Tuesday, May 08, 2007

AMD Is Doomed!

I constantly get accused of being slanted/biased/pro/fan in favor of AMD. So, I'm going to try to see the world through the eyes of an unabashed Intel enthusiast. This is not an easy point of view to maintain though because Common Sense keeps getting in the way.

It's so wonderful to be able to rely on a manufacturer with so much volume capacity and a steady production supply.

CS: AMD is now making processors in three different FABs (FAB 30, FAB 36, and Chartered) plus TSMC and UMC for Chipsets.

Yes, but it is so liberating to know that I can overclock the stuffing out of my wonderful C2D chips.

CS: Yes, this is great for the 1/10th of 1% of people who do this but it doesn't effect the vast majority of computer purchasers; factory stock clock speed remains the most important. Also, it has been suggested by die-hard overclocking Intel enthusiast Ed Stroligo that Penryn's 1600MHz FSB speed will put the brakes on overclocking.

Sure, but AMD is drowning in debt; they were stupid to buy ATI.

CS: Buying ATI was the only way to compete with Intel's in-house chipset advantage and AMD's debt was recently refinanced. Of course the debt isn't good and AMD needs to start making money but it isn't going to drive the company out of business in the next year.

Intel has been so open about C2D and their 45nm process; AMD has said almost nothing so they must be hiding something.

CS: AMD was just as secretive about K8. What reason does AMD have to sabotage current sales by promoting a chip that isn't available yet? Also, AMD doesn't have anywhere near Intel's advertising budget. Making a big splash at launch couldn't hurt.

But Intel has already demonstrated 45nm; AMD must be having a problem.

CS: AMD produced its own 45nm SRAM test wafers just 3 months after Intel. If AMD is going to release 45nm mid 2008 then these chips would tape out about mid 2007. How can AMD demonstrate a chip before it is taped out?

Intel is a full process generation ahead of AMD.

CS: Isn't a process generation 2 years? If Intel leads AMD by six months wouldn't this be ¼ of process generation? Also, AMD has indicated that it is going to move to 32nm in just 18 months while Intel is still looking at 24 months. This should make AMD and Intel about equal at 32nm.

AMD is going to be delayed because they are using immersion technology. Intel was smart to stick to the old dry process.

CS: IBM and AMD have been using a pre-production prototype immersion scanner for quite some time and are reporting the same defect rate as dry; there is no reason to assume a delay.

You can't believe AMD. Intel is much more reliable.

CS: You are suggesting that the company that lied about developing 64 bit technology for X86 (when it was already on the Prescott die) is more honest?

Intel is a far better engineering company; they will win just by being better.

CS: If Intel is so much better at engineering then why was Williamette uncompetitive with K7? Why did it take the Northwood revision for P4 to take the lead? And, why did they cancel Tejas, the original Nehalem, and Whitefield? Also, if Intel can win just by being better why did they bully motherboard manufacturers to delay supporting both K7 and K8?

You can't prove Intel did that.

CS: The K8 launch was not attended by a single motherboard manufacturer; why do you think that was?

Maybe they just knew how poor K8 was.

CS: You mean the architecture that would go on to dominate Intel's products for two years?

Well, if AMD is so great why don't they have quad core?

CS: This isn't really that much different than when AMD released the excellent X2's and all Intel could offer in response were the Smithfields which ran slow and hot. It took Intel about the same amount of time to reply with C2D as it will take AMD to reply with K10.

Yes, but Intel's MCM approach is brilliant; it is much more flexible and economical than AMD's plan to put four cores on one die.

CS: True but MCM also has poor scaling. With four cores, Kentsfield is really only getting the equivalent of three. This is one of the reasons why AMD is confident that they can match Intel in quad core.

AMD will never be able to keep up. Intel has tons of unused clock ceiling.

CS: Then why is Intel bumping the thermal limits on Kentsfield at 3.0GHz with a stock HSF? And, why is Intel waiting until 45nm to release anything faster?

That's . . . just a coincidence. Intel has always had better process technology.

CS: You mean like back in 2000 when their PIII couldn't keep up with K7?

That was a fluke; just look at history!

