2007: The Second Half
Amid all the rumblings and rumors there signs of fundamental differences between this year and last. In almost every aspect of processors AMD and Intel have swapped places. This has left a virtual vacuum of analogy for AMD and Intel supporters alike since both are reluctant to compare their favorite to the competition. The situation today is not exactly the same but some comparisons do provide a view of where things are likely to go.
We can add up the various ways that Intel and AMD have swapped places and there are quite a few. In early 2006, AMD's K8 was the undisputed leader ahead of Intel's Presler and Yonah offerings. Today, C2D is the undisputed leader ahead of AMD's Opteron and Athlon 64 offerings. In late 2006, Intel introduced quad core which AMD has taken nearly a year to match. Today, AMD is ready to offer native quad core which it will take Intel about a year to match. In early 2006, Intel previewed the native dual core 2.93Ghz Conroe which looked great and then it was a matter of waiting for Intel to actually get them out the door in volume. Today, AMD has previewed native quad core 3.0Ghz K10 which looks great and once again it is a matter of waiting for AMD to get them out the door in volume. In 2006, Intel was recovering from revenue shocks caused by AMD's K8. Today AMD is recovering from revenue shocks caused by Intel's C2D. In 2006, Intel introduced a new architecture that was far ahead of its previous generation offerings while AMD was only able to offer secondary upgrades such as small clock increases, virtualization, and faster memory speeds. Today, AMD is offering a new architecture that is far ahead of its previous generation offerings while Intel is only able to offer secondary upgrades such as small clock increases, SSE4, and faster bus speeds.
It really is remarkable how similar each company's situation is to its competitor's last year. This is most fundamentally true on the desktop. I suppose Intel supporters would point out that AMD is not likely to take the top performance spot in Q4 when Phenom is launched as Intel did when Conroe was launched. That is true. However, I suppose AMD supporters could point out that AMD was never in the heat and power crunch that Prescott was. Mobile is the most fundamentally different. Intel took mobile by storm when it launched the Centrino platform and since that time AMD has only been slowly chipping away with Turion. Merom was nearly the opposite of Conroe. While Conroe added tremendous value to the desktop as it replaced the sagging P4 line, Merom on the other hand actually had worse power draw than Yonah. Intel finds that having conquered the battery life and wireless LAN issue years ago that it has no place left to take mobile to gain an advantage. Turion has only had a small effect on Intel's mobile share but AMD should be fully competitive in 2008 with Griffin and Puma. It also looks like most of Intel's tweaks with Penryn are to try to stave off the coming attack from K10 Opteron. Intel is putting up a good fight with lower power draw, more cache, and faster FSB but it won't be enough. The fact is that when you've taken back as much server share as Intel has the only place left to go is down. SSE4 could be a big boost in HPC however Intel has already made most of its gains in the HPC low range with Woodcrest so Penryn would most likely be an upgrade to existing systems. SSE4 could be a boost in the top range but currently Intel has little presence there.
With Intel certain to have small losses in server share and no real change from the previous situation in mobile that leaves the desktop as the main battleground. AMD's average volume share in 2005 was 18% up noticeably from about 16.5% average for the previous several years. AMD's average volume share in 2006 was 23% and even though Intel has been fighting hard it remains at 23% in Q2 07. The actual price cuts have been a lot less than most people imagine. Intel's overall ASP is only down 16% from the nearly steady value of about $99 that it had been for three quarters. AMD's drop is similarly down 17% from the previous three quarter average of $60. AMD has had a desktop ASP drop of 42% since Q1 06 while Intel has dropped 38% in the same time. In Q2 07 AMD's desktop ASP was $49 versus Intel's $83. AMD's desktop ASP is substantially lower than Intel's but if it remains steady AMD will make more money as its margins improve with cost savings from 65nm. Although Intel's reorganization has so far only brought tiny changes in cost reduction it could see more in the 2H of 2007.
This does bring up the question of whether Intel will be able to bring additional price pressure to bear against AMD. Intel's Q2 07 earnings suggest that Intel reached its lower limit in pricing in Q2 and that it would need additional cost savings to be able to price lower. This plus the Q2 07 reductions in ASP for both server and mobile make it unlikely that we will see much in the way of lower prices during the rest of 2007. However, Intel is almost certain to resurrect this tactic in some fashion in 2008 as its ramping 45nm production reduces costs again. This should be interesting since AMD's ramping K10 desktop production should raise its desktop ASPs. It wouldn't surprise me to see a substantial bump of AMD's desktop ASP to $60 with a small cut of Intel's down to $79. This is possible if everything goes well and Intel still wants to keep prices down. Otherwise I would expect Intel to pull its desktop ASP back up to its preferred level of $99 and AMD to increase its to a preferred level of $70. This will likely be dependent on Intel's flash spin-off not being a $300 Million a quarter drain and AMD's new chipset division earning a profit. Mostly this means that Q4 07 will be more of a skirmish than the major battle that was expected. Presumably this will become a genuine battle in 2008 as Intel ramps Penryn while AMD ramps K10.
Some people seem to assume that Intel will ramp 45nm quickly and have large volumes available in 2007. However, the following ramp graph from Intel shows that 45nm will only be about 3% of production before the end of 2007. 45nm won't be a significant desktop volume for Intel until Q2 08 with crossover occuring the following quarter. Again, this why 2008 will be the real battle.
149 comments:
Talking about ASP should provide a much more accurate picture than just discussing the latest posted cpu prices.
scientia said: This does bring up the question of whether Intel will be able to bring additional price pressure to bear against AMD. Intel's Q2 07 earnings suggest that Intel reached its lower limit in pricing in Q2 and that it would need additional cost savings to be able to price lower.
Q2'07 was CAPEX heavy which hit intel by 2-3% points in its margins. Q3'07 margin targets is 52% with significantly less CAPEX which Intel can sacrifice in a price war if needed. At this margins AMD is set to lose ~$500M per quarter. How much more price pressure do you want Intel to provide when it's quite obvious its already very severe for AMD.
roborat
Yes, that makes sense but AMD is likely to have a margin bump of 5% in Q3.
From this article: :link:
Four of the five Wolfdale processors will have core frequencies of 3.16GHz, 3.0GHz, 2.83GHz and 2.66GHz, while the frequency of the remaining one is unknown. All five Wolfdale processors will feature 6MB L2 cache, detailed the sources.
Three of the four Yorkfield processors will have core frequencies of 2.83GHz, 2.66GHz and 2.5GHz with the remainder as yet unknown. Three will include 12MB L2 cache while the 2.5GHz version will have 6MB
----
Interesting isn't it? Out of all Penryn dual and quad core CPU's they only have 2.5GHz Quad Core with 6MB cache (meaning that it is made out of two dual core dies with half the cache disabled). Are they planing to assassinate AMD with second hand Quadies?
Are they planing to assassinate AMD with second hand Quadies?
They are trying to sell everything they make including "broken" cores. Nothing wrong with that, even in the future they can replace the "broken" core with native 3MB L2 parts which will lower the price or increase the profit.
I just don’t understand why Intel didn’t make any Core 2 Quad with 2MB + 2MB L2 parts; performance would go down because of the Double dual core? Even if Intel released the Q6600 266$ with two 2MB cores instead of the two 4MB, they would still sell those easily at that price with even more profits.
INTEL has lots of pricing ability. Recall the bulk of 45nm capital has been spent and they should have all that capacity starting to ramp. The 65nm capacity is now fully leveraged. INTEL margins were down, BUT don't forget that they still made billions!
Barcelona at the speeds they are releasing will offer little relief. WIth it being native quadcore reveue per wafer will actual be down at current yields.
INTEL can easily lower prices and drop margins another 2-5% and still turn a profit and cause AMD another billion dollar 2nd half loss. Its more a question does INTEL want to see their stock drop to the high teens do drive AMD into BK and single digit stock prices. I guess Paul will decide how weak a financial AMD he wants going into 2008. Personally I'd push AMD to another couple 300 million loss quarters. That will insure their 45nm never comes and they are stuck milking 65nm till 2009.
AMD is finished
Scientia we know that Intel managed to get back in 6 months the market share they lost due to k8 superiority. Now please try to imagine all those users with 965 or P35 chipsets which are easily upgradeable to quadcores (both 65nm and 45nm). Who will even bother to buy an inferior phenom chip (which will also require a mobo) when they can buy an Intel quad core dirt cheap (which will easily outperform all phenom chips up to 2Q08 at least)?
Now please tell me where exactly do you see a battle between K10 and penryn? There will be no battle unless miraculously AMD release a 2.6 - 2.8 phenom before 2Q08 (which we can surely assume that won't happen).
I agree with lex. AMD is finished.
gdp77
"Scientia we know that Intel managed to get back in 6 months the market share they lost due to k8 superiority."
I'm sorry but this statement is completely wrong. The actual situation was mentioned in the article and I have no idea how you missed it. AMD's average volume share for years was 16.5%. This was before K8. This rose to 18% in 2005 and 23% in 2006. In Q2 07 AMD's volume share was 23%. The only gain that Intel has had is in server share.
"all those users with 965 or P35 chipsets which are easily upgradeable to quadcores"
True. But surely you must understand that the number of upgrades will be less than 1% on the desktop and far less for notebooks. The strongest area of upgrades is servers.
"Who will even bother to buy an inferior phenom chip"
Inferior to what? I'm sure that Phenom will be as good or better as any Intel processor with a similar price.
"(which will also require a mobo)"
??? Why you need a new motherboard to upgrade? Quad Phenom will drop into socket AM2 or F.
" when they can buy an Intel quad core dirt cheap (which will easily outperform all phenom chips up to 2Q08 at least)?"
Well, because your statement is false. The more expensive quad cores will out perform AMD's quads but not chips at the same price. Also, AMD will almost certainly have 3.0Ghz quads by Q1 08 so Intel is unlikely to have much more performance with their fastest 3.33Ghz chips.
"There will be no battle unless miraculously AMD release a 2.6 - 2.8 phenom before 2Q08 (which we can surely assume that won't happen)."
AMD is likely to have 3.0Ghz in Q1 but even if they didn't you must be smart enough to understand that the battle will not be with the tiny fraction of chips at the top. The main battle is in the high volume mid range.
lex
It is clear that Intel is unwilling to have a quarter worse than Q2 07. This severely limits their ability to drop prices arbitrarily. Which means that Intel can only succeed if they can reduce costs faster than AMD.
AMD is likely to have 3.0Ghz in Q1 but even if they didn't you must be smart enough to understand that the battle will not be with the tiny fraction of chips at the top. The main battle is in the high volume mid range.
Well, if u got the fastest chip then u can set both the price and performance for the middle range. That was the story with K8, C2D, etc.
AMD will almost certainly have 3.0Ghz quads by Q1 08
I am going to screenshot that one and post it here after 6 months. We will see who is right and who is wrong.
gdp77
"Well, if u got the fastest chip then u can set both the price and performance for the middle range. That was the story with K8, C2D, etc."
I suppose that is an interesting superficial argument but it is completely false. K8 had better prices because at any given price Prescott pulled more power and was hotter than K8. We see a similar situation with C2D where it has lower power draw but also much stronger SSE performance at a given price. Suggesting that this is only related to the top speed is absurd.
Sorry if the following text sounds weird, I've been pulling double shifts at work to finish the project and am a "bit" tired.
scientia
"Today, AMD is ready to offer native quad core which it will take Intel about a year to match."
True but being native means nothing, delivered performance is where everything is. Nobody cares if you have tiny midgets remembering numbers instead of CPU in their box for as long as they do their job fast enough.
So far I haven't seen glued quadcore being too bad. Native vs glued won't matter for almost anyone.
"In early 2006, Intel previewed the native dual core 2.93Ghz Conroe which looked great and then it was a matter of waiting for Intel to actually get them out the door in volume"
Yes, they showed them months before launch and launched them together with all the other Core2's. AMD showed their 3GHz models and don't even have a timetable of when will they be availiable. All AMD has showed were 1.6/2GHz models couple of months before launch and will launch at those speeds.
Also Intel has showed their 45nm CPUs at 3.33GHz quite some time ago and from what I knoe those will be released by the end of this year.
I'd say there is quite a lot of differences between showing those CPUs.
"Today, AMD is offering a new architecture that is far ahead of its previous generation offerings while Intel is only able to offer secondary upgrades such as small clock increases, SSE4, and faster bus speeds."
So it is. Pipe and tick-tock are out-of-sync so they will be leapfrogging each other, assuming one or the other won't lag behind some time.
"Turion has only had a small effect on Intel's mobile share but AMD should be fully competitive in 2008 with Griffin and Puma"
Yes, things will improve for AMD but Intel said it intends to have over 40% (or was it >50%?) of its mobile CPUs at 45nm by Q2 next year. So far we don't know what will 45nm bring to Intel mobile solutions but increased performance with reduced power are pretty much guaranteed. I have a few doubts about AMD increasing performance.
"Intel is putting up a good fight with lower power draw, more cache, and faster FSB but it won't be enough."
Enough for what exactly?
"The actual price cuts have been a lot less than most people imagine. Intel's overall ASP is only down 16% from the nearly steady value of about $99 that it had been for three quarters. AMD's drop is similarly down 17% from the previous three quarter average of $60."
How much has AMD and Intel ASP dropped year-over-year during the last three quarters? From what I understand you are talking about the ASP AMD had Q406-Q207. If not then please be more specific, it wasn't really clear in your article.
Btw, how high ASP would AMD need to break even assuming that costs and marketshare remain constant?
"AMD's desktop ASP is substantially lower than Intel's but if it remains steady AMD will make more money as its margins improve with cost savings from 65nm."
