38.

Since Brisbane is around and ramping 1/3 at 300mm allows the other 2/3 to be taken offline without a supply reduction.
This again implies that Kuma will be the main parts coming out of Fab 36 until Q4 as it should be half the size of Barcelona.

Once AMD reaces their volume targets the ramp to Barcelona will happen. By Q4 there will be a serious cancellation of 90nm Opterons.

Then Fab38 will have 65nm and 45nm qualified with 300mm wafers. Then the 65nm chips will be 75% of shipments.

I expect that FishKill should drop the tape out announcement of Shanghai in the next few months.

And as I expected Intel is not going to go volume with 3GHz Clovertown since I know it's at 140W maybe 150W.

Quad core will take awhile to break 3GHz. I really don't expect it until 45nm.

Unknown said...

Fair enough. I probably just misinterpreted your 2X defect density remark.

Actually, Roborat's statement is incorrect as well. Defect Density will not change if you are running a die 2x the size - and neither will yield for that matter. What will happen is defect density will stay constant but yeild may be much better than .5x for the 200mm^2 part. Just to throw out some real numbers - if I assume 300mm wafers and a constant (zero) defect density for a hypothetical 100mm^2 x2 part and a 200mm^2 part, I get a minimum of 1.03x (edge factor on the wafer) for the 100mm^2 x2 die. At a 'normal' defect density, I get 1.39x advantage.

As for depreciation - it is an expense, but it is not a 'cash expense' as it is an artifact of an accrual based accounting system. However, it should be included when comparing 2 products as it does represents the use of the capital equipment. The only time it should be ignored when comparing 2 products is if you never plan to purchase new equipment - otherwise, new purchase of equipment comes into play which is a cash charge. Given AMD and Intel are forced to have high CAPEX every year, it is common sense to include depreciation expense when comparing the cost advantage of their products.

Ho Ho said...

thekalif
"And as I expected Intel is not going to go volume with 3GHz Clovertown since I know it's at 140W maybe 150W.


Quad core will take awhile to break 3GHz. I really don't expect it until 45nm."


linky

"Still, the company has a plenty of headroom with its 65nm chips and has plans to introduce 3.0GHz Intel Xeon “Clovertown” chips going forward"

"there will be an improved version of the Intel 5000-series chipset aimed at HPC/WS market segments that features 1600MHz processor system bus (PSB) coming in the second half of the year to support higher-performance Xeon “Clovertown” chips with operation at 3GHz."

Christian H. said...

"Still, the company has a plenty of headroom with its 65nm chips and has plans to introduce 3.0GHz Intel Xeon “Clovertown” chips going forward"

"there will be an improved version of the Intel 5000-series chipset aimed at HPC/WS market segments that features 1600MHz processor system bus (PSB) coming in the second half of the year to support higher-performance Xeon “Clovertown” chips with operation at 3GHz."





I know about that but as I said VOLUME shipments won't happen as it will be a 140W chip, maybe more unless they tweak the process severely. I mean lok at how Clovertown ramps with speed.

The slowest models are 68W, then 80W, then 120W. The next level will be at least 140W which makes it look less desirable.

Ho Ho said...

thekalif
"The slowest models are 68W, then 80W, then 120W. The next level will be at least 140W which makes it look less desirable."

First, where did you get that 68W number? Isn't that from AMD?

L5310 will be 50W at 1.6GHz.

Intel has quadcores at 80W TDP from 1.6GHz (5310) to 2.4GHz (5345). For 2.66GHz they had to increase the TDP number. Who wants to bet that 3GHz one won't be >120W?

Also Intel has demonstrated its 65nm quadcores running at >3.5GHz running Alan Wake. Would that put those OC'd things at >>200W?

Woof Woof said...

Intel's TDP according to their spec sheets != Max Power

Traditionally, Intel CPU Max Power is abt 11% higher than their TDP rating (see prev spec sheets pre Woodcrest).

Also, the 5000 chipset memory controller takes up abt 29W TDP if you use two banks of memory. 30+W if you use four.

So 80W Intel part (in AMD terms)
= 80*1.11 + 29W
= 117W

enumae said...

Scientia I hope I don't offend you by doing this.

This is off topic...

I feel that the people who view this blog probably spend quite a bit of time on line, and as such I am hoping I may get a few minutes of your time.

I am working on a new website and I am hoping some of you could take a look and possibly tell me what you think of the layout and color scheme.

The site is in the very, very, very early stages, but the contact button works, so thanks in advance to anyone who takes the time to comment.

Here is the link

abinstein said...

roborat: "Take for example R&D did screw up and the actual production yields turned out to be crap. And you now have tools in the Fab and some more tools coming for your planned ramp. The question now is do you slow down your RAMP or do you continue to RAMP while fixing your process? Trust me. Slowing down your ramp to leave your Fab and your depreciating tools underutilized while causing you to not meeting your customer commitments will only make things worse."

Roborat, what you said do not support your argument. Production ramp depends on yield because fabs will be making "dumb shrinks" before the new design. So if Barcelona does have some design issue negatively affecting its yield, AMD would just slow down or halt the ramp and keep making more Brisbane.