CS: Okay, Intel's first boom period started when they introduced the ATX motherboard standard. This ended when AMD released K7 four years later. Intel's second boom period began at the beginning of 2002 with the release of Northwood on 130nm and ended at the beginning of 2004 when Prescott had to be released as Celeron D. AMD's first boom period lasted just one year in 2001 when PIII was unable to clock and Williamette wasn't quite competitive with K7. However, AMD's second boom period started in 2004 and ended when Intel released C2D in 2006. So, Intel's booms have a downward trend from 4 years to 2 years while AMD's have an upward trend from 1 year to 2 years. It seems likely now that neither company will be able to maintain a lead for more than 1 year in the future. This would suggest that Intel's current boom will end with the introduction of Barcelona but will probably return with the release of Nehalem.

Hah! Intel will hold onto the lead with faster clocks; AMD can't keep up with 2.5GHz.

CS: The Inquirer suggests that AMD will release quad core at 2.9GHz.

That's just FUD.

CS: Well, 2.5GHz would have been fine when Intel's fastest quad core speed was 2.66GHz. Isn't it likely that Intel's recent boost of quad core to 3.0GHz also encouraged AMD to match with 2.9GHz?

Even if it's true, it won't matter; Intel will bury K10 with Penryn.

CS: Aside from SSE4, isn't Penryn just a Merom with more cache and a faster FSB?

Maybe, but Nehalem is a whole new architecture.

CS: I've already said that Intel may regain the lead with Nehalem but isn't it true that this isn't quite a new architecture in the old sense?

Of course it is; what do you mean?

CS: Well, it took four years before to design a new architecture. Intel is moving to Nehalem just 2 ½ years after C2D so it doesn't seem likely that it is a whole new core but more of an upgrade. Also, won't Intel encounter the same upgrade hurdle with Nehalem with a new socket that AMD is running into now with socket 939 versus AM2? And, with it's modular core design AMD should be able to deliver its own upgrade to K11 in 2009.

Oh, come on! Everyone knows that people prefer C2D. Intel's ASP is almost double that of AMD's.

CS: Yes, but if you look at HP's website there doesn't seem to be any real difference in the prices for actual systems. I remember back in 2002 and 2003 when Intel systems commanded a price 10-20% higher than AMD systems. And, HP offered more Intel systems than AMD. Today, the price looks about the same on both desktop and mobile, and the number of systems offered is identical on both desktop and mobile. Most likely the sharp drop in AMD's ASP was due to competition from discounted P4's and AMD's desire to hold onto its volume share.

Yeah, like that worked. AMD lost tons of marketshare. They lost everything they gained in 2006.

CS: Well, AMD took a big hit in revenue, no doubt. But, I haven't yet seen volume numbers. The only comment I've seen suggests that AMD's volume drop was only about 1/3rd of the revenue drop.

Intel will bury AMD with low prices.

CS: AMD's new DTX motherboard standard seems to be pretty popular and it was designed to reduce costs on desktop systems.

AMD needs more layers to make its chips; Intel's chips will still be cheaper.

CS: But perhaps not at 45nm. Doesn't Intel's use of the old dry process require multiple passes that make up for AMD's extra layers?

AMD is losing money now and it will just get worse as Intel ramps C2D.

CS: There is no doubt that AMD was at a big disadvantage with half of its production still at 90nm on a 200mm FAB but now as AMD ramps both 65nm and 300mm production its costs should drop. Also, as Intel phases out P4 how can it maintain lower prices without dropping its ASP?

Intel has quad core and 45nm production is a lot cheaper.

CS: Yes, quad core was an advantage but that advantage is lost once Barcelona is released. Also, the last estimate I saw was that 45nm production would account for less than 1% of Intel's volume in 2007; how much could that help with costs?

Yes, but Intel has Apple and their marketshare isn't even included in the PC numbers.

CS: Apples' desktop sales have been stagnant for the past three years. And, they seem to be charging a premium price for no additional features. For example, for the price of a MacBook with 15” display you can get an HP notebook with a 17” display, twice the harddrive capacity and still save $500. You could easily save $1,000 dollars over the price of MacBook with a 17” display. At these prices I would seriously doubt that very many parents are buying these for their kids. This seems more like the market for middle aged professionals who have been using Macs for awhile and don't want to switch. However, I'm positive that there is going to be increased pricing pressure on desktop systems and even on notebooks now that Intel doesn't have the market cornered with Centrino. I just don't see the average buyer continuing to shell out a premium price for Apple computer systems over the next two years. My guess is that Apple has peaked as a computer company and is now shifting to consumer electronics.

AMD is just a one trick pony!