Isn't AMD already almost entirely on 65nm? How big part of its revenue comes from 90nm? Knowing that should tell us what kind of savings could AMD get from getting rid of that.
"This does bring up the question of whether Intel will be able to bring additional price pressure to bear against AMD. Intel's Q2 07 earnings suggest that Intel reached its lower limit in pricing in Q2 and that it would need additional cost savings to be able to price lower. This plus the Q2 07 reductions in ASP for both server and mobile make it unlikely that we will see much in the way of lower prices during the rest of 2007"
Technically Intel has had two pricecuts this year: one in April and one in July. First in beginning of Q2, second in the beginning of Q3. Both of those cuts were known months before they arrived. Q3 one was know already in Q1 this year. That makes me somewhat doubtful that Intel makes so sudden changes in their pricing plans.
Also I'm quite certain there will be additional pricecut when first 45m CPUs arrive later this year. Though technically not a cut, just a shift. There will be new top performing CPUs and older ones will drop by one level. Still even that kind of shift will be duifficult for AMD to match.
Another interesting battle will be between AMD and NVIdia. Last quarter graphics department made 200M revenue and 50M losses while NV had records in both. OEM's were supposed to like slower R600 based GPUs but rumours say they are not that interested any more. We shall see more about that in Q3 though.
"Yes, that makes sense but AMD is likely to have a margin bump of 5% in Q3."
How much will it help?
aguia
I just don’t understand why Intel didn’t make any Core 2 Quad with 2MB + 2MB L2 parts;"
Me neither. The tasks that matter to me can run on 2M dualcore just fine and bandwidth limit is nowhere near in sight. I'm quite sure some 2.13GHz/2+2M quad for $200 or less would be quite good for me.
scientia
"The strongest area of upgrades is servers."
Are you sure? The time I worked at one company that managed around 10-15 different servers never upgraded any of those. If performance wasn't good enough they just bought new ones and gave the old ones new tasks they could handle. In fact I've never heard of anyone buying anything else for servers than additional RAM and HDDs.
"I'm sure that Phenom will be as good or better as any Intel processor with a similar price."
I'd say that until Phenom doesn't have enough volume it will be priced higher than whatever Intel offers. If it is not then something is very wrong with AMD financial thinking.
"??? Why you need a new motherboard to upgrade? Quad Phenom will drop into socket AM2 or F."
Somehow I doubt there will be a lot of people who bought servers during last year and want/need to upgrade them soon. I'm not saying there are noone, just they will be insignificant. There will be much more people buying whole new systems and for them the drop-in compability doesn't matter.
"Also, AMD will almost certainly have 3.0Ghz quads by Q1 08"
Are you so sure that the demo with the 3GHz quad means they will get it out so soon when they haven't talked about too high clock speeds in any near future?
" Suggesting that this is only related to the top speed is absurd."
K8 is more than competitive in the low-end but still that doesn't seem to help too much against Core2. I wonder why so. Could it possibly because there are lots of people out there who see Core2 as leader in performance and get themselves a CPU with the same name but significantly lower performance without even thinking that there might be better alternatives at given price?
ho ho
Well, looks like your comment is going to need a lot of corrections.
"True but being native means nothing"
This is simply false. Intel delivered dual MCM core P4's but didn't actually match K8 until it had native. K10 native quad will be superior to Intel's MCM quad Penryns in certain ways. This is why Intel is dumping MCM quad and moving to native quad with Nehalem.
"Yes, they showed them months before launch . . . AMD showed their 3GHz models and don't even have a timetable"
Intel didn't have a timetable at first either.
"Pipe and tick-tock are out-of-sync "
False, they will be sync starting with Bulldozer in 2009.
"but Intel said it intends to have over 40% (or was it >50%?) of its mobile CPUs at 45nm by Q2 next year."
Q3 is crossover and mobile normally comes after desktop so I would assume Q3.
"So far we don't know what will 45nm bring to Intel mobile solutions but increased performance with reduced power are pretty much guaranteed."
Really? I'll bet you made the same false assumption with 65nm.
"I have a few doubts about AMD increasing performance."
No, Griffin is not focused on performance. If you'll look back at history you'll see that Intel was beating Turion with a lower performaing but better battery life Pentium M.
"Enough for what exactly?"
I thought that was obvious. It won't be enough to prevent the loss of server share for Intel.
"Isn't AMD already almost entirely on 65nm?"
No. At mid year 90nm still accounted for 40% of AMD's production volume.
"Another interesting battle will be between AMD and NVIdia . . . slower R600 based GPUs"
You completely missed the chipset battle which AMD seems to be winning.
"Are you sure? The time I worked at one company that managed around 10-15 different servers never upgraded any of those."
Yes, as I've already explained the upgrade market is small for desktops. The vast majority of computer buyers will never open the case much less change cpu's. It is not unusal to swap out thousands of processors for high end HPC servers. The common mid and low end servers are not usually upgraded though.
"I'd say that until Phenom doesn't have enough volume it will be priced higher"
You are missing the point. He tried to claim that AMD Phenoms wouldn't sell because Intel cpu's would be cheaper. The fact is that AMD will price as high as they can and still sell.
"Somehow I doubt there will be a lot of people who bought servers during last year and want/need to upgrade them soon."
You are again missing the point. He claimed that a new motherboard would be required which is not true.
"Are you so sure that the demo with the 3GHz quad means they will get it out so soon when they haven't talked about too high clock speeds in any near future?"
I figure 6 months which would be Q1 08.
"K8 is more than competitive in the low-end but still that doesn't seem to help too much against Core2. I wonder why so."
Because even low end C2D's still have much better SSE.
ho ho
Okay, I found what you were talking about and it is Q2.
Digitimes: Intel 45nm notebook CPUs to account for 50% of total shipments in 2Q08
45nm-based Penryn processors for Santa Rosa Refresh will launch in first quarter of 2008 and are expected to account for 20% of total notebook CPU shipments for the quarter. In the second quarter, CPUs supported by the Montevina platform will launch and account for 15% of shipments, while the proportion of Santa Rosa Refresh CPUs will increase to 35%.
So, yes, 50% of mobile processors should be 45nm in Q2 08.
False, they will be sync starting with Bulldozer in 2009.
No, ho ho is correct. They are out of sync. If you examine both companies roadmaps based on when they say they would launch new products.
The Barcelona (new arch) is lined up with Penryn (Merom shrink). Shanghai (Barcelona shrink 2H'08) is lined up with Nehalem (new arch 2H '08). Bulldozer (new arch 1H'09 - mid'09) lined up with Westmere (Nehalem shrink 2H'09)
(Shrink is probably not the right word to use here. It simply means the porting of the existing design to a new process but adding a few features - more cache, instructions, tweaks etc). So AMD's new arch is always up against Intel's shrink of the previous arch.
Wallachian
Q2 06 - Woodcrest
Q3 07 - Barcelona
Trailing by 5 quarters.
Q4 07 - Penryn
Q3 08 - Shanghai
Trailing by 3 quarters.
Q4 08 - Nehalem
Q2 09 - Bulldozer
Trailing by 2 quarters.
Q4 09 - Westmere
Q1 10 - AMD 32nm (assuming they stay on schedule)
Trailing by 1 quarter.
scientia
"Intel delivered dual MCM core P4's but didn't actually match K8 until it had native"
And Netburst being simply massively inferior had nothing to do with it? Funny that for me going to dualcore P4 gave great speed increase over the older singlecore, pretty much every task got twice as fast as with singlecore. Later going to Core2 gave additional performance but that was simly from faster CPU.
"Intel didn't have a timetable at first either."
At first, yes, but it did have it a few months before it released them.
"Really? I'll bet you made the same false assumption with 65nm. "
History doensn't always show what future will bring. First Penryn benchmarks have showed us both, I can't think of a reason why should mobile CPUs be any different, especially considering the new sleep states.
"You completely missed the chipset battle which AMD seems to be winning."
I didn't miss it, I just wasn't talking about it. Graphics was what I was talking about. Also that 50M loss included chipsets too. Also AMD can't sell more chipsets than AMD has marketshare. NV sells for both.
"It is not unusal to swap out thousands of processors for high end HPC servers"
Somehow I doubt that HPC servers have are that big part of the market.
"You are again missing the point. He claimed that a new motherboard would be required which is not true."
It is true for all the people who didn't buy a new system during last year, that was my point.
"I figure 6 months which would be Q1 08."
My money would be on H2 next year. We'll see who got closer.
"Because even low end C2D's still have much better SSE."
My point was that similarly priced low-end CPUs also have similar performance, despite that Core2 has much better performance in some areas. Do you not agree that in <$150 CPUs AMD generally has better offer for the money? If you do then why doesn't those sell as much as Core2's?
Are you really so sure that AMD won't push any of its roadmaps back? It pushed K10 back for 1-2 quarters. Also you could say that Intel pulled its 45nm forward by 1.
My point is that trying to predict what will happen in 1-2 years or later will not be all that accurate, especially considering AMD financial position.
2% of all shipments for the 45nm processors seams very low to me.
Does this mean:
-It will be very late Q4 (December)?
-Initial ramping of the 45nm Intel FAB will be slow?
-Bad yields?
-Still give preference to mature 65nm products?
-Most parts will be quad core so it will cut the production in half (4% in production became 2% in products)?
aguia
"2% of all shipments for the 45nm processors seams very low to me."
Way higher than Barcelona numbers.
"-It will be very late Q4 (December)?"
11 November
"-Initial ramping of the 45nm Intel FAB will be slow?"
IIRC they only have one 45nm FAB now but it'll have two additional soon next year.
"-Most parts will be quad core so it will cut the production in half (4% in production became 2% in products)?"
45nm quadcore takes less die area than 65nm dualcore
You completely missed the chipset battle which AMD seems to be winning.
What battle?
The OEM battle?
Because almost all motherboards selling in my country for AMD platforms are NVIDIA, one or two are the AMD 690 but that’s about it.
Without IGP even if I would like to buy one AMD 480 or 580 I couldn’t, there isn’t one computer store that sells those. Maybe this also explains the almost inexistent crossfire systems.
ho ho,
45nm quadcore takes less die area than 65nm dualcore
Current Core 2 Duo: 143 mm2 at 65 nm
New Core 2 Duo: 107 mm2 at 45 nm
Quad new Core 2 Duo 214mm2 at 45 mm
Ho Ho
"And Netburst being simply massively inferior had nothing to do with it? Funny that for me going to dualcore P4 gave great speed"
It may have worked for you but it was obviously false in general cases. If the simple MCM design had been adequate Intel would not have bothered with the much more complex hybrid bus design in Tulsa.
Most of the rest isn't worth quoting. Basically you have great confidence in Intel with no proof and are pessimistic about AMD, again without proof.
"My money would be on H2 next year."
So, you believe that it will take AMD an entire year to release a demoed chip. That is absurd.
aguia
I added Intel's 4nm ramp slide to the article so that people could see that Penryn won't be battling anything in 2007. The 45nm ramp for Intel had been exagerated over and over with people claiming Q3 introduction and crossover in Q1. We can see that the Q4 production is tiny and good volume isn't until Q2.
Ho Ho is also incorrect about the 45nm/K10 numbers. Intel won't at any point have a higher ratio of 45nm than AMD has of K10. For example AMD will begin ramping Shanghai before Intel is at 100% 45nm. Even in absolute numbers Intel should not pull ahead with 45nm until Q2 08.
However, I'm not really trying to talk about 2008. It is far too early to see what will happen with Penryn and K10 in Q1 08, much less the rest of the year.
Finally, there are 35 different motherboards now that use the 690G chipset. These should be available in this quarter. Also, the AMD chipset has been more popular than the nVidia chipset and this has not changed.
he 45nm ramp for Intel had been exagerated over and over with people claiming Q3 introduction and crossover in Q1.
How much do you think this will affect Intel Tick-Tock campaign?
They explicitly said new architecture in every 2 years with something in the middle. Intel "new" architecture was released in July/2006, however Penryn didn’t see the light of the day in July/2007 (I’m betting late December/2007) with paper launch in November 2007 and send to review in September, to keep people talking about Penryn putting Phenom in the shadow.
So do you think Nehalem will also be late?
Sorry about the die size numbers, I don't know what was messing up then. Still they are considerably smaller than similar things on 65nm so cost cannot be the factor for not ramping it fast.
I also didn't have time to ansver some of the older posts.
scientia
"This is why Intel is dumping MCM quad and moving to native quad with Nehalem."
Perhaps the reason is that Intel has put a few numbers together and found that MCM brings in more money on 65nm and at the beginning of 45n than monolithic die?
"It may have worked for you but it was obviously false in general cases."
What would be a general case? Servers are only a tiny part of the market, around 10% for AMD according to your claims
"If the simple MCM design had been adequate Intel would not have bothered with the much more complex hybrid bus design in Tulsa."
Servers need a lot of bandwidth, you can't push too much data over a single FSB bus. Also things like databases need quite a bit of synchronization over threads, doing it faster will increase performance. Most home users do not use databases every day and for multi(mega)tasking MCM over single FSB is good enough.
"So, you believe that it will take AMD an entire year to release a demoed chip. That is absurd."
I call getting 50% clock speed increase in half a year absurd.
"I added Intel's 4nm ramp slide to the article so that people could see that Penryn won't be battling anything in 2007."
Still, having one >3GHz cpu to replace the current 3GHz ones is enough to cause a price shifts all over the place. Unless of course intel decides to drop current 3GHz to around $800 and sell the new extreme version at $999/~$1100.
"Intel won't at any point have a higher ratio of 45nm than AMD has of K10."
