AMD: All Dressed Up, But No Place To Go
AMD's current position is frustrating at best. Although AMD seems to have gotten its ATI offerings in order, its K10 offerings lag expectations in almost every way.
AMD's previous 65nm ATI 29xx GPU offerings looked a bit dismal compared to nVidia's. The high power draw and low performance justified more than one lukewarm review. The best ATI product, HD 2900XT, was only close to the nVidia 8800GT (at higher power draw) but nowhere near the 8800GTX. AMD's response to this seemed more than a little strange. They said that they intended to compete against GTX with Crossfire. However, it seemed difficult to imagine anyone really wanting double the power draw and heat of 2900XT. It seems though that with the new 38xx series on the 55nn process, AMD has gotten the power draw under control. And, with the reduced die size they can probably sell them at the reduced 2900 price and still make money. And, it looks like 38xx will actually make Crossfire a viable competitor against GTX.
AMD's new 55nm based 7xx chipsets look like winners of the first order. And, AMD's Overdrive utility looks very good. It is amazing to think about tweaking the overclock on a system with a factory control panel. The GPU's, the chipsets, and the Overdrive utility dress up any new AMD system fit for a ball. The problem is that these systems require an equally good processor but, so far, AMD has failed to deliver.
But, it isn't just one problem with K10; it is many. From lagging volumes to a clock ceiling of just 2.3Ghz to slow NorthBridge and L3 cache speeds to no clear indication of improvement over K8. Not only is there no clear indication of when things might be fixed there is not even a clear indication of what the problem actually is. Low volumes would suggest poor yields and ramping problems. Low clocks could suggest either process problems or architecture problems. The lack of 2.4Ghz speeds was blamed on a bug that won't be fixed until the next revision. This is in contrast with both reviews that overclocked to 2.6Ghz with no trouble and AMD's own 3.0Ghz demo. Under normal circumstances AMD should be able to deliver a demo chip in volume about 6 months later. That AMD does not appear ready to deliver 3.0Ghz K10's anytime in Q1 suggests a big problem of some kind. The poor performance also contrasts with statements from different sources who suggested that K10 had great performance at higher speeds.
The reviews aren't much help either. OC Workbench shows performance for Phenom close to that of Kentsfield at the same clock with a few exceptions. For example the Multimedia - Int x8 iSSE3 score indicates either a compiler problem or an architecture problem. Phenom's score is only about 1/3rd what it should be after turning reasonably good scores in the other categories. The Cinnebench scores are odd since Phenom appears to speed up more than 4X as all four cores are used compared to 3.53X for Kentsfield. A really good speedup would be 3.8-3.9X while 4.0 should be impossible. Something funky is definitely going on to get 4.35X. Phenom also falls off quite a bit in the TMPGenc 4.0 Express test.
The Anandtech Phenom Review suggests that Intel might be having some small difficulty with clocks on Penryn. However, this difficulty would be so small compared to AMD's that it is hardly noticeable. It only means that 3.2Ghz desktop chips won't be out until Q1. Since this is also when AMD is releasing 2.4Ghz this could easily put Intel at a 33% faster clock. Other statements in this review suggest that Intel may have streamlined its internal organization considerably which would also be bad news for AMD. This review also mentions that K10's L3/NB can't clock higher than 2.0Ghz. Anandtech does however confirm how nice AMD's Overdrive Utility is. The Anandtech benchmarks show generally slower performance for Phenom at the same clock as Kentsfield with higher power draw for Phenom. So, the Anandtech scores don't really show us where a problem might be.
I was hoping that when the November 2007 Top 500 HPC list came out that I would get some new test scores that would settle the question about whether K10 was better than K8. I downloaded the latest Top 500 list as a spreadsheet. I was delighted to see a genuine Barcelona score. HPC systems typically have well tuned hardware, operating systems, and software so benchmarks from these should be accurate. I was hoping the HPC numbers would avoid any question of unfavorable hardware, OS, or compiler.