CS: Intel had mobile 486 chips. AMD didn't get into mobile until K6 but caught up to Intel completely with Turion. Intel had the powerful Pentium-Pro server chip. AMD didn't get started in servers until Athlon MP but had caught up in servers by Opteron. On the desktop, AMD was way behind Pentium with K5, was briefly faster than PII with K6, outpaced both PIII and Williamette for a year with K7, led Prescott P4 for 2 full years, and now seems poised to regain the lead with K10. This seems like more than one trick.

AMD was so arrogant; they just sat on their hands for four years.

CS: Northwood P4 was introduced in 2002; that's four years to C2D's launch in 2006. AMD released K8 in 2003 and that is also four years to K10 in 2007. Was Intel just sitting on its hands for four years? It seems clear that AMD made a mistake or two during those four years and lost about a year while it shifted direction to K10. Intel made mistake after mistake with Tejas and then Whitefield but got lucky because the Israeli team had an architecture that could be beefed up into C2D. The problem is though that while Intel used to run with the completely separate Pentium M architecture and the substantially different Xeon architecture they have now cut back to just a single architecture. Woodcrest today is almost identical to Conroe just as Opteron is to Athlon 64. Intel has lost its former advantage of having multiple research projects to reduce risk. Today, in this regard, it is equal with AMD.

But, everyone knows that Intel is the best.

CS: If Intel is the best then HPC is very curious. Intel was dominant in HPC with its 32 bit Xeon servers. However, it has lost ground with 64 bits until, today, Intel only has a fraction of AMD's presence in the most power supercomputers. One has to wonder why there haven't been a flood of Woodcrest based supercomputers announced.

But, but, Intel has to win because . . . they are Intel!

CS: It seems that may no longer be enough.

13 comments:

Christian H. said...

Wow, that title surprised me at first. Though we don't always come to the same conclusions, we do agee that AMD is a great company with even more potential than before.

I like that CS: thing. It reminds me of Tom's Forum. They say somethng stupid I mention the opposite.

You're right that AMD will start to increase their revenue. They are reportng lots of orders that will roll ut in Q2 based on DTX and the Live Home Cinema for Turion.

I recently got an HP Home Ofiice catalog and it had ZERO Intel machines.

And now OCZ has eleased two new RAM SKUs SPECIFICALY FOR AMD. I would bet that OCZ has seen K10.

It is also worth noting that B0 is prety good as the final revision. I would say that means that designing K10 for 65nm and 4 stressors is paying off.

I long told people that Brisbane isn't the best example of AMDs 65nm process since it was a shrink from basically 130nm (though X2 came with 90nm?)

I also told thm that AMD woul dbe able to raise money very easily but talk became "buyout by PE."

Sure enough they got $2B($1.5B after Morgan Stanley). I'm glad they waited to release real numbers for Barcelona.

I honestly see bettet than 100% scaling as shown in the SPEC benches.

I mean just going quad (without bottlenecking) should get them 80% then when you add the awesome improvements, I would have to say the execs hae been VERY conservative.

K10 is a BEAST.

Scientia from AMDZone said...

theKhalif

"I recently got an HP Home Ofiice catalog and it had ZERO Intel machines."

Yes, I got the same catalog. I assume it is a targeted mailing because my last HP purchase was AMD. I think they have the same thing with Intel.

However, the point I was making is that Intel used to have a premium price at HP and this is now gone in spite of C2D.

Polonium210 said...

Scientia, you have once again repeated the error that AMD produced
test SRAMs 3 months after Intel when in fact AMD produced these at the SAME time as Intel, namely January 2006!

Christian H. said...

Yes, I got the same catalog. I assume it is a targeted mailing because my last HP purchase was AMD. I think they have the same thing with Intel.

However, the point I was making is that Intel used to have a premium price at HP and this is now gone in spite of C2D.



That maybe it since I did buy TWO AMD systems mobile and desktop. I just thought that interesting.

It obviously corresponds with what you've seen in terms of pricing - there is a parity in models and AMD seems a little ahead on price.

A little sarcastic humor goes a long way, doesn't it?

Scientia from AMDZone said...

heat

Actually I never said anything about AMD's being a smarter choice. As far as I can tell you can get a good system from HP with either processor for about the same price. Now, if you honestly doubt this then you could check the HP website yourself or you could check the review at Sharkys. There is currently no price advantage for an Intel system although for a higher price you can certainly get a more powerful dual core or quad core system.

Scientia from AMDZone said...

Actually, I didn't make up the Intel enthusiast arguments; I'm just repeating back what other people have said.

Of course, after doing this article I've been told that I completely overlooked the argument that because Intel is ahead now that they will always be ahead. The sad part is that the people who said this were serious.

netrama said...