How fast does AMD intend to ramp K10? Anyone remembers how fast Intel got to shipping more Core2 (quad) Xeons than older CPUs? IIRC quadcores sold like hotcakes and were quickly accepted. Can AMD possibly repeat that considering it will not have as much performance as Intel for most tasks for the same price, at least not this year and possibly not before well into Q1 next year?
"Finally, there are 35 different motherboards now that use the 690G chipset"
Number of different models is pretty much meaningless, what we need is data on number of sales. Fact is that AMD only sells chipsets for AMD cpus. NV also sells for them and Intel has nearly 75% of marketshare. Against who exactly is AMD winning the chipset war you are talking about?
You said before that "mobile comes after desktop". Usually that has been true since mobile CPUs need to be using less power and mostly it takes time to deliver such chips at a new architecture/technology. Doesn't the fact that Intel intends to ramp mobile before desktop hint that they have good thermals and power usage on 45nm?
Btw, do you know how big part of overall sales do the mobile CPUs make up for Intel?
aguia
"How much do you think this will affect Intel Tick-Tock campaign? "
It shouldn't effect it at all.
"Intel "new" architecture was released in July/2006, however Penryn didn’t see the light of the day in July/2007"
Right, but Intel released 65nm in Q4 05. 45nm comes out Q4 07 which is right on the two year cycle. BTW, this is in direct contradiction to people who claim that Intel "moved Penryn forward" to Q4 07. Intel didn't move 45nm forward; they just kept it from slipping.
"So do you think Nehalem will also be late?"
No. I think Nehalem will be released Q4 08.
No. I think Nehalem will be released Q4 08.
But thats late... ;)
July 2006 + 2 years = July 2008
Unless you count 1 year after Penryn release.
Ho Ho
"Still they are considerably smaller than similar things on 65nm so cost cannot be the factor for not ramping it fast."
The reason has nothing to do with a preference by Intel. The 45nm ramp is the fastest that Intel can get the new process distributed and producing at good yield.
"Perhaps the reason is that Intel has put a few numbers together and found that MCM brings in more money on 65nm and at the beginning of 45n than monolithic die? "
No. Intel is moving to a monolithic die to be competitive in performance with K10. MCM reduces cost and performance both. It should be common sense that Intel would not take the hit on a die size more than double that of Penryn if they didn't have to.
"Most home users do not use databases every day and for multi(mega)tasking MCM over single FSB is good enough."
True but not as good as native which is why Intel moved to native dual core for C2D and is moving to native quad for Nehalem. This has nothing to do with a preference by Intel; it is in fact simply the fastest that Intel can deliver native versions. For example, it is about 2 years between X2 and Barcelona and just over 2 years between C2D and Nehalem. The extra time for Nehalem makes sense because C2D doesn't have a Crossbar like K8.
"I call getting 50% clock speed increase in half a year absurd."
Strange because Intel did the same thing with C2D 2006. The demo versions were slower yet two revisions and 6 months later Intel was at 3.0Ghz.
"How fast does AMD intend to ramp K10?"
I can only estimate it. As far as I know AMD has never given a ramp projection. Based on the 65nm conversion it looks like AMD could do it in 3 quarters. This would mean 100% K10 by mid year (June/July) 2008.
"Against who exactly is AMD winning the chipset war you are talking about?"
AMD's 690G chipset is more popular on AMD boards than nVidia's.
"Doesn't the fact that Intel intends to ramp mobile before desktop hint that they have good thermals and power usage on 45nm?"
I don't know. The talk lately on the Apple boards has been that the 45nm mobile chip will not be as good on power draw.
"Btw, do you know how big part of overall sales do the mobile CPUs make up for Intel?"
40% of the total cpu revenue.
Aguia
"But thats late... ;)
July 2006 + 2 years = July 2008
Unless you count 1 year after Penryn release."
No. Woodcrest was early. This would have been the normal tick tock schedule:
Q4 05 - 65nm Presler
Q4 06 - 65nm (new) Woodcrest
Q4 07 - 45nm Penryn
Q4 08 - 45nm (new) Nehalem
Q4 09 - 32nm Westmere
Presler was the first 65nm chip but by the tick tock upgrade schedule it should have been Tulsa. However, Intel was late with dual core in 2005 so that pushed Tulsa back. The proper tick tock schedule would have been 65nm Tulsa in Q4 05 and then the new architecture Woodcrest in Q4 06 then Penryn Q4 07.
scientia
"The reason has nothing to do with a preference by Intel."
I was actually replying to aguia who claimed that going to monolithic core would cut its production in half.
"No. Intel is moving to a monolithic die to be competitive in performance with K10."
It could also be that it didn't make much sense before. With Nehalem they had time to do it properly. Penryn is just an improved Core2.
"MCM reduces cost and performance both"
Yes, it does. Question is by how much. I still remember I was very close to buying one of those ~$450 x2 3800's but I decided that much better OC'ing and considerably cheaper (~$350) 2.8GHz P4 was better for the money. Had the price difference been smaller I had definitely bought X2.
"True but not as good as native"
I'll ask again: how big is the speed difference between native and non-native CPUs. Give me some rough number tha would accounmt for most used tasks for the market that buys most of the CPUs (== the ones who companies target their CPUs). I know that for some tasks there is a speed difference between native and non-native but it is almost never too big, at least not for vast majority of usecases. If you believe otherwise please try to explain it.
"Strange because Intel did the same thing with C2D 2006. The demo versions were slower yet two revisions and 6 months later Intel was at 3.0Ghz."
How long before K10 launch did AMD show its 3GHz CPUs? What were the clock speed it launched those CPUs? Am I wrong to say that AMD showed 3GHz roughly 1-1.5 months before launch and will launch at 2GHz max whereas Intel had 2.93GHz availiable from day one? Can you spot the difference?
"This would mean 100% K10 by mid year (June/July) 2008."
That is excluding mobile CPUs, of course.
"AMD's 690G chipset is more popular on AMD boards than nVidia's."
You mean on the roughly 20% of the market. NV also sells chipsets for some of the CPUs that make up the rest of the market. Also do you have any numbers that would support your claim? Something like revenue comparison would be nice. As I said before the sheer number of different configurations is pretty much the most meaningles metric one could think of.
"I don't know. The talk lately on the Apple boards has been that the 45nm mobile chip will not be as good on power draw."
Anywhere that I could read about it?
I still remember I was very close to buying one of those ~$450 x2 3800's but I decided that much better OC'ing and considerably cheaper (~$350) 2.8GHz P4 was better for the money.
ho ho,
you must be confusing something, the price of one X2 3800+ at launch I think it was around 350$.
And that processor OC easily. The performance difference was also abysmal on some applications.
ho ho,
you could have saved money too:
Look here:
Power Consuming
full system power consuming at load:
AMD X2 3800+ 166W
Intel P4 660 235W
Pentium D 840 292W
Pentium D XE 313W
Intel dual core system not only performed slower, it consumed almost 2 X in energy.
Curious too, same processor with HT enabled consumed more 21W, seams little but its almost as one Pentium M.
I have shown percentages for Intel's 45nm ramp.
aguia
"you must be confusing something, the price of one X2 3800+ at launch I think it was around 350$."
I was quoting the prices those were sold in Estonia. The time I was buying my CPU x2 cost around $300 in US and P4D ~$240. EU adds additional 18% tax to that and of course all those middlemen who brought the CPU here also want to earn something. Pretty much every piece of electronics is considerably more expensive in EU and Estonia than in US.
"And that processor OC easily. The performance difference was also abysmal on some applications."
When I later got to compare my P4d @3.9GHz against x2 @2.7GHz in the applications I used they were pretty much equal. As I said Intel was considerably cheaper for me.
About power consumption, those are 90nm dualcores. I had 65nm one that used considerably less power.
INTEL can easily lower prices and drop margins another 2-5% and still turn a profit and cause AMD another billion dollar 2nd half loss. Its more a question does INTEL want to see their stock drop to the high teens do drive AMD into BK and single digit stock prices. I guess Paul will decide how weak a financial AMD he wants going into 2008. Personally I'd push AMD to another couple 300 million loss quarters. That will insure their 45nm never comes and they are stuck milking 65nm till 2009.
So what you'r sayign is that Intel can flew their "monopoly-muscle" all they want and it doesn't matter?
Right now the price of C2D and C2Q at the same clock speed could be considered "below cost."
If C2D at 2.4GHz is selling retail for $229 and C2Q at 2.4GHz is selling retail for $299, that means the second chip, FSB packaging and tweaks are only costing consumers $70.
Things that make you wonder.
Christian Howell
If C2D at 2.4GHz is selling retail for $229 and C2Q at 2.4GHz is selling retail for $299, that means the second chip, FSB packaging and tweaks are only costing consumers $70.
Intel have basically discontinued the 2.4 GHz C2D so its price is no longer valid for this comparison. Here is the proper comparision (Newegg prices):
E6550 (2.33 GHz) - $179
E6750 (2.66 GHz) - $212
Q6600 (2.4 GHz) - $290
Once demand settles down, the Q6600 will drop another $10 or so. So the premium for the extra two cores and packaging is more like $90. From a purely time + materials standpoint this sounds reasonable to me, considering how little AMD dual-cores are selling for nowadays. $90 leaves room for a bit of profit (considering that most of the 775-pin package is unchanged) and is certainly not predatory pricing.
Nevertheless, it's clear to me that Intel are pricing the Q6600 this low specifically to put AMD in a tough spot. If Intel weren't trying to price AMD out of business, they would certainly place a 200%+ premium on quad cores just like with their Extreme Editions or AMD's FX series.
"When I later got to compare my P4d @3.9GHz against x2 @2.7GHz in the applications I used they were pretty much equal. As I said Intel was considerably cheaper for me."
You can buy an Athlon X2 at 2.6GHz that consumes only 65W, while I don't know if there is any 90W or lower P-D over 3GHz, and it takes a 3.9GHz P-D to basically match the performance of a 2.6GHz A-X2.
Pentium D might be cheaper, not because its easier to make, but because nobody wants them. Of course except Intel die-hards like you who can possibly claim its "advantage". Pentium D's run slower, consume more power at idle and under load. The cheapest P-D box you can buy won't be cheaper than that of A-X2, yet the latter will be better performing.
As I have said before, at this moment, the smart purchase for mid or low end desktops are both Athlons. C2D is the winner at and above E6600 or E6550. Of course not everybody is smart.
Right now the price of C2D and C2Q at the same clock speed could be considered "below cost."
Untrue. As long as Intel continues to post large profits well in excess of $1bn a quarter there's no chance of AMD claiming that Intel is using predatory pricing. AMD started this nonsense with pricing dual core CPUs below $70. When Intel introduced C2D the price for the entry level E6300 was $183. A reasonable, fair price. AMD's insistence to "gain market share at all cost" even if it meant sacrificing profit is what got them into this mess.
If Intel can price the Q6600 at $266, and sell a ton while making profit there's certainly nothing illegal about that.
abinstein
"Pentium D might be cheaper, not because its easier to make, but because nobody wants them. "
I as a customer couldn't care less why one is cheaper. For the tasks I ran it delivered more bang for buck than X2. Had I waited for a few months for AMD to drop their prices things would have been different.
Btw, why is AMD CPUs currenyly cheaper than Intel ones? Because nobody wants them or because they are cheaper to produce? It cannot be the latter because then AMD would not be making losses.
"The cheapest P-D box you can buy won't be cheaper than that of A-X2, yet the latter will be better performing."
I've been telling the story of why I bought P4D instead of X2 several times and every single time you have said that completely missing the main reason.
"As I have said before, at this moment, the smart purchase for mid or low end desktops are both Athlons."
Depending on needs, there are situations where even the lowest end Core2 can beat similarly priced X2's. In general case, yes, x2 will be better.
fujiyama
Stop being a troll. I expected K10 to arrive in late Q2 07 so late Q3 07 is one quarter late. I've mentioned this before.
"It really is remarkable how similar each company's situation is to its competitor's last year"
That premise is simply laughable. ON first glance Scientia's makes what sounds like a logical review of the past couple years but he simple forgets a few things.
1) INTEL has had the dominate market share, greather then 70% for the past many years.
2) INTEL has made billions profit / quarter during the best of years and hundreds of millions in the worst of quarters
3) AMD has lost hundreds of millions in worst quarters and millions in average quarters
4) AMD made tens to a hundred million in its best quarters
5) INTEL out invests in core technology and manufacturing that insures it continues to keep a 1 year lead minimum in technology. That results in the ability to offer double the cache, or 1/2 die size for a comparable design
6) AMD is not only behind on technology but uses inferior SOI that greatly hampers cache performance. That directly is why they have IMC wasting valuable silicon capacity and why they have very small caches. They are now even more behind in being late with IBM HighK/metal Gate not appearing before 32nm.
The only reason AMD is still around is INTEL is simply too greedy to drive them out of business. INTEL can easily in 3 quaters choose to price AMD out of busines. It would be prefectly legal as INTEL would still sell at a profit.
Simple bottom line, AMD is only around because INTEL chooses to allow it. If INTEL executes there is little that Hector, Dirk, Henry can do. They can survive, churn out a nice design, but they will always loose. The only reason they are still around is because INTEL allows them.
Root for AMD all you want, try and argue why AMD will make a comeback. Maybe for a short moment it'll look like they got a chance, but the bottom line is they are only around because of either INTEL incompetence or intel executive choice.
AMD is irrelevant.
I found this comment funny in a recent interview with Hetor. It only goes to show that its all smoke an mirrors about what is wrong with Barcelona
Q What caused the Barcelona delay?