I began crunching numbers. I normalized the scores based on the number of processors and clock speed. I was happy to see very tight clustering at 2.0 for dual core Opteron. Then I ran the numbers for Barcelona and it showed 4.0. This was double the value for dual core. This seemed to be a big problem to me since with twice as many cores and double the SSE width it would seem that K10's top SSE speed should be twice per core or four times larger for Barcelona. In other words, I was expecting something around 8.0 but didn't see that. This would suggest a problem.
However, I then ran the Woodcrest and Clovertown numbers. Woodcrest clustered tightly at 4.0. This was not a surprise since it too has twice the SSE width of K8. Unfortunately, the Clovertown numbers continued to cluster at 4.0. This was a surprise since (with twice as many cores) I was expecting 8.0 for Clovertown as well. So, unfortunately, these scores are inconclusive. K10 is showing the same normalized score as Clovertown but neither is showing any increase in speed over Woodcest. The questions still remain unanswered.
The bottom line is that AMD is having problems. Whether these problems are due to process, design flaws, or unfixed bugs is hard to say. AMD is also under the gun on time. For example if Phenom is only hitting 2.4Ghz in Q1 then that leaves no time to start production of Shanghai in Q2 for a Q3 release. I suppose it is possible that AMD could fix a serious design flaw with the Shanghai release but it is by no means certain. What is certain though is that AMD will have to do much better to have any chance of increasing its share up to profitable levels in 2008.
122 comments:
Hi Scientia:
Well, seems that AMD is facing some serious problems. To deny it would just be delusional.
They really need to ramp up clock speeds and fix any possible bugs with the upcoming revisions. L3 cache SHOULD run at processor speed and not at northbridge speed. Also, I expect Shangai to reduce the cycles needed to acces L3 cache.
ATi's Radeon 3800 is catching up to nvidia's 8800GT (performance/watt wise) and I really expect R700 to blow out of the water any Nvidia offering.
Also, according to some news I've read a few days ago (can't remember source), AMD seems confident to fight Nehalem with Shangai.
what should we expect from Shangai (perfomance wise)?
Is it just a "dumb" die shrink or is it really an overhaul of the K10 made right?
2)Will Bulldozer use more than a 4-issue design
The alarm bells about AMD's 65nm process have been ringing for some time. The fact that Brisbane has not clocked above 2.7GHz 11 months since it's (very soft) launch at 2.6GHz ought to be indication enough.
The key here are process variation resulting in curret leakage resulting in an inability to hit thermal limits at desired clockspeeds. Throw in K10's comparitively big die where there is a lot more area for variation to result in excessive leakage and you have a recipe for big problems.
Here are some links for instructive reading. The first link in particular in that it indicates how totally different Intel's and AMD's approaches are to the problem of process variation. The two are not comparable.
http://www.metarainc.com/news/First_Albany_Capital.pdf
http://investorshub.advfn.com/boards/read_msg.asp?message_id=21353453
http://www.xcpus.com/forums/news/5216-eetimes-barcelona.html
Happy reading!
ATi's Radeon 3800 is catching up to nvidia's 8800GT (performance/watt wise) and I really expect R700 to blow out of the water any Nvidia offering.
The 3870 does quite well. It's not as fast as the 8800 GT, but it's cheaper. You can also use Crossfire on either AMD or Intel chipsets which is a bonus.
Nvidia's new high end is also scheduled for 2008. I'd expect R700 and G100 to have similar performance.
Also, according to some news I've read a few days ago (can't remember source), AMD seems confident to fight Nehalem with Shangai.
I've not seen any information on AMD 'being confident', but they have Shanghai vs. Nehalem in 2H08. Shanghai will have a larger 6MB L3 cache, and as well as further IPC improvements. I'd expect these to be similar to what Intel did with Penryn, providing ~5% boost to IPC. It will be key for AMD to ramp up frequency of the Shanghai product, they'll need at least 3.5GHz IMO.
Intel has discussed some of the Nehalem details, but there's still a lot we don't know.
Accurate analysis of the current state I would say, in summary 2 out of 3 ain't bad.
Interestingly it is the ATi derrived parts that actually look good now, the AMD originated part does not, wonder what the people who left from ATI's side or the organisation think of that?