But, I haven't yet seen volume numbers. The only comment I've seen suggests that AMD's volume drop was only about 1/3rd of the revenue drop
I think AMD might have increased its numbers in mobile space. The decreased revenue figure is to be blamed due to Dell and a weak dollar. Meanwhile for Intel I suspect it was able to charge a decent ASP from APPLE

InTheKnow said...

Scientia, I must say I'm rather disapointed in the tone of this piece. To me it screams that you either agree with me or you are an idiot. It's not the message I would want to send on a blog that invites open discussion. Hopefully, I'm reading more into this than you intended.

And for the record I don't think AMD is doomed, but I wouldn't put any money into their stock for a while. I think they have a rough year ahead of them.

Intel has been so open about C2D and their 45nm process; AMD has said almost nothing so they must be hiding something.

CS: AMD was just as secretive about K8. What reason does AMD have to sabotage current sales by promoting a chip that isn't available yet? Also, AMD doesn't have anywhere near Intel's advertising budget. Making a big splash at launch couldn't hurt.

This all may well be true, but it is hard to deny that all the pre-release data on Conroe hurt Intel when they were down.

Intel is a full process generation ahead of AMD.

CS: Isn't a process generation 2 years? If Intel leads AMD by six months wouldn't this be ¼ of process generation? Also, AMD has indicated that it is going to move to 45nm in just 18 months while Intel is still looking at 24 months. This should make AMD and Intel about equal at 45nm.

I think 6 months is rather generous. If you choose crossover as your reference point (no accusation of paper launches that way) then Intel hit crossover last June from what I could find on the web. The only thing I could find on AMD was that they expected to hit crossover sometime in Q1. I have not been able to find anything indicating they have actually hit crossover. If you know differently please fill me in. So by my metric, this would put them at 11 months and counting.

Now Intel claims they will hit crossover on 45nm in the middle of '08. When does AMD expect to achieve that milestone? Again, I haven't been able to find anything. If your claim of closing the gap is correct, they should be making similar claims.

You can't believe AMD. Intel is much more reliable.

CS: You are suggesting that the company that lied about developing 64 bit technology for X86 (when it was already on the Prescott die) is more honest?

Unlike the company which has consistently revised it's expectatations downward after every report that everything was going okay?

Frankly, I think it's foolish to think either company is completely trustworthy. It seems to me they both seem to employ plenty of spin doctors.

Maybe, but Nehalem is a whole new architecture.

CS: I've already said that Intel may regain the lead with Nehalem but isn't it true that this isn't quite a new architecture in the old sense?

Of course it is; what do you mean?

CS: Well, it took four years before to design a new architecture. Intel is moving to Nehalem just 2 ½ years after C2D so it doesn't seem likely that it is a whole new core but more of an upgrade. Also, won't Intel encounter the same upgrade hurdle with Nehalem with a new socket that AMD is running into now with socket 939 versus AM2? And, with it's modular core design AMD should be able to deliver its own upgrade to K11 in 2009.

If Intel is to be believed, they have been working on this for 4 years. They have 2 design teams on a 2 year offset. That means each team puts out a new architecture every 4 years, just like you are saying.

Yeah, like that worked. AMD lost tons of marketshare. They lost everything they gained in 2006.

CS: Well, AMD took a big hit in revenue, no doubt. But, I haven't yet seen volume numbers. The only comment I've seen suggests that AMD's volume drop was only about 1/3rd of the revenue drop.

But isn't a 2% volume share loss a big hit at a time when AMD has said they are doing all they can to hold volume share? It certainly isn't a move in the direction the Hector is saying he want's to go. No doom and gloom here, but there is no way to try and portray this as a good thing for AMD.

I believe that I have raised cogent and rational questions regarding several of your points. Does the fact that I question your viewpoint make me an idiot? The tone of this piece would lead me to believe I must be a low grade moron. But I remain an unconvinced low grade moron.

Scientia from AMDZone said...

intheknow

No, I wrote this one as mostly humor because I keep seeing the same silly arguments repeated by Intel enthusiasts.

That was a typo on the process technology. It should say 6 months behind on 45nm and roughly equal at 32nm. I corrected it in the article.

Yes, there is no doubt that a 2% loss in volume is bad for AMD since this puts them back to Q3 06 levels. But I keep seeing the revenue numbers reported as volume with people stating confidently that AMD lost 6% volume to Intel (which would be much worse).