A It's awfully complicated. We're doing something that nobody has ever done. As strong as our competitor is, they have not done a quad core on a chip (four chip cores, or brains, laid out on a single piece of silicon instead of on two pieces attached together). . . . Every time we ran into a gotcha (or technical glitch), it created a six-week or so hole in the schedule as we went back and fixed it. We hoped we wouldn't get many of those, but in the Barcelona case, we got more than we thought. By the time we got through fixing them all, we were six months-plus late from where we originally wanted to be.
Hector is somehow trying to blame the delay of Barcelona on the fact it is 4 cores on a single chip. Or impress us wiht AMD's feat to do something INTEL couldn't. How about AMD losing 1.5 billion dollars, that is something INTEL couldn't do either. Could it be yield, could it be the fact getting 4 cores with similar performance a problem. Or could it be the fact the design is busted and things like clock skew and power consumption trying to clock the whole damm chip is the problem. Damm why would a company late on a technology node be so stupid to try and integrated a huge quadcore before going to 32nm. Who does AMD think it is, INTEL. ONly intel has the capacity and engineering muscle to do this with Merced, Williamatte and other huge die done before their time. AMD bet, and bet wrong. Its a bet the board should have never let Hector make and now AMD is ruined.
ONly intel has the capacity and engineering muscle to do this with Merced, Williamatte and other huge die done before their time
Intel have sucessfully engineered and produce the Tulsa CPU. That's a whopping 1.4bn transistors. Of course, most of that is due to the massive 16MB shared L3 cache. Tigerton will be cheaper to produce; being only 580m transistors. (Tulsa had a 1MB l2 for each core; then a massive 16mb shared L3. Tigerton simply has a 4MB shared L2 on each die; for a total of 8MB L2 cache.
scientia
"I expected K10 to arrive in late Q2 07 so late Q3 07 is one quarter late."
Yes, you did. AMD expected to launch it sooner than Q2.
Technically speaking, Barcelona is not late. AMD taped it out at around August last year. And in CPU design, it theoretically takes 4-5 quarters (12-15 months) from tapeout to launch.
Now AMD is coming in at the earlier end of that - 12-13 months. So August to September is about right. AMD is launching it on time.
The problem is with their big-mouthed management - Henri, Randy and Mario. I mean, do these guys even know what it is to design chips. Did AMD seriously believe that it could launch a 65nm monolithic quad core with a new microarchitecture/new interfaces/new cache technology off an A2 stepping. Theoretically that is the only stepping they could have launched for the mid-q2 or end-q2 timeframe. Realistically speaking you need to go through a few major revision steppings (B, C) to get a fully working bug-free CPU (bugs -> functional, speed)
I think AMD executives just got a dose of reality with their Barcelona launch. There was quite a lot of hopeful lines been dropped by their executives, which were unrealistic. Wonder what else will turn out unrealistic in Barcelona...
Yes, you did. AMD expected to launch it sooner than Q2.
Never heard that. I always heard middle 2007. That could be end Q2,or beggining Q3 (July/August).
Also Nehalem will be taped out in October 2007, arrives Q4 2008 much in line with the Wallachian said.
Scientia, I caught a post about this article on HardOCP.com's forums before it went nuclear again... I would enjoy picking your brain about this anaylst from Goldman Sach's article about AMD vs Intel and their predictions/anaylsis for the next several years.
http://www.swsoft.com/r/summit/2007/Barcelona_presentation.pdf
Thanks in advance, keep up the good work.
"My guess is they internally already pushed the schedule forward long before they started talking about Q2/midyear release."
AMD had internally expected server K10 to launch in early Q2, followed by desktop some time in Q3. Eventually they had to delay server to late Q3 and desktop to late Q4. This is an almost 6-month delay for server and more than a quarter for desktop.
"Did AMD seriously believe that it could launch a 65nm monolithic quad core with a new microarchitecture/new interfaces/new cache technology off an A2 stepping. Theoretically that is the only stepping they could have launched for the mid-q2 or end-q2 timeframe."
Are we sure that Barcelona taped out last December? I thought it taped out earlier than that, but December was the first time it got the chips back (on 65nm).
"I caught a post about this article on HardOCP.com's forums before it went nuclear again..."
I know you intended to direct this to scientia, so ignore me as you wish. But sorry for being blunt, that document looks to me like a nice summary of lots of crap... :p
Untrue. As long as Intel continues to post large profits well in excess of $1bn a quarter there's no chance of AMD claiming that Intel is using predatory pricing. AMD started this nonsense with pricing dual core CPUs below $70. When Intel introduced C2D the price for the entry level E6300 was $183. A reasonable, fair price. AMD's insistence to "gain market share at all cost" even if it meant sacrificing profit is what got them into this mess.
If Intel can price the Q6600 at $266, and sell a ton while making profit there's certainly nothing illegal about that.
And I'm sure you believe that.
lex
You are making the same tired arguments that Intel fans have always made. Intel is big and I like Intel therefore they will win. Your theory that Intel isn't really trying is what is laughable.
Christian M. Howell responded to:
If Intel can price the Q6600 at $266, and sell a ton while making profit there's certainly nothing illegal about that.
with:
And I'm sure you believe that.
From Wikipedia:
...the U.S. Supreme Court has set high hurdles to antitrust claims based on a predatory pricing theory. The Court requires plaintiffs to show a likelihood that the pricing practices will affect not only rivals but also competition in the market as a whole, in order to establish that there is a substantial probability of success of the attempt to monopolize. If there is a likelihood that market entrants will prevent the predator from recouping its investment through supra competitive pricing, then there is no probability of success and the antitrust claim would fail. In addition, the Court established that for prices to be predatory, they must be below the seller's cost.
It doesn't mattter what any of US believe. It is what can be proven in a court of law. And the simple fact of the matter is that Intel may not be making a lot of additional money on quads over duals, but they aren't losing money either. They have advantages in economy of scale that AMD doesn't, and the law does not mandate a level playing field. Only that the pricing not be below cost.
How much do you think the incremental silicon, packaging, and test costs are? I can almost guarantee that it is far less than the incremental $90 in revenue for a similarly binned dual. It just means that the dual parts are higher margin- and there is a valid business case for the segmentation strategy:
1. Drive quad adoption with low prices (seed the market)
2. Maximize revenue by maintaining ASP/margin on the low end (better margins on low bind dual, which sell more than quads)
So where's the beef? AMD and Intel bet on different horses- AMD the "cool, native" one, Intel on the "time-to-market" one. I think we know which one wins from a business perspective. And these ARE businesses first- even if they aren't always well run ones.
Abinstein said...
"I know you intended to direct this to scientia, so ignore me as you wish."
It's all good Abinstein, I agree with your assessment probably more than you realize. I just wanted to see what Scientia has to say about it/debunk it, since it is from a Goldman Sachs analyst.(which I feel their analysis is equivalent to a "sach" of $&*+ if you ask me heh heh)
"But sorry for being blunt, that document looks to me like a nice summary of lots of crap... :p"
Refer again to the above for my comments. 8^]
Ho Ho
"AMD expected to launch it sooner than Q2."
I'm sorry but you are wrong. Q2 was the earliest that K10 was ever expected. The six month behind comment comes from the frequency, not the launch date. In other words AMD expected to launch in late Q2 at the frequencies it will have in Q4.
I realized that I didn't specifically link to the AMD internal server roadmap in my blog (which got the roadmap from the blog of Hiroshige Goto). It's quite clear that internally AMD did expect Barcelona server to be out around early Q2 2008. Barcelona server chips are late, with respect to AMD's internal plan, for almost 2 quarters.
abinstein
"But sorry for being blunt, that document looks to me like a nice summary of lots of crap... :p"
Can you or anyone else at least comment on the triplechannel DDR3? That surely looks interesting and it was the first time I heard about it.
scientia
"The six month behind comment comes from the frequency, not the launch date."
What makes you think that?
Lou Ceifer
I looked over the Novakovic Presentation. It's a bit like a curve ball. At first it seems to be going straight but then it veers further and further off the path. By the last few pages his presentation is nothing but a fairytale. I did a page by page error correction but that took 900 words which is nearly the length of an article. I guess this means I'll have to do another one soon and talk about 2008 and 2009.
lex
Do you honestly think you are the first person to argue that because Intel is bigger and makes more money that they will crush AMD? These argument have been used over and over since AMD introduced K5. It just amazes me that each person who does this thinks they are saying something new.
Things change quarter to quarter. It is beyond absurd to insist that the current profitability is valid next quarter, let alone next year. I see no reason to waste time with an "Intel is bigger and therefore better" argument.
Scientia
I did a page by page error correction but that took 900 words which is nearly the length of an article.
I would like to see that, maybe you could post it and keep it separate from a 2008 and 2009 article.
It would be nice to focus the comments attention to one thing, a break down of the PDF.
Just my opinion.
Although Intel's reorganization has so far only brought tiny changes in cost reduction it could see more in the 2H of 2007.
I don't know that I would call a $2 billion in savings this fiscal year small. Add that to AMD's bottom line and their entire outlook would be rosy. Even for Intel, that's like an extra quarter of profit for the year.
And IIRC that is supposed to increase to $3 billion next year.
Q2 06 - Woodcrest
Q3 07 - Barcelona
Trailing by 5 quarters.
Q4 07 - Penryn
Q3 08 - Shanghai
Trailing by 3 quarters.
Q4 08 - Nehalem
Q2 09 - Bulldozer
Trailing by 2 quarters.
Q4 09 - Westmere
Q1 10 - AMD 32nm (assuming they stay on schedule)
Trailing by 1 quarter.
Aside from whether or not AMD can do this, it seems a pretty expensive decision to me. Ramping 3 tech nodes in ~3 years seems like they are failing to milk all the value out of their investment before moving on to the next node.
All in all, an odd decision for a company that is insisting that the customer doesn't care which technology node a processor is made on.
Sure there is a cost benefit to moving to the next node. I'm not disputing that. But I find it interesting that at the same time they are publicly saying that it doesn't matter what tech node you are manufacturing on, they are trying to move heaven and earth to catch up.
Scientia, nice job digging up Intel's 45nm ramp plan. I really think that is a more meaningful metric than process introduction.
What I'd really like to see is AMD's ramp plan for the same node. I would have to treat it with a healthy dose of skepticism if it is too much steeper than Intel's.
Sure they only have to ramp 2 fabs, but they also have to keep them producing while they ramp. While Intel's ramp involves 2 new fabs starting up from scratch. I'd call the difference there a wash.
InTheKnow
"I don't know that I would call a $2 billion in savings this fiscal year small."
You're just repeating the numbers that Intel gave as their goal back in 2006. However, Intel has only saved about $200 Million so far in 2007.
It looks like Intel will meet about half of their 2007 goal if nothing else changes. This would be a real savings of about $600 Million in 2007 but it would save them $1 Billion in 2008. To meet Intel's 2007 goal they would have to eliminate 2/3rds of the flash memory losses which are running $300 Million a quarter. If Intel can do this in Q3 they'll have a real savings of about $1.2 Billion in 2007 but it would save $2 Billion in 2008
"And IIRC that is supposed to increase to $3 billion next year."
Again, you are repeating Intel's stated goal which they have not begun to meet.
To reach the 2008 goal they need to completely eliminate the flash loss plus get rid the other 6,000 or so employees.
Scientia
However, Intel has only saved about $200 Million so far in 2007.
I believe that Intel's savings are planned to come from OpEx.
While they may spend it else where OpEx is down $842 million in the first 6 months of 2007 vs the first 6 months of 2006.
They were aiming for an OpEx of $10.7 Billion for 2007 (Q1 confrence call) vs $11.97 Billion for 2006, which would be $1.27 Billion in savings for 2007.
Things change quarter to quarter. It is beyond absurd to insist that the current profitability is valid next quarter, let alone next year.
LOL, yup AMD is up to it lose 500 million for 3 quarters and it'll bounce back in a quarter to make 1.5 billion to recoup it all.
Remember when all INTEL had was Prescott then Cedar Mill for what 6 or 7 quarters. If you recall they still made billions a quarter. But eventually MS went down and AMD cleaned their clock.
Things are different now the gorilla got its mojo. AMD simple can't make the transistion from 90 to 65 to 45 then to HighK metal gate in 2 years. They simply don't have the capacity, invested the R&D nor most important have the knowledge or methodology.
Sorry AMD is behind, its because they are small, because they don't have enough MS to generate a revenue base to invest the required $ to develop the technology and manufacturing capacity.
If we were talking about a AMD with 15 billion in revenue and 2 billion in profits I'd shut up. But we aren't so the facts are the facts.
Its not old news its the facts. Facts that suck for Hector, Dirk and Henry. It don't matter what you spin AMD don't got the revenue base to support advance technology required to compete. You can argue about design, architecture and a great roadmap for 2009 and 2010. Bottom line AMD doesen't have the cash to execute the plan.
Size matters here little dicks
lex -
"AMD simple can't make the transistion from 90 to 65 to 45 then to HighK metal gate in 2 years. They simply don't have the capacity, invested the R&D nor most important have the knowledge or methodology."
I don't know how you can miss it, but AMD is not making the transition alone, at least until 32nm. In terms of process, AMD, IBM, Chartered, and a few Japanese companies are all on the same boat. Do you believe all of them will sink but your beloved Intel?
AMD will survive even if its process technology is 9 months behind Intel. It will do okay if the lag is just 3-6 months. It will do well if there is no lag or even lead.
Anand has some Wolfdale figures out and about. These results are about 5%, so less than the assumed 10% increase mentioned in the past, perhaps because no SSE4 benches were included (quite rightly in my opinion).