In regards to Phenum, I take the TLB bug like I take their explanation for increased cache latencies on 65nm K8, with puzzlment. It does not make sense that it would appear at a certain speed only, it would just become more frequent. Also, if there is a BIOS workaround why no put out the faster chips like they are doing with the slower? I think it is an excuse only and they cannot do 2.4+ currently in enough numbers to support the market.
Another disappointment that may be connected is the L3 cache speed. If it is going to be stuck at the northbridge speed then as the cpu speed increases they are going to have a bigger delta between the two which will reduce scaling. This is, as pointed out by scientia, opposite to what was claimed on scaling and backed up with results at 3Ghz from various sites. Maybe this is why the 3GHz machine demo'd by AMD never had benches done in the past? Increasing the L3 cache to cpu speed might cause TDP issues or yield or both I guess.
The IPC is yet another disappointment, forgetting the silly 40% claim for a moment I was hoping that a general value of 10-15% above Core2 would be present that would go some way to making up for Intels clock speed advantage, however it seems more like a negative 10-15%. I would say that as soon as Core2 results came out last year AMD realised thay they had been caught by surprise and ould not be able to catch up on the IPC side and were hoping for pure speed to somehow be got with later spins. This hasn't happened yet and I think they are still waiting for a goldern spin to turn up at the 11th hour to save the day.
The main problem I see for AMD is that even if they get 65nm sorted out, Intel has Nehalem next. We know the process is going to deliver, so the only unknown is whether the architecture will. However, we do know that Intel is going to get rid of certain areas it currently has bottlenecks with, FSB and memory controller for instance. So it's hardly likely to come out only 1-5% better than Penryn, it is more likely to be higher, nearer to Core2 type jump.
Even if AMD 45nm is perfect will it be able to match that? Will it be perfect even, I have my doubts. AMD say it is ramping up in the first half of next year but is there any sightings of it out in the wild yet? A demo machine running windows? I may have missed this I have to admit. But if it is not being demo'd is time running out?
We have all benefitted from K8 being so strong in the past, even if you bought Intel, looks like a weak K10 is doing nobody any favours.
erlindo
Perhaps saying that the 7xx series chipsets will "blow nVidia out of the water" is a bit strong. The truth is that the ATI 6xx series chipsets have been the favored chipsets for AMD boards, ahead of nVidia's offerings. I would simply expect this state to continue with 7xx. Some of the current preference may simply be due to having a chipset from the same factory.
periander
There are some errors in the first article:
wafers are moved in a mini environment, or Foup (front opening unified pod). Thus, one of the largest sources of random variation in a process, people, has been minimized in Intel's fab.
AMD has increased the number of people on its factory floor
This is incorrect. The same Foup's are used in FAB 36. The reason for this error was that the piece was written in May 2006 after FAB 30 had expanded production by 50% (hence more people on the FAB 30 floor) but before FAB 36 was producing any real volume. The situation today is the opposite with FAB 36 ramping while FAB 30 is being dismantled.
The company now monitors thousands of engineering parameters and is monitoring the process as it occurs in order to eliminate variation.
AMD has similar tool sensors in place and does similar real time monitoring of the process.
The article is primarily about Intel and only mentions AMD in passing. It does not accurately describe AMD's current operations.
The second link discusses specs printed by AMD in February 2007. This information is nine or ten months out of date.
The third link is a discussion on xcpus forum. There isn't a lot of solid information here:
1.) AMD may or may not have better pMOS performance than Intel.
2.) AMD may or may not be having trouble with gate thickness and therefore balancing gate leakage versus clock.
3.) AMD may need high K to fix what may or may not be current problems. I think the earliest for high K would be Bulldozer.
Was there something in this discussion that I missed?
randy
Yes, I agree. At this point we have no idea how much Nehalem might rely on hyperthreading for example or even if its IPC is higher than Penryn's. Not enough information.
andy
Yes, that's what I was saying. Everything looks good except the processor. They really seem to have gotten the GPU problems fixed.