I am always willing to discuss points about the markets and technology. I'll give you some examples:

Rationally, AMD's Q1 loss was bad and I would hope that they can cut this loss in half in Q2 and reduce this to zero loss in Q3 with profit back in Q4. I don't think this view is unreasonable yet I've seen it stated with confidence that AMD would continue to lose $500 Million every quarter in 2007 or even that AMD's losses would excelerate as Intel ramped C2D and AMD would lose $4 Billion total during 2007.

I've seen people dismiss K8 as a fluke when clearly AMD gained ground with both K6 and K7. I've likewise seen people dismiss AMD's foray into servers with Athlon MP and then their larger success with Opteron. And, similar dismissals of AMD's mobile efforts.

I don't think of anyone as an idiot who brings up a rational point. For example, I really don't know if K10 will perform and obviously AMD's debt is serious if they don't get back to making money. But why do people automatically assume that K10 won't perform while assuming that Penryn is a killer design? And, why would anyone assume that AMD would go bankrupt rather than admitting that they have a good shot of getting back on their feet based on their chipsets, DTX standard, and K10? I assume that costs will continue to drop as they ramp both 300mm and 65nm.

Finally, I don't understand the idea that when obvious AMD fans are claiming that K10 will leave Clovertown for dead that a lack of complete pessimism towards AMD somehow puts me in the same category.

Incidently, I thought about adding an AMD section but it seemed too cumbersome to do a three way conversation. This would have included statements like "Intel must be falling apart because they just closed a FAB and fired 1,100 people." Obviously this is silly.

Aguia said...

Intel is a full process generation ahead of AMD.

Isn’t everyone too?
Who besides Intel is already at 65nm and is already planning 45nm production in few months?

Besides didn’t AMD have to transact from their 130nm to (IBM) 130nm SOI? Obviously delaying 90nm transaction?

And isn’t Intel the largest manufacture in the world and with more factories? Shouldn’t they be the first and the number one?
I'm not expecting that the smaller be first, or there was something wrong at Intel.

Aguia said...

I know this is a little off topic and is already old, but with some google search I got this article:

Intel is TERRIFIED of AMD

I don’t know if this story is true since I'm still a little young but it was interesting to read, I don’t know if someone know anything about this.

Scientia from AMDZone said...

aguia

I read the article and it is a bit exagerated. Intel was not in the red when they got the IBM contract. The 8080 had two problems which is it required three different voltage levels and an external clock circuit. The Z80 had internal clocking and only needed one voltage level. The 8080's biggest problem was that it couldn't clock very high so performance was limited.

The 8080 was cheaper than Z80 (not the other way around) and was used for the next decade as an embedded controller. The 8085 was the more finished version of the 8080. It was being used but not as much as Z80 because Z80 was cheaper.

However, the author very strangely overlooks the 6502 which was the true most popular processor at the time. It cost less than Z80 but was just as powerful. The 6502 was used by Apple, Commodore, and Atari. It was used in the Atari home video game and the Mattel Intellivision game. 6502 was also used in perhipherals like printers and harddrives. And, later when it had been completely outclassed by the Motorola 68000 and Intel 80286 processors it was still used as a keyboard controller.

QDOS was not a ripoff of CP/M. QDOS used I/O abstraction to avoid having to write hardware level drivers for every peripheral as CP/M required.

Finally, the author leaves out the most important point of all. IBM did not want to buy processors from a single source. Dual sourcing was very common at that time since government contracts required it. As a result of this AMD began second sourcing the Intel chips. This is what put AMD on the path to where it is today.

The part that the author leaves out however is that Digital Research went on the develop their own version of MSDOS called DRDOS. However, MS put discrimating code in Windows that caused it to issue a bogus error message if it detected DRDOS. Also, DR created a superior graphic interface called GEM that was both smaller and faster than Windows. Unfortunately, GEM did not catch on.

The author does not mention that when MS first demoed Windows it was nothing but a graphic shell with no actual OS functions. Nor does he mention that the head of the Windows team who was from Xerox PARC quit in the middle of the project and that MS then put its lead MacOS programmer in charge. He was appalled at the poor design compared to MacOS but it was too late to change it. So, he had to just try to patch everything and make it work. This led to a hodgepodge of Xerox PARC mainframe and Mac GUI concepts. This however was thrown out when MS developed Windows NT whose lead developer wrote VMS. If you add one letter to VMS you'll see where W NT came from.

The significance of Apple to MS is that until Apple had MS develop applications for the Mac, MS was an OS only company.

Aguia said...

Thanks Scientia,
for the reply and explanation.