So it's not earth shattering, but of course it was never meant to be, just cheaper, slightly faster and slightly less power. The overclock was impressive on early motherboards and stepping though, I think Intel has got 45nm down pat.
It was interesting to compare the die shrink effect on cache latencies, here going up in speed whereas AMD got slower for 65nm. I still think that is not due to the supposed allowance for bigger L2 sizes, seems more to do with production amounts.
If Phenom can beat Conroe/Kentsfield then it should beat Penryn as well you would assume, but by reduced amount.
I don't know how you can miss it, but AMD is not making the transition alone, at least until 32nm. In terms of process, AMD, IBM, Chartered, and a few Japanese companies are all on the same boat.
In other words, only Intel has the expertise necessary for 45nm high-k process on it's own. Without IBM's help your beloved AMD would be stuck on 90nm indefinitely.
Anand has some Wolfdale figures out and about. These results are about 5%, so less than the assumed 10% increase mentioned in the past, perhaps because no SSE4 benches were included (quite rightly in my opinion).
5% IPC benefit is quite impressive with no SSE4. Of the tests Anandtech ran, the following could be heavily optimised for SSE4:
-DivX/Xmpeg
-Windows Media Encoder
-3dsmax 9
-Cinebench
-Lightwave 3D
I think it's fair to say that on average, when including SSE4, you'll see ~10% IPC gain with Penryn.
SSE4 won't give any definite advantage to Intel, AMD can implement SSE4 by the time it only begins to be used. SSE4 can't be counted as and Intel advantage at least not for long. Not to mention the fact that it takes months to include it in production compilers (can't make a stable release of a compiler in a matter of weeks) then the software comes, which either is simply recompiled (hardly ever happens that) or modifications to the algorithms are required to take full advantage of new changes, or in lots of cases hand coding the new SSE4 version/code path from scratch is done.
By the time SSE4 actually starts to be used, AMD can implement it too.
Now if Barcelona really were all that, why would executive Henri Richard resign?
abinstein
"I'm quite sure my $200 X2 4600+ barebone (MB+CPU) bought 6 months ago perform everything faster than your Intel P-D."
How much did the same thing cost in Q1 06?
Also that P4D was replaced around a year ago, I don't have it any more. I got around $150 for it when I sold it.
abinstein,
the reason that has driven ho ho to buy the Pentium D is the same that drives many people to buy AMD today, unbeatable price.
However at that time ho ho could have bought one single core Athlon that would be much cheaper and has fast or even faster than the Pentium D while consuming 1/4 (or less with C&Q) of the power of the Pentium D.
Everyone have their reasons.
On a side note
Henri Richard
is set to resign with Barcelona's launch.
Hector finally admits that Barcelona has given them technical problems and
thus the 6 Month delay.
His job was to let people know how competitive Athlon X2 are, but he apparently failed.
Not only that as the all platform too, here is a good example that the Intel platform is not that perfect:
Intel slow IDE/SATA controllers
Exactly why the latest Intel desktop chipsets have an apparent 80 MB/sec ceiling for sustained transfer rates with the SSD products is still a mystery to us and the drive manufacturers. We are still testing other Intel chipsets and will report these tests results and any updates from Intel or the drive manufacturers in our next article. In the meantime, using this drive with the Intel ICH9R provides the speed of Wile E. Coyote while we liken the NVIDIA 680i to the Road Runner: just a little faster and apparently a little smarter when it comes to SSD products.
It seams that the guys at Anandtech doesn’t know what happened with the Intel system, but I will explain since my company has been upgrading computers and keeping their old HDD.
What is happening is that NVIDIA is much faster, but no one noticed, but we (me and my colleagues) at our company did real tests and the "old" 40GB IDE Seagate hard drives worked much faster on the new NVIDIA Motherboards than on the Intel. Intel has bottlenecks not only on their FSB, there are more.
While all those benchmarks and charts look great at web sites, real tests must be made, and we do real tests.
How many time it takes to:
- Boot the OS with all the extras we put in.
- Copy files from and to the HDD.
- Move files from and to the HDD.
^ Important test nobody does.
- Delete files from the HDD.
- Load the applications we normally use.
- Backup files.
- Basic tasks like time it takes to load the Word, IE, to check Virus on HDD.
One that seams its missing is Hard Drive defragment, but the best defrag that I know is format and copy.
From all those tests in some, NVIDIA chipsets outrun Intel about 2X faster.
Are you sure? My more than a year old p5w DH deluxe works nicely with 1333MHz FSB (official BIOS for that) and I'm quite sure if Intel will ever produce 1.6GHz FSB desktop chipsets I can run those CPUs on the motherboard aswell, mine works nicely at FSB 1.85GHz.
That was only possible because of this:
Intel chipset supplies to be tight until next year
Intel is having limited chipset production, by allowing the "old" chipset to work with the new processor allows them to meet market demands for their chipsets otherwise they would lose sales for ATI, NVIDIA, VIA and SIS.
I think if this wasn’t happening you would have to buy new motherboards for the 1333Mhz FSB processors.
Ho Ho -
The problem in discussing with you is that you know no focus, and thus your logic jumps here and there and is fundamentally flawed.
"So without any benchmarks to prove it you assume there are no problems with AMD?"
The issue is never any benchmark. I was comparing X2 4800+ and E6400, and X2 4200+ with any Pentium-D under 3.2Ghz. The AMD counterparts perform better and are cheaper, period. If you want benchmarks to backup, there are plenty, and it is your problem who can't see, not mine.
"If it is so good then why is AMD not selling those enough to earn profits? "
Isn't it obvious enough that Richard, as head of AMD's marketing team, is responsible of the poor marketing performance? Or do you expect I to tell him what/how to do better?
Again, you are trying to use such logic that, first, because AMD products perform worse, thus they are priced cheaper, then, because AMD processors are cheaper, thus it must mean they perform less. Man, this is official, you have no logic at all!
The fact is taken from all 3rd parties benchmarks and pricing, AMD's Athlon64 X2 4800+ is a much more cost-effective purchase than Intel's E6400; similarly, Athlon64 X2 5600+ is a much more cost-effective purchase than Intel's E6550, and Athlon64 X2 6000+ is a much more cost-effective purchase than Intel's E6600 & E6700. Try to spin your way to deny these simple facts.
Ho Ho, you reminded me of the infamous "proof" used by the Roman church about the existence of the God:
God exists, because the Bible says so.
Bible must be speaking of the truth, because it's the book of God.
Even a "loyal" catholic like my mother do not agree with such a logic. And she's my mother who doesn't make a living through technology. You ought to be ashamed of yourself!
"Are you sure? My more than a year old p5w DH deluxe works nicely with 1333MHz FSB ... mine works nicely at FSB 1.85GHz."
I see you're lucky to get a MB that overclocks well. Let us know if your lovely motherboard supports E6550, or maybe Q6550 when it is released?
Aguia
I think if this wasn’t happening you would have to buy new motherboards for the 1333Mhz FSB processors.
No, a simple BIOS update is all it takes for many of the older C2D boards to support the 1333 FSB Conroes / Kentsfields. Not only that, many of them even support Wolfdale / Yorkfield. I'm very happy that my P5B Deluxe/WiFi purchased in September last year is on that list, as I will now almost certainly buy a Yorkfield in six months or so for a nice performance gain over my current E6600. Ironically, the Yorkfield will probably cost me less than I paid for that E6600 but with a simple drop-in upgrade I get:
- Twice the cores
- 5% to 10% IPC gain per core
- SSE4 capability
- All within roughly the same power envelope.
Now that's progress.
abinstein
AMD's Athlon64 X2 4800+ is a much more cost-effective purchase than Intel's E6400
E6400 is no longer a valid data point, as it has effectively been discontinued and is priced artificially high.
Athlon64 X2 5600+ is a much more cost-effective purchase than Intel's E6550, and Athlon64 X2 6000+ is a much more cost-effective purchase than Intel's E6600 & E6700
For general purpose computing this may be true, but for SSE and in many games the Conroes still have the advantage in those comparisons you listed. Anyway, you must know by now that AMD's pricing strategy is widely attributed to Hector Ruiz's quest to "break the monopoly" and not to Henri Richard. There's nothing that Richard could have done, Intel's marketing advantage is too great. Between Fab 30 and 36, AMD had too much capacity to supply the demand and couldn't afford to underutilize the fabs, so the chips had to sell. Due to Intel's far greater marketing budget, AMD had to undercut Intel's pricing to avoid piling up millions of CPUs in warehouses.
Abinstein
No wonder Richard is forced to resign.
I wrote this on another blog but it's also relevant here so I'll copy it:
I don't believe that Henri was fired. Not having an immediate replacement implies that Henri left largely of his own volition. Hector would be nuts to assume both jobs: he knows it's impossible to effectively evangelize Barcelona to clients around the world while simultaneosly running the company. I'm sure he'll scrape someone up in the next couple weeks and put him/her through a crash course in Henri's role and meet all of Henri's contacts, but not having an immediate replacement is a strong indication that this wasn't Hector's doing.
No, the more likely story is that Henri simply saw the writing on the wall, which is that K10 will not have enough IPC gain over K8 to compete with Penryn and raise ASPs to where they need to be to turn the ship around. He cashed out before the stock begins to really tank.
"I don't believe that Henri was fired."
Nobody says he is fired. He was undoubtedly under pressure, and possibly under fire.
"No, the more likely story is that Henri simply saw the writing on the wall, which is that K10 will not have enough IPC gain over K8 to compete with Penryn..."
Blah blah blah... our Axel is making the same old FUDs. Funny that everyone that I know who's actually seen Barcelona at work sees its superior IPC over Yorkfield, but you say Henri can't? Maybe it's just you can't?
Other than mpeg4 encoding, Phenom X4 performs better than Penryn Q6xxx at even 15% slower clock, period. That means a 2.5GHz Phenom will be comparable to 3.0Ghz Penryn.
abisten wrote
Blah blah blah... our Axel is making the same old FUDs. Funny that everyone that I know who's actually seen Barcelona at work sees its superior IPC over Yorkfield, but you say Henri can't? Maybe it's just you can't?
OK. Since you talk about everyone that you know who's seen Barcelona at work sees its superior IPC over Yorfield...implies you have hard data. Let's have it, please post it here, and please don't post regurgitated AMD CRAP posted on YouTube or some moronic Analyst Day PPTs.
Axel -
"OK. Since you talk about everyone that you know who's seen Barcelona at work sees its superior IPC over Yorfield..."
Lets just say that you are FUDing when you say otherwise. Everyone can always wait for the hard data when Phenom is released. Intel is trembling. You aren't. I'm not sure who's smarter.
"No, remember that all of the 1066 FSB Conroes have been rendered obsolete by the 1333 FSB pricing. Newegg prices:
E6420 (2133/1066/4MB) - $186
E6550 (2333/1333/4MB) - $179"
This is again nonsense. My Dell machine running the E6600 can't accept E6550 nor E6750, period. Are you saying my 9-month-old Core 2 Duo is not obsolete? Congratulations! You are agreeing with what I said 6 months ago!
"Let's look at the 6000+: $170 on Newegg. The E6750 ($205) is significantly faster than the 6000+ in general computing and much faster in SSE."
This is again pure nonsense. E6750 is faster in media applications than X2 6000+, but equal or slower in general computing. Or maybe you don't even know what general computing is, and thought it is only media encoding?
Oh, I wrongly attributed gutterat's rumbling to Axel... my bad. But what I said stands true. Those who tested both Phenom and Penryn do confirm to me that Phenom has better IPC and draws less power. AMD's new chipset helps there too.
The fact is clear, whether spoken to gutterat or anyone, that Axel spoken nothing but FUD.
Abinstein said...
Intel is trembling
I suppose that your close personal friend Paul Otellini confided this bit of information to you? Or are you the one spreading FUD now?
I see you're lucky to get a MB that overclocks well. Let us know if your lovely motherboard supports E6550, or maybe Q6550 when it is released?
My homebuilt system, with an ASUS P5B Delxuse motherboard, was first built with an E6600. A BIOS revision allowed for quad core CPUs and 1333mhz FSB CPUs. The latest update adds full compatability for 45nm Penryn processors. So now I can (and intend to) install a new Yorkfield quad core CPU either late this year or early next year. Obviously Nehlaem will require new chipsets, sockets etc. but when changing from the FSB system to the new CSI/QuickPath it's rather unavoidable.
If your going to yammer on about a motheroard being slightly more expensive for an Intel system, it's worth noting that an AMD system needs more expenisve ram (DDR2-800) for maximum performance while a Core 2 based system gets on fine with cheaper memory (DDR2-533 for 1066FSB, DDR2-667 for 1333FSB). You can also get motherboards with both DDR and DDR2 and both AGP/PCI-E for Core 2 (Like the ASrock Dual-VSTA) that allow for the use of older components, with a later upgrade to DDR2/PCI-E. You won't find that kind of flexbility in an AMD system motherboard.
As a former Socket 939 system owner I'll say that AMD is far from perfect as well. AMD billed Socket 939 as the platform of the feature, with plenty of upgradability in it. Instead, in Q2'06 AMD releases Socket AM2, and kills of Socket 939. Just like that. So when someone like me is faced with buying 1) a new motherboard, 2) new memory and 3) a new CPU it's no suprise I bought a Core 2 Duo. That E6600 is still faster than any X2 CPU today. (After I upgraded to a Q6600, I built another cheap system to use that E6600)
This is again nonsense. My Dell machine running the E6600 can't accept E6550 nor E6750, period.
Then you need a BIOS update from Dell. It's not Intel's responsibility if the manufacturers don't provide the neccessary BIOS update.