I've heard rumors about two different non-errata bugs that needed to be fixed. However, there still seems to be something else, some larger problem. The simple fact is that the Integer IPC on K10 should be at least 20% above K8 and the SSE3 performance should be nearly double. True, K10 could end being one clock speed slower at the same clock as Penryn (much as K7 was compared to P4) but that would still be a big improvement over K8. But, AMD would still need to get its clock speeds up.
randy allen wrote:
I've not seen any information on AMD 'being confident', but they have Shanghai vs. Nehalem in 2H08. Shanghai will have a larger 6MB L3 cache, and as well as further IPC improvements. I'd expect these to be similar to what Intel did with Penryn, providing ~5% boost to IPC. It will be key for AMD to ramp up frequency of the Shanghai product, they'll need at least 3.5GHz IMO.
OK. I've found the source on what I've said:
quote:
AMD plans to be ready with its 45 nm version of K10 in the second half of 2008 and it believes and hopes that this is enough to fight Nehalem.
Deneb 45nm will support the existing socket AM2+ and it will work in AMD 790FX and the rest of the 7x0 chipset family while you will have to chance the motherboard to make the Nehalem work
Deneb 45 nm quad core should come before Nehalem, as the latest roadmap we’ve seen indicates that Intel won’t launch its native quad core CPU until the very last Q4 2008.
We hope for AMD’s sake that Deneb will be out a bit sooner than that.
Source Here
...So I guess your 5% guesstimate is a bit off for them (AMD) to believe something like that.
AMD is not dressed up at all, rather it was caught with the pants down ...
pop
Your characterization is obviously incorrect. Getting caught with their pants down would imply that they were doing something that they weren't supposed to be doing. There is no indication however that AMD hasn't been working hard on both process technology and cpu design.
As far as I can tell, IBM is the current best process partner for AMD. Perhaps you can suggest a better one.
AMD is trying to improve 65nm and working on 45nm. Is there some other course of action that they could take?
Finally, if you think AMD is performing so badly with processor design then perhaps you could name a company besides Intel that is doing better. VIA perhaps?
The truth is that Intel had similar problems when AMD moved to copper interconnects a year sooner. AMD simply needs to do better.
I just saw a very interestign story on Fudzilla that I beleive sheds a VERY BRIGHT LIGHT on K10s "issues."
Basically, unlike Opteron which had one core with only disabling HT links, but K10 is THREE DIFFERENT CORES basically where there is 23xx\83xx whixh is HT2, 22xx\82xx(?), and 6xxx\7xxx\9xxx.
That means that if a Barcelona chip doesn't meet the standard it would then need the HT adjusted for HT1 and CAN'T be a Phenom\Kuma as Phenom uses HT3.
I can't believe that the OEMs picked now to grow some balls and demand a customized solution.
THEY SUCK.
Why don't they do this to Intel, I wonder?
Christian M. Howell *Edited*
The alarm bells about AMD's 65nm process have been ringing for some time. The fact that Brisbane has not clocked above 2.7GHz 11 months since it's (very soft) launch at 2.6GHz ought to be indication enough.
That's not actually true as AMD still has to sell 90nm chips at the higher speeds. If they are totally cherry-picking 5000+ BE, then perhaps but they all seem to run above 3GHz without exotic cooling (nearly every reviewer got above 3GHz).
AMDs big problem right now is that people suck.... up to Intel so Intel can ... force them to take what they get. OR ELSE.
Where is the FTC? I guess getting lobbied.
christian
Could you link to which Fudzilla article you are talking about? I'm not seeing it.
"That means that if a Barcelona chip doesn't meet the standard it would then need the HT adjusted for HT1 and CAN'T be a Phenom\Kuma as Phenom uses HT3."
I'm not sure what you are talking about here. HT 3.0 is fully backwards compatible with HT 2.0 and 1.0. An HT 3.0 capable processor should work on an HT 1.0 motherboard.
That's not actually true as AMD still has to sell 90nm chips at the higher speeds.
No one cares about high AMD bins as they compete with mid Intel bins. The thermal traits of AMD's 65nm is hardly different than their 90nm so the OEMs don't care how big the die is. The notion that AMD's 90nm has to be sold at 2 bins ahead of their 65nm to be attractive is funny.