It seems Barcelona's design has been stable enough that AMD's confident of an end-of Q2 release. As we know, the first batch will have clock speed in the range of 2.1-2.3Ghz, which is not particularly good, and it makes sense to ramp slowly at these speeds to allow both the process and the design mature.

abinstein said...

thekhalif: "The reason Barcelona will be such a small part of 07 revenue is simple: there is no 65nm K8 Opteron (I am surprised they didn't shrink the HE or SE).This means that most server chips will still be dual Opteron."

AMD can just make K10-derivative Opterons at 65nm, and leave K8 at Santa Rosa.

"Barcelona is not ALL of the chips, just the 2P and 4P server chips."

Quad-core Opteron 1000 series will be in production by the end of 2007.

"They still need room for the volume SKUs like Kuma/Rana and to some extent Agena.

Since 2P/4P is a drop in the bucket and AMD is after volume they will ramp the desktop chip first and Dell and Sun will get MOST of the Barcelona SKUs."


Desktop sales are important to increase revenue, but have much lower profit margin. AMD's "Barcelona strategy" for 2007 is clearly to lineup the server makers behind Opteron, HyperTransport, and Direct Connect Architecture. It is important for AMD to do so first before the release of CSI from Intel.

Scientia from AMDZone said...

AMD's initial release of K10 will be quad core opteron. The initial clocks will only be 2.3Ghz. The 2.3Ghz K10 Integer speed should be faster than a 2.6Ghz K8 however, not quite as fast as the 2.8Ghz. It will be much faster in terms of SSE and obviously with twice the cores it would be considerably faster for multi-threaded code.

Sometime in Q3 AMD is supposed to bump this to 2.5Ghz which should easily beat the 2.8Ghz Opteron in every way but would only equal the Integer speed of 3.0Ghz K8's. It's been suggested that the dual core versions of K10 will hit 2.9Ghz but it would only take 2.7Ghz to jump ahead of 3.0Ghz K8.

It is obvious that AMD will want a fair number of quad core K10's to counter Clovertown and it will also want to deliver the very small volume but highly visible QFX upgrade. Then AMD moves on to the dual core versions of K10 Opteron. This will leave FAB 30 building Semprons and single core Athlon 64's.

FAB 36's 65nm capacity will be absorbed by mid range Brisbane X2's although Chartered will help with this; and with 65nm mobile which is desperately needed. Chartered may build some of these too but it is not likely at all that Chartered will build K10 in 2007.

I expect AMD to cover the server segment fairly well with K10; think in terms of 25-50% coverage rather than robo's ridiculous 1% estimate. I'm also certain that AMD will cover FX since it is very small volume, however I doubt K10 will be available for much on the desktop except at the high end. AMD won't start catching up with K10 production until Q2 08. By catching up I'm referring to K10's being available in the desktop midrange.

It wouldn't make sense for 45nm production to start on FAB 38 since the tooling wouldn't even be qualified on 65nm at that point. 45nm production will start on FAB 36 with FAB 38 qualifying its tooling in early 2008 to replace 90nm production with Brisbane versions of single core Athlon 64 and Sempron. This would then allow the last of the 90nm tooling to be removed. FAB 38 should be ramping capacity until end of 2009.

Intel shouldn't deliver CSI with IMC on the new socket until 2009. AMD will have time to release DC 2.0 before then. It also appears that CSI will be hampered by having to have a switching hub to route signals. Further, HT 3.0 includes functionality that CSI lacks such as a split mode, a power saving mode, and a long distance mode as well as being hot pluggable. CSI will probably be more competitive with a second version in 2010.

Roborat, Ph.D said...

Scientia:... I expect AMD to cover the server segment fairly well with K10; think in terms of 25-50% coverage rather than robo's ridiculous 1% estimate.

Let me just clarify. I said less than 1% of AMD's total server output in 2007. That's volume and that's for the entire year 2007 . You can look back up the posts to confirm what I said.

You may want to clarify your "25-50% coverage" before you start comparing it to my statement. Bearing in mind that AMD has been shipping non-Barcelona Opterons for two months now.

Common sense tells me you speak of production % between K8 and K10. I still think its a bit high considering Hector's statement that they will not ramp. To me that's manufacturing's code word for single line production, very limited release. Take that in perspective, Intel's was committed to ramp Core2 simultaneously across all segments and yet 6 months later was only 25%-30% share.
I honestly think you're over estimating again Scientia.

enumae said...

Scientia
It would seem my statement about northbridges was correct, its the southbridge that almost all Nvidia chipsets lack.

I know this isn't a server motherboard, but AMD's new chipset uses a Southbridge on the 690G chipset.

When I found that, I then found that the EPoX EP-AD580XR does as well, there are probably others.

Christian H. said...

Intel has quadcores at 80W TDP from 1.6GHz (5310) to 2.4GHz (5345). For 2.66GHz they had to increase the TDP number. Who wants to bet that 3GHz one won't be >120W?

Also Intel has demonstrated its 65nm quadcores running at >3.5GHz running Alan Wake. Would that put those OC'd things at >>200W?



Hey, I was nice enough to give them 12W. Running a game at a speed rating says nothing about power usage. Physics says that as you increase speed you increase heat.
That in my opinion is why there isn't a 3.2GHz Core 2. It'll start to go the way of NetBurst.

BTW, Sci,

There is a link to some seti numbers using what I assume is a Barcelona 2P system.

It basically destroys a 2.66GHz Clovertown.

WOW!!

«Oldest ‹Older   1 – 200 of 224   Newer› Newest»