If a company makes a motherboard with two Socket 1270F sockets for Opteron but never releases a BIOS update for Barcelona compatability, does that make it AMD's fault that this motherboard cannot support Barcelona?
Those who tested both Phenom and Penryn do confirm to me that Phenom has better IPC and draws less power. AMD's new chipset helps there too.
Either provide proof or it's not true. Lets see what I can come up with, shall we?
I have a reliable source at Intel that says that Nehalem has indeed taped out and is running at 4Ghz already. In general terms, IPC is improved by 95% over a Core 2 Duo CPU.
Obviously, that message above is total BS. But you've done the same thing and expect everyone to believe you? Pfft.
abinstein said...
Oh, I wrongly attributed gutterat's rumbling to Axel... my bad. But what I said stands true. Those who tested both Phenom and Penryn do confirm to me that Phenom has better IPC and draws less power. AMD's new chipset helps there too.
The fact is clear, whether spoken to gutterat or anyone, that Axel spoken nothing but FUD.
I am calling your bluff. Let's see the data. Phrases like What I said stands true. and Those who tested both Phenom and Penryn do confirm to me that Phenom has better IPC and draws less power. is merely propaganda.
Those who tested...right. Let's see the hard data, which you don't have the gonads to provide, eh?
abinstein, i got tired of reading your biased opinions about c2d vs A64 x2 performance. I mean, seriously, grow up and realize that there is absolutely no reason to buy AMD today, unless u want the absolutely cheaper product.
It is not Henri's fault that AMD cpus cannot compete. I am wondering when Hector is going to "pay" for the mistakes he has made the last year...
Abinstein is making up stories, claiming to know people that are testing AMD's next generation parts and that sort of thing, yet he refuses to provide even a modicum of evidence to back this up.
We know what Penryn can do; Intel has shown that on average is provides a 10% IPC advantage.
Barcelona is simply too little too late when compared with Penryn.
AMD has 1.5bn in cash left, they have to keep taking on more debt and they're so desperate for money that they are selling off scrap parts from FAB30. Losing $600m a quarter and will run out of cash soon.
When they are out of cash, BK will quickly ensues. Once they file for bankruptcy, it's easy to see what Intel will do.
Look at this:
http://contracts.corporate.findlaw.com/agreements/amd/intel.license.2001.01.01.html
Go to part 6, section 6.2 Termination for Cause.
It lists that the license will be terminated if:
(1) the filing by the other party of a petition in
bankruptcy or insolvency;
(2) any adjudication that the other party is bankrupt or
insolvent;
(3) the filing by the other party of any petition or answer
seeking reorganization, readjustment or arrangement of
its business under any law relating to bankruptcy or
insolvency;
(4) the appointment of a receiver for all or substantially
all of the property of the other party;
(5) the making by the other party of any assignment for the
benefit of creditors;
(6) the institution of any proceedings for the liquidation
or winding up of the other party's business or for the
termination of its corporate charter;
(7) the other party undergoes a Change of Control. For
purposes of this Section 6.2(b)(7), "Change of Control"
shall mean a transaction or a series of related
transactions in which (i) one or more related parties
who did not previously own at least a fifty percent
(50%) interest in a party to this Agreement obtain at
least a fifty percent (50%) interest in such party, and,
in the reasonable business judgment of the other party
to this Agreement, such change in ownership will have a
material effect on the other party's business, or (ii) a
party acquires, by merger, acquisition of assets or
otherwise, all or any portion of another legal entity
such that either the assets or market value of such
party after the close of such transaction are greater
than one and one third (1 1/3) of the assets or market
value of such party prior to such transaction.
That will pretty much kill the transfer of the x86 license with Intel. According to the contract, the license does not transfer with a sale, acquisition, merger, or bankruptcy of AMD. It just gets revoked.
AMD loses it's x86 license. Even if they are acquired by IBM, Samsung etc. the same thing can and will happen. Intel will drive AMD into the ground with low prices. If AMD tries to find a buyer, Intel pulls the license and AMD as a CPU manufacturer is finished.
There is absolutely no reason to buy Intel today, unless u want the absolutely slower and more expensive product.
There is absolutely no reason to buy AMD today, unless u want the absolutely slower and more expensive product. 250W 4x4 anyone? Q6600 is comparable in performance costs less than half and uses less than half power. 95W vs. 250W.
AMD is finished. BK in Q2'08.
No, a simple BIOS update is all it takes for many of the older C2D boards to support the 1333 FSB Conroes / Kentsfields.
axel,
Yes it's just a bios update, but again just because of this:
Intel chipset supplies to be tight until next year
Show me a link where Intel in the past increased their FSB speed for their older chipsets?
If this isn’t authorized by Intel and was deliberately done by the mobo makers, again show me a link where they in the past increased the FSB of their motherboard to support new CPUs?
There is absolutely no reason to buy AMD today, unless u want the absolutely slower and more expensive product.
Here is a reason from you:
Obviously Nehlaem will require new chipsets, sockets etc.
There is no upgrade path for Intel after Penryn. Who is thinking in buying a new system or will buy one current X2 or will wait for Phenom. Both are upgradeable.
You can also get motherboards with both DDR and DDR2 and both AGP/PCI-E for Core 2 (Like the ASrock Dual-VSTA)
You won't find that kind of flexbility in an AMD system motherboard.
Really? The AMD version of that board already existed so long time ago that some have already forgot:
Dual-VSTA
Dual-SATA2
And that isn’t from Intel it’s from VIA and the other from Uli. That’s why it’s a shame that Uli, VIA and SIS are all going down.
So when someone like me is faced with buying 1) a new motherboard, 2) new memory and 3) a new CPU it's no suprise I bought a Core 2 Duo.
That’s why if some one is going to buy a new system today will go for AM2/AM2+ because of Phenom and the upgrade path. As soon as some one buys a new system based on Penryn will have no upgrade path.
I see in 2H/2007 more AMD systems sold, and in the 1H/2007 even more AMD system sold.
Again, because of this: Obviously Nehlaem will require new chipsets, sockets etc.
Wow! You can buy an AMD system and upgrade to quad core in the future! Who would have thought that? Too bad I did that months ago on my Intel system.
Only suckers buy crappy AMD dual core crap to upggrade to quad core later. Buy intel and get quad core now for just $266.
AMD is finished no cash or anything like that left. BK in Q2'08.
Really? The AMD version of that board already existed so long time ago that some have already forgot:
Can I put DDR ram in a Socket AM2 CPU on that board? Will that work? Nope.
AN AMD CASH ANALYSIS
At the end of Q2'07: 1.5bn in the bank. They gain 1.5bn from another loan. Total, 3bn. But they have to pay Morgan Stanley $2bn for the loan to be payed off. Thus, AMD has $2bn left. But then by the end of the quarter, they lose $600m more.
By the end of Q3'07 AMD will have only $400m in cash, plus whatever they can sell off in their garage sale.
giant,
You have reading problems?!
The upgrade capacity has nothing to do with the number of cores.
So what you are saying is it its impossible for AMD to release one new faster quad core in two years that works in current AM2 motherboards?
For (you) Intel, Penryn is end of the road for sure.
That motherboard support DDR and DDR2.
You might think that, but AMD has a history of screwing customers over. Socket 754 in 2003. Screw that, we'll go to 939 in 2004! Less than two years later they ditch that for socket AM2. AMD's 45nm CPUs will likely be slower than their 65nm CPUs. We've seen this from AMD before, Brisbane core 65nm is slower than Windsor 90nm at the same frequency with the same amount of cache.
My Q6600 at 3Ghz (same speed as QX6850 stock) is more than twice as fast in video encoding than AMD's fastest CPU. AMD quad core stuck at 2ghz will not help that at all. Even 2Ghz CPUs are very scarce for the OEMs. AMD stuffed the channel and screwed over it's channel partners earlier in the year, and now less and less companies will do business with AMD.
You will be lucky to see 2.5Ghz Barcelona from AMD before they BK in Q2'08. In fact, it might even be sooner than that. The only way for AMD to avoid BK is to sell off it's fabs. During the price war Intel is laughing all the way to the bank with $1.3bn profit while AMD LOSES MORE THAN $7M PER DAY.
AMD is a pathetic whinging company. They can't compete with Intel and they will BK and be crushed like ants because of it. Good riddance.
AMD BK Q2'08.
Aguia said...
well you cant put any Am3 cpus in Am2/+ sockers so whats the difference to Nehalem with Socket B/H ?
Good riddance.
AMD BK Q2'08.
The problem with democracy is that it has a tendency to allow the foolish majority control the fate of all...
You don't have a clue how horribly AMD's BK would affect you or I, do you? Do you even understand what the words competition, monopoly, price-war, etc mean? Have you even completed a single economics class, or at least read a book or two about the subject on your own time?
Sometimes I get really depressed about the future, because I realize that the majority, people like you, are the ones who, in aggregate, control my fate.
Aguia
If this isn’t authorized by Intel and was deliberately done by the mobo makers again show me a link where they in the past increased the FSB of their motherboard to support new CPUs?
I'm not sure what you're getting at. Of course it's the mobo makers that are supporting the faster FSB, not Intel. Intel's P965 Express chipset does not officially sanction 1333 FSB operation. You won't find an Intel-branded P965 motherboard that will overclock to 1333 FSB.
But the reason many mobo makers can guarantee 1333 FSB on P965 motherboards is because Intel's P965 chipset runs perfectly fine at 1333 FSB. It's obvious that the P965 was designed from the beginning to run fine at 1333 FSB, though Intel themselves will not say so to you or I.
This has nothing to with tight chipset supplies. The mobo makers are simply taking advantage of the capabilities of the chipset, and would have updated the BIOS for this regardless of how tight the P35/X38 chipset supply would be. It just makes sense because:
- It needs no official Intel sanction (there hasn't been any public announcement by Intel that P965 supports 1333 FSB).
- It keeps customers happy with an upgrade path.
- Competition between the mobo makers would have ensured it. All it takes is one to do it and the others would have followed suit.
From AMD: Advanced Micro Devices Inc. is talking about the Luther Forest Technology Campus as potentially becoming home to "the world's most advanced microprocessing plant."
More hot air. That's nothing. The fab that will exist fifty years from today, but does not exist now, will be the most advanced fab in the world!
AMD knows that D1D is the most advanced fab in the world. Running with 45nm high-k process. What does AMD have? Oh yeah. 'FAB36'. A fab with a 65nm process that can't clock a dual core CPU above 2.6Ghz.
E6550 vs. 5600+
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3038&p=7
AMD fragged once again.
abinstein
"Let me tell you this simply: Either provide proof or stop FUDing."
Yes, please do that.
aguia
"Show me a link where Intel in the past increased their FSB speed for their older chipsets?"
Show me in the past when AMD could use more than single type of RAM. You can't? Does this make it invalid to say it can't in future?
"Who is thinking in buying a new system or will buy one current X2 or will wait for Phenom. Both are upgradeable."
For how long? My 975 will work for me from P4 to Penryn, not all that bad I'd say, comparable to s939. How long could you live with dualcores on s939? A whole year?
spam
"You don't have a clue how horribly AMD's BK would affect you or I, do you?"
I don't know about him but I certainly do. That's the main reason why I'm dissapointed in how AMD performs.
abinstein
"I knew all you can refer to is AnandTech and others on-line sites who got more money from Intel than from anyone"
Do you have any better?
Axel,
They have to do it otherwise they would lose huge sales to NVIDIA for example that already had 1333 Mhz FSB officially even sooner than Intel.
Ho ho,
Show me in the past when AMD could use more than single type of RAM. You can't? Does this make it invalid to say it can't in future?
AMD never had been in the chipset business, you should know that, like Scientia already said in the past AMD was forced to develop 750 and 760 chipset. But that’s it.
I could provide you lots of links with chipsets from VIA and SIS that supported two types of memory. Remember the famous K7S5A? No? The problem is that when guys like you that only buy Intel and only know Intel, think that the others can’t do it too.
For how long? My 975 will work for me from P4 to Penryn, not all that bad I'd say, comparable to s939. How long could you live with dualcores on s939? A whole year?
Well I don’t know what is 975?!
Regarding the s939 in my country there are s939 dual cores for 80€, it’s a great upgrade if you really want/need a dual core processor. Besides virtualization what is that processor lacking?
aguia
"Well I don’t know what is 975?!"
Intel chipset that came out with COre2.
"Regarding the s939 in my country there are s939 dual cores for 80€"
Good for you. Cheapes x2 we have in Estonia is some x2 4600 left into some warehouse with a pricetag of well over €200.
"Besides virtualization what is that processor lacking?"
Upgradeablity and performance for one thing. DDR1 costs way more than DDR2, at least in Estonia the price difference is around 2x. Just imagine how many people bought s939 just for the x2 and had it outdated soon after with no future what so ever.
13ringinheat said...
Abinstein has no clue what barcelona can do and for that matter no one does since there are no benchmarks out there except AMD's powerpoints no leaks nothing......
The mere fact that the henri resigned soo close to AMD's most crucial launch goes to show that even he has no hope for barcelona yet this might be to big of a dose of reality
However i will give abinstein this AMD processors are all low end yard sale garbage and if you want a decent rig Intel is the only company that will do......
I like how scientia will jump on an intel sympathizer if they say something wrong or stray from the topic at hand but all of abinstein's posts yet have been pointless BS and no word from scientia yet...... beacon of unbiasedness indeed........
please abinstein show us a benchmark where the 6000+ is better in general computing than the E6750 because i have yet to see it better in any application that is used by the main stream public
Ringinheat, I took the time to edit and repost your comment because your criticism is worth noting. Abinstein has indeed been instigating most of the recent flaming. I apologize for letting the comments get out of hand. It wasn't because of bias or favoritism on my part but simply that I've been occupied working on my next article.
Every post with direct insults like fanboi, fanboy, idiot, and nut as well as other language has been deleted.
I guess I also have to repeat that when someone says something insulting DO NOT quote it in your reply. It doesn't do any good to delete the original if everything is quoted somewhere else. Therefore, quotes of insults have been deleted as well. Nor do I feel like going through around 25 posts and editing them for repost.
I have copies of everything that has been posted including everything deleted by the original author. If you posted something you want back up again make a request and I'll hunt it up and repost.
However, most of this discussion has been a waste of time. It isn't important why an executive left AMD, this happens all the time. I would be a lot more concerned if Dirk Meyer or someone in development were leaving.
Also, there is no doubt that C2D is faster on desktop benchmarks both in total and per clock.
I've mentioned using PGI before to have a fair test but even with PGI, C2D will still be faster than K8.
Opteron is doing better in server benchmarks but that is the only area that I know of where AMD's cpu's are competitive. On the desktop, AMD only holds a competitive position by having a lower price.
In terms of money, AMD's position is likely to improve in Q3 and Q4 due to cost reductions from 65nm and other factors (like maybe ATI being profitable again). This is helpful in terms of the threat of bankruptcy but no change in benchmark position. Desktop versions of K10 are not going to be out in real volume until Q1 08 while Intel hits its own desktop volume with Penryn in Q2 08.
Other than mpeg4 encoding, Phenom X4 performs better than Penryn Q6xxx at even 15% slower clock, period. That means a 2.5GHz Phenom will be comparable to 3.0Ghz Penryn.
Look who is talking FUD. Abinstein, you never cease to amaze me.
abinstein
"Other than mpeg4 encoding, Phenom X4 performs better than Penryn Q6xxx at even 15% slower clock, period. That means a 2.5GHz Phenom will be comparable to 3.0Ghz Penryn."
Where are you getting this from? I've only seen comments from Fuad and Sood. I'm a bit more inclined to believe Sood.
And for the record, if you were to benchmark Phenom at 3GHz you would see that it kicks the living cr_p out of any current AMD or Intel processor — it is a stone cold killer (at 3GHz, now imagine how it would perform if they could squeek some more juice out of it?).
I’m guessing that AMD will be able to launch some parts at higher clocks than it is currently showing in its roadmaps, and if the company can get these chips on shelves in a timely fashion, I think it could be a major coup and could even be the impetus for the turnaround the company so desperately needs.
Okay, so this would mean that a 3.0Ghz Phenom is faster than a 2.93Ghz Kentsfield. If what he is saying is true then it is possible that Phenom at 3.0Ghz is as fast as Penryn at 3.33Ghz.
ho ho,
Intel chipset that came out with COre2.
That chipset already existed before Core 2 Duo. And I'm not sure if every motherboard that uses that chipset will work with Core 2 (Conroe).
Also don’t forget that the ones that don’t bought one (very expensive) 975 motherboard didn’t got the same luck even if they bought one socket 775 motherboard, it doest run Core 2 Duo CPUs. Almost all nvidia 5xx line and all Intel <965 line don’t work with those CPUs.
Take Asus for example:
Intel 975X
Not all work with Core 2 and some strange specs there too, on the same motherboard lists this:
Intel® Quad-core CPU Ready
Compatible with all FSB except Quad Core
It is or it isn’t compatible?!
Maybe giant could explain that, since he says only AMD socket 939 have no upgrade path.
Good for you. Cheapes x2 we have in Estonia is some x2 4600 left into some warehouse with a pricetag of well over €200.
There are also lots of Opteron for s939 starting at 90€. Great deals too.
Very good cheap upgrades. Regarding the memory even if more expensive (1GB DDR 45€, 512MB 24€) it is worth the premium since performance difference is not that big if any, and the price of migration to other system, Intel 775 or AMD AM2 is much higher. For example just one Core 2 Duo CPU is higher than one socket 939 X2/Opteron + 1GB DDR, 150€ total. Core 2 Duo 150€ + motherboard 130€ + 2GB RAM DDR2 80€ (360€).
giant
" Socket 754 in 2003. Screw that, we'll go to 939 in 2004!"
939 came out in late 2003 and was a direct response to Intel's 800Mhz FSB. Remember that 754 matched the 2.2Ghz Barton in bandwidth. It's difficult to claim that socket 754 was bad because it is still used as socket S (DDR2) for mobile.
"Less than two years later they ditch that for socket AM2."
Actually, that would be 3 years later. Further, 939 chips continued to be made until Q1 07. AMD disposed of these chips in Q2 with a 2% hit on the gross margin.
"AMD's 45nm CPUs will likely be slower than their 65nm CPUs."
If you think this then you know nothing about cpu architecture.
"Brisbane core 65nm is slower than Windsor 90nm at the same frequency with the same amount of cache."
Brisbane is the same core with a different cache controller. K10 is far different.
"AMD stuffed the channel"
Do you have any source besides roborat for this?
" and screwed over it's channel partners earlier in the year,"
Wrong quarter. The shortages ocurred in Q4 06.
"and now less and less companies will do business with AMD. "
Wrong again. AMD recovered its share in Q2.
Well that Raul Sood wrote a superb article, it seamed that it was written by me :)
In other words I completely agree with him, in fact in one of my post here regarding the slow Intel SATA/IDE bus we at my company don’t use 3dmarks, sisoft or any crap that gives us random numbers for any thing that we don’t use a cant use and can’t really evaluate!
We test all the OS and applications on the machines and use our own metrics to evaluate the machine performance, and for that we don’t need some stupid software developed by some company that as nothing better to do than invent some test that for them it measures “real” CPU/HDD/GPU/USB/… performance, yeah right…
Scientia
If what he is saying is true then it is possible that Phenom at 3.0Ghz is as fast as Penryn at 3.33Ghz.
He certainly does hint that he's personally benchmarked K10 and he further hints that it's much faster than Kentsfield per clock. If true, we're in for an exciting 2008. But then again, Sood has always been somewhat excitable and quick to evangelize new technology after seeing a couple impressive benchmarks. He's much like Kyle Bennett of HardOCP in this regard, lacking some practical grounding. Sood's outlandish forecasts for 4x4 are painful to read today:
"Real world benefits for gamers & enthusiasts",
"Consider, for instance, the World of Warcraft player who has multiple characters under different accounts. With a properly configured [4x4] system it’s possible to load two instances of the game at once and trade items back and forth, or skin with one while killing with the other."
LOL.
My guess regarding Sood's most recent espousal of K10 is that he saw a few handpicked memory bandwidth limited benchmarks and drew some rather optimistic conclusions about overall performance.
Even Hector plays down Barcelona's competitiveness with Penryn, only stating that K10 will "regain position in the server market" and that "Intel and AMD are going to be very competitive in every space going forward."
Regarding Henri, high level executives don't make sudden exits on the eve of a product launch if they expect the product to be astounding. No, they cash out after the stock has rocketed post-launch. Evidently Richard doesn't expect Barcelona to floor the market with its performance and he doesn't expect to see the share price rocketing up anytime soon. That either means Barcelona is underwhelming, or that it's competitive but he doesn't think the price war will ease up anytime soon, leaving AMD's ASPs in the gutter and its share price with it.
axel
No, I think what Sood said is accurate but you have to consider what it means. I wouldn't personally get that excited unless K10 was, say, 30% faster at the same clock. However, as you say, Sood is more easily impressed so I'm thinking suppose K10 is 15% faster than Kentsfield at the same clock. That would be enough for 3.0Ghz Phenom to match 3.33Ghz Penryn.
Oh, as far as the money part goes; that is not related to K10's performance. That is related to Intel's advantage during 2008. I mention some of this in my next article.
Intel-favorers,
The facts are straight. AMD processors are smarter buys. Compare the two below:
1. AMD Athlon64 X2 4800+: $108
2. Intel Core 2 Duo E4300: $121
AMD runs faster and is cheaper.
Now compare the two below again:
1. AMD Athlon64 X2 6000+: $169.99
2. Intel Core 2 Duo E6550: $178.99
AMD runs faster and is cheaper.
It seems to me that AMD processors are priced so low because of two reasons: 1) Intel is essentially a marketing company and it pays to get favorable reviews unconditionally, 2) there are many "unwise" buyers.
Scientia, do whatever you want with the comments, it's your blog. Apparently you're more interested in keeping people coming rather than keeping the truth. It's nothing but shame.
"Abinstein has no clue what barcelona can do and for that matter no one does since there are no benchmarks out there except AMD's powerpoints no leaks nothing......"
The point is not how much I knew. I only tell you what I heard from someone who actually seen both Penryn and Phenom running. The point is those Phenom doubters, who claim the chip doesn't offer better IPC, have offered not but FUDs.
"However i will give abinstein this AMD processors are all low end yard sale garbage and if you want a decent rig Intel is the only company that will do......"
Take a look at my comment above and you'll see from another (more correct) point of view, what you said is equivalent to this -
"AMD processors are all low priced bargain buys and if you want an overpriced rig Intel is the only company that will do"
I'm sorry if this could hurt any Intel lover, Intel reseller, Intel worker or buyer's heart. But this is the simple fact. Go to newegg and compare the prices. Simple as that.
giant -
"Abinstein is making up stories, claiming to know people that are testing AMD's next generation parts and that sort of thing, yet he refuses to provide even a modicum of evidence to back this up."
No, I am not making up stories, but those who actually tested both Phenom and Penryn told me such "stories," if you like. The story is short and simple: Phenom has better IPC than Penryn.
gdp77 -
"abinstein, i got tired of reading your biased opinions about c2d vs A64 x2 performance."
Just search this phrase "X2 4800+: $108" on this comment area. You will realize that you got tired too easily. The fact, again, is stright and simple: A64 x2 are smarter buys than c2d.
poke -
"Look who is talking FUD. Abinstein, you never cease to amaze me."
No, this is no FUD. As I said, I was merely transcripting what I heard. Maybe you will be amazed by Phenom when it is released. Just wait and see.
Abinstein, I heard from a friend of a friend of a friend that he has an old Pentium Pro that will crush a 3Ghz Phenom.
Of course the above statement is absolute rubbish, but it makes my point. All you can offer is 2nd or 3rd hand information. And you expect us to accept it as fact.
Not so long ago, I offered my professional opinion on a subject within my area of expertise based on first hand experience. You chose to question it, which is fine. That is your right. Just don't be surprised when others don't show any greater inclination take your 2nd or 3rd hand accounts at face value.
"Of course the above statement is absolute rubbish, but it makes my point. All you can offer is 2nd or 3rd hand information. And you expect us to accept it as fact."
Whenever I say some 2nd hand information, I specify it as so. However, do you or do you not accept that buying Athlon64 X2 is simply the less expensive and better performing choice? Again, search for the phrase "X2 4800+: $180" on this page and read the paragraphs around it.
"Not so long ago, I offered my professional opinion on a subject within my area of expertise based on first hand experience."
Sorry, I must've missed it.
However, do you or do you not accept that buying Athlon64 X2 is simply the less expensive and better performing choice?
Well, considering I just found an E6600 at the link below. At $196, I'm not sure the Athlon64 X2 is a better deal at $180.
http://slickdeals.net/?permadeal=10125
That aside, it depends on what you want to do.
For current games with no futureproofing then the Athlon is a better deal. However, with the current trend in games going to raytracing and enhanced physics, I would say the C2D chip is the better way to go if you don't want to upgrade in a couple of years.
If you do a lot of media encoding (and just because you don't doesn't mean that no one else does) then go with C2D.
For word processing, web browsing, email, etc. a $180 chip is overkill anyway so go for the cheapest deal you can find.
Having said all of that, I won't buy AMD until they put a fab in the states. I have no vested interest in supporting the EU economy without getting a reasonable portion of the investment going back into an economy that benefits me. Call it enlightened self interest.
"Having said all of that, I won't buy AMD until they put a fab in the states. I have no vested interest in supporting the EU economy without getting a reasonable portion of the investment going back into an economy that benefits me."
I can see this clearly, and I'm sure a lot of enthusiastic posters here have similar thoughts like yours.
If you are an Intel employee or share holder, then such rationale is justified. However, if you are just an Intel customer, you should know that it is the very existence of AMD that has made the Intel processors better and cheaper every year.
The best way to contribute to the economy is to buy the most cost-effective product.
AMD 6000+ $169
http://www.microcenter.com/single_product_results.phtml?product_id=0258266
Core 2 E6600 $129
http://www.microcenter.com/single_product_results.phtml?product_id=0253108
$40 less for a faster product.
Since you don't like Anandtech, we'll use Xbitlabs this time:
http://xbitlabs.com/images/cpu/dualcore-roundup/charts/average.png
Those results also highlight a key point, the E4400 is faster than the X2 4800+, a fact that you claimed was untrue earlier on.
the full set of benchmarks not just the average graph can be seen too
http://xbitlabs.com/articles/cpu/display/dualcore-roundup_8.html#sect0
"Since you don't like Anandtech, we'll use Xbitlabs this time:"
That page is as good as nothing. You don't know what the application is, you don't know what the peripherals are, you don't even know how the programs are compiled or tested. I guarantee you there are Intel money behind. Is this your level of performance measurement?
abistein
You look very silly now.
I have VERY GOOD INFORMATION on Barcelona and Wolfadale system performance but will NOT go out of my way to show the results.
Scientia has a prior post of mine, which he deleted, challenging you.
Your Wait until Barcelona...and Or Wait until Phenom are really tiresome and, frankly, empty.
Systems are making their way into the hands of people who will let a few curious types slip in and take some teaser numbers and gently let them slide out onto the web.
There will be NO WAY to conceal Barcelona numbers before the official launch, just like Wolfdale and Yorkfield numbers are making their way out now.
But we didn't really need the numbers anyway because they were in front of us _all along_
"the full set of benchmarks not just the average graph can be seen too "
Just compare the "full set of benchmarks" with what AnandTech tested 9 months ago. Wow, so in 9 months E6400 actually grew faster! I didn't know Intel had implemented bio-like technology into its processors?
It's obvious that the xbit lab benchmark is more a joke than AnandTech. Basically it is hand-picking benchmarks to justify the ridiculous pricing disparity.
"You look very silly now.
I have VERY GOOD INFORMATION on Barcelona and Wolfadale system performance but will NOT go out of my way to show the results."
If you can't quote my words correctly, then please don't. You only make yourself silly.
"Your Wait until Barcelona...and Or Wait until Phenom are really tiresome and, frankly, empty."
It is an advice. It is not the advice giver who is empty, it is those who doesn't listen to advice but determined to FUD along that are.
As I said, and you and many others obviously can't read, I merely transcribe what I heard. I don't even ask you to believe me, just wait and see. But if you FUD along, then like it or not, I'll make sure your FUDs are marked as FUDs properly.
"There will be NO WAY to conceal Barcelona numbers before the official launch, just like Wolfdale and Yorkfield numbers are making their way out now."
So what's the poor Barcelona numbers? Do you have them?
"But we didn't really need the numbers anyway because they were in front of us _all along_"
Benchmark numbers, especially the enthusiast ones, are easily manipulated, usually by big money. Simple things like optimization can easily affect performance by 20% on the same architecture, not to mention the choice of benchmarks.
No one is going to believe these wild claims that Xbitlabs is being paid by Intel. They even have 'Powered by Opteron' logos on their site. Anandtech also recently published a dual core shootout between Opteron and Woodcrest Xeon 5100 in energy efficiency. The Opteron won and was give the praise it was due by Anandtech. In fact, this praise was quoted by AMD in their analyst day presentation.
Abinstein, you are a fool. You should have left this one alone and tried to dodge the facts like you always do. Now you have three options:
1. Bravely admit defeat like a man.
2. Silently admit defeat like a coward.
3. Continue to pointlessly argue like a clown when it is blazingly clear you have been defeated.
aguia
"And I'm not sure if every motherboard that uses that chipset will work with Core 2 (Conroe)."
That first motherboard on the list is what I have and I know it works with quadcores, even with the 1333MHz ones. The second one was kind of odd, I'm not sure what keeps it from supporting Core2.
"There are also lots of Opteron for s939 starting at 90€. Great deals too."
True, Opterons are a cheaper than X2's but still almost twice as expensive as their AM2 counterparts. For the price of 2.6GHz dualcore I could get brand new Q6600. Cheapest dualcore I could find for s939 was 1.8GHz for €100. For same money I could get 2.5GHz AM2, a whole 40% speedboost. Alternative would be e2160 with comparable performance for €20 less money.
"For example just one Core 2 Duo CPU is higher than one socket 939 X2/Opteron + 1GB DDR, 150€ total. Core 2 Duo 150€ + motherboard 130€ + 2GB RAM DDR2 80€ (360€)."
Yes, it is more expensive but it'll also be considerably faster than the s939 based thing. Also there are (much) cheaper motherboards and RAM availiable.
I can get decent MB with great OC capability for around €110-120 with 45nm quad and 1333MHz FSB support. If you don't want to OC subtract €20-40 from that. 2G 800MHz DDR2 costs around €65. Add to that €150 for E6550 or €245 for Q6600. Total cost is €325 with OC MB and around €285-305 without, add €95 for quadcore. Those are Estonian prices, I'm sure they would cost less elsewhere. In US subtract at least 18% from them. With that you've got massive speedboost compared to the older machine. Also the dualcore should have lower power usage. You could get some money back by selling your old stuff, there are people out there who don't want to upgrade all of their PC at once.
scientia
"However, as you say, Sood is more easily impressed so I'm thinking suppose K10 is 15% faster than Kentsfield at the same clock."
Heh, that's new for me, suggesting performance lead over how easily ipressed someone is :) Excuse me if I remain sceptical.
abinstein
"The point is not how much I knew. I only tell you what I heard from someone who actually seen both Penryn and Phenom running."
You mean the same guy who you quoted to have been running K10 since early this year and seeing some massive performance increases and clock speeds even though AMD couldn't clock them over 2GHz?
"No, I am not making up stories, but those who actually tested both Phenom and Penryn told me such "stories," if you like. "
Ok, you are not making up the stories, the people who tell you them are. If you claim otherwise why not introduce some of those people to us so we could hear the stories first-hand?
abistein rambled a bit, looking ever so silly as always.
I don't even ask you to believe me, just wait and see.
I have seen enough to know my own Barcelona vs XXXXXXXX reality.
You will eat your words. I will supply the condiments, free of charge.
Giant -
"No one is going to believe these wild claims that Xbitlabs is being paid by Intel. They even have 'Powered by Opteron' logos on their site."
I'm not sure who you called a fool. If above looks good enough for you to prove something, then the fool is obviously you sir.
And unfortunately I have to say that you don't know benchmarking. It's so apparent, that these sites don't even spend effort to synchronize themselves, that the benchmarks of Intel processors are "floating points". If this is not evident enough to you that the benchmarks are rigged then you have no observability.
"Anandtech also recently published a dual core shootout between Opteron and Woodcrest Xeon 5100 in energy efficiency. The Opteron won and was give the praise it was due by Anandtech."
Saying otherwise will only destroy AnandTech's own credibility because Opteron's energy efficiency has been proven objectively by more people than Intel could buy.
"In fact, this praise was quoted by AMD in their analyst day presentation."
Justified.
"You will eat your words. I will supply the condiments, free of charge."
At least I'm better than some Intel propellers who make FUDs here and there and (by definition) never take responsibility.
So what do you expect Barcelona IPC to be? I say, as measured by SPECint_rate and SPECfp_rate, Barcelona will be 10% to 30% better than Tigertown. For 2P and 4P, the lead will be even greater.
What I heard, however, is that Phenom X4 has better IPC than Yorkfield, but the latter has clockrate advantage. I have said this so many times, but to my amazement, even when some of you got tired of hearing it, you still don't get it.
Ho Ho -
"You mean the same guy who you quoted to have been running K10 since early this year and seeing some massive performance increases and clock speeds even though AMD couldn't clock them over 2GHz?"
I don't remember ever quoting such a guy saying such a thing. Clock speed and massive performance increase?
"Ok, you are not making up the stories, the people who tell you them are."
How do you know they are making up stories? FYI, they are the same people who showed me Conroe's SPEC_rate around last April (which proved Conroe has some 20% higher IPC than K8, way below some Intel "enthusiasts" 40% claim).
"If you claim otherwise why not introduce some of those people to us so we could hear the stories first-hand?"
You simply can't. Unless you speak Japanese. Well, I'm saying too much. But there are billions people speak Japanese, so pick your pick.
abistein wrote:
So what do you expect Barcelona IPC to be? I say, as measured by SPECint_rate and SPECfp_rate, Barcelona will be 10% to 30% better than Tigertown. For 2P and 4P, the lead will be even greater.
Sorry to inform you that you are showing your ignorance. IPC on its own is worthless. You have to also care about clock frequency.
You should also care about the version of SPEC 'rates' you want to compare against because one might favor AMD more than Intel.
You would also probably care about what compilers, flags, operating systems, bitness are used because these will have an impact on the underlying results.
You might also care about the platforms atop which the CPUs are hosted. The performance you are likely to achieve on a scale-up Japanese system with 8-16 sockets is probably not the same as what you would get on an IBM System X 8-16 socket system. Go figure.
SPEC int rate and FP rate, assuming the 2006 versions, are not really relevant: they are only relevant to AMD marketing. This further proves to me that you don't have any real performance numbers to talk about. We've been hearing so many data points about SPEC INT rate this and SPEC FP rate that but never anything about straight up versions of SPEC INT and SPEC FP. Why is that? (hint: rhetorical question)
Unfortunately for you: I can't tell you anything about the performance of Barcelona and XXXXXXX because I would be breaking several NDAs, and I want to continue getting access to these cutting edge systems, thank you very much.
Here's a hint: the blue team can sleep very well at night :)
BTW, there is no such thing as a 'Tigertown' CPU.
gutterrat -
It's funny to see you say I have to consider this and that when comparing Barcelona. Apparently people like you spend no effort of such when taking "Core 2 reviews" where essentially only one type of application is benchmarked.
"Sorry to inform you that you are showing your ignorance. IPC on its own is worthless. You have to also care about clock frequency."
You should learn to read before you criticize. I said "Phenom X4 has better IPC than Yorkfield, but the latter has clockrate advantage" (just search this sentence). Funny to see your statement "pre-fragged" by mine. Pity.
"Unfortunately for you: I can't tell you anything about the performance of Barcelona and XXXXXXX because I would be breaking several NDAs,"
First you talked about Wolfdale, then you talk about XXXX... Wolfdale is dual-core, and I'm quite sure dual-core K10 won't have as much advantage against dual-core Penryn. If the XXXXX... is Nehalem, then I don't doubt Barcelona have little advantage against it; if it's Harpertown/Tigertown, then it wins by higher frequency, not better IPC.
"Here's a hint: the blue team can sleep very well at night :)"
If Penryn tops at 3.33GHz while Phenom at merely 2.8GHz, then Intel should have no fear of losing the performance title. With Phenom at 3.0GHz, however, ...
"SPEC int rate and FP rate, assuming the 2006 versions, are not really relevant: they are only relevant to AMD marketing."
Funny you can say such things at the same time calling other ignorant.
SPEC rate measures performance where performance is most needed. Any SPEC benchmark is far more meaningful than pushing a game from 40fps to 100fps.
Sorry I didn't finish editing one paragraph in my previous comment:
If Penryn tops at 3.33GHz while Phenom at merely 2.6GHz, then Intel should have no fear of losing the performance title. With Phenom at 3.0GHz, however, only skewed benchmarks can save Penryn. :)
abinstein
"It's so apparent, that these sites don't even spend effort to synchronize themselves"
Can you list a few sites that one can count on when reading benchmarks?
"I say, as measured by SPECint_rate and SPECfp_rate, Barcelona will be 10% to 30% better than Tigertown"
I wonder what happened to "over 40% in FP".
"How do you know they are making up stories?"
Without proof their stories aren't worth much. Anyone can say anything and without proof all those stories have equal credibility.
"FYI, they are the same people who showed me Conroe's SPEC_rate around last April (which proved Conroe has some 20% higher IPC than K8, way below some Intel "enthusiasts" 40% claim)."
Why are you quoting those "enthusiasts" instead of Intel? Intel specifically said 40% increase over Netburst. Did they lie?
"Unless you speak Japanese."
I don't but I have friends who do.
" there are billions people speak Japanese"
You are off by around two orders of magnitude :)
"If Penryn tops at 3.33GHz while Phenom at merely 2.6GHz"
Do you have any ideas when will either of those CPUs be availiable? Same question for 2.8/3GHz Phenom.
Ho Ho -
"I wonder what happened to "over 40% in FP"."
There are claims that Core 2 Duo is 40% faster than K8. We know that such claim is only true for certain applications under certain way of benchmarking. In average, Core 2 Duo is just 20% faster than K8 at the same clock.
I hope you don't need me to guide you further and can readily see why there's difference between the 30% and 40%.
"Why are you quoting those "enthusiasts" instead of Intel?"
Just to show those Intel enthusiasts are plain wrong. So you do agree that they are wrong, don't you? I had hard evidence as early as last May that they were wrong, even though I wasn't able to disclose the numbers. Yet those Intel enthusiasts FUD along till even today.
The fact is, for general purpose computing, Core 2 Duo has about 20% IPC lead over K8.
"You are off by around two orders of magnitude :)"
Many people speak Japanese. A rough estimate is twice the Japanese population (many of them are Aisan), some are Americans, too. Japan has 0.3 billion people, and that makes the total Japanese-speaking population close to a billion.
If I was wrong by two orders of magnitude, then Japan would have less than 1/10th of its current population. Maybe if you plan to drop a few more atomic bombs on that island?
abistein rambled yet again...
Don't pussyfoot around the issue. I know how to read. I don't have to respond to anything anyone, let alone, you and the rest of the amdzone crew pumps.
Rev 10h may be a good part but it's unfortunate that AMD's speedpath and process issues, let alone SHOWSTOPPER BUGS, ended up costing it about a year.
A whole year wasted down the drain. Flush that turd, please.
Just remember: You will eat your words. I will provide the condiments free of charge.
K10 is going to have to be about 25% faster to keep up with Penryn. It is looking like that may indeed be the case.
Gutterat, how much increase in IPC do you think K10 will actually have?
Scientia asked,
Gutterat, how much increase in IPC do you think K10 will actually have?
Question is relative to K8, right?
15% on average with gcc, 20% with PGI IFF they fire on all cylinders on integer-style benchmarks.
GutterRat -
I have nothing else to say to you but this again:
You should learn to read before you criticize. I said "Phenom X4 has better IPC than Yorkfield, but the latter has clockrate advantage".
If you get confused by Yorkfield and Wolfdale, then you should use the condiment free of charge yourself fisrt.
abistein,
Please don't keep fanning the fire.
The only one that'll be burnt will be you.
It will all become fairly evident within the next 2-3 weeks.
Post a Comment