If they are totally cherry-picking 5000+ BE, then perhaps but they all seem to run above 3GHz without exotic cooling (nearly every reviewer got above 3GHz).
I thought you gave no credence to overclocking. Whatever the case, the low volume 3.2GHz Black Edition is 90nm, while the also low volume 2.6GHz Black Edition is 65nm. And AMD's 90nm overclocks better.
I'm not sure what you are talking about here. HT 3.0 is fully backwards compatible with HT 2.0 and 1.0. An HT 3.0 capable processor should work on an HT 1.0 motherboard.
Yes, they are compatible but according to The Inq, OEMs didn't want to re-qualify systems to run with a chip with the full 4 16 bit ungangable links in an nForce 3600.
here is the Fudzilla article
I thought you gave no credence to overclocking. Whatever the case, the low volume 3.2GHz Black Edition is 90nm, while the also low volume 2.6GHz Black Edition is 65nm. And AMD's 90nm overclocks better.
Excuse me for not knowing how you know what I think about OCing, but that has nothing to do with whether that may be a clue that Brisbane can be retail at 3GHz.
Low volume isn't usually $129 - $169. The 90nm process is at least a year older. It should be better.
Also, Brisbane is a shrink actually from 130nm (dual core was 90nm but the design was created at 130) so it has obviously reached the limit for the DESIGN not the process.
Sci,
The href tag didn't appear.
Barcelona 22xx
Ho Ho said ...
erlindo
"I really expect R700 to blow out of the water any Nvidia offering."
I'd be surprised if R700 was more than twice as fast in games as hd3870. Sure, multichip sounds very nice but when you start thinking about how it can be implemented you can see how much problems it really has. Just imagine how one could make a 4P Opteron board with 32bit HT3 links fit on GPU board and how much would it cost.
"Also, according to some news I've read a few days ago (can't remember source), AMD seems confident to fight Nehalem with Shangai."
They also said that Barcelona will be 40%+ faster than high-end Clovertown in various workloads and now it seems they are quite a bit behind in all but bandwidth limited scenarios. Now they claim to only be be competing so assuming history repeats itself we can predict that Nehalem will absolutely destroy it.
It's fun to use past experience to predict future events, isn't it?
"what should we expect from Shangai (perfomance wise)?"
I'd say not more than 5-10% IPC increase, bigger L3 could help a bit but probably not too much.
randy allen
"It will be key for AMD to ramp up frequency of the Shanghai product, they'll need at least 3.5GHz IMO"
Yes, they will need it. Good luck in getting there, though. I bet 45nm won't be released at higher than 2.5GHz, by that time 65nm is hopefully near 3GHz.
andyw35
"Also, if there is a BIOS workaround why no put out the faster chips like they are doing with the slower?"
Why do you think you can use BIOS tweaks to work around it? I have seen no signs of it being doable. After all the bug is near L3, BIOS can do little there.
"I think it is an excuse only and they cannot do 2.4+ currently in enough numbers to support the market."
That is exactly what I think.
"This is, as pointed out by scientia, opposite to what was claimed on scaling and backed up with results at 3Ghz from various sites."
Well, the whole K10 performance has been shown to differ quite a bit from what was claimed by AMD. I wouldn't take claims of random people too seriously (superlinear scaling, suddenly gets superfast at >2.4GHz, 3Ghz blows everything out of the water etc). Those claims are just redicilous.
"The IPC is yet another disappointment, forgetting the silly 40% claim for a moment I was hoping that a general value of 10-15% above Core2 would be present that would go some way to making up for Intels clock speed advantage, however it seems more like a negative 10-15%."
Actually there is not big surprise there. Sure, SSE got much better but integer and FPU ALUs are still quite similar to what is in K8 so I didn't expect it to be better than Core2. Guys much smarter than me said K10 will have lower integer IPC than Core2 and is a bit faster in floating point, just as K8 was. Now with Penryn there is even bigger IPC difference with integer and they are on par on floating point workloads.
Now I guess abinstein and/or Scientia start telling me to read the architecture manual. Well, I've read it and I still can't find the reason why K10 should be as good as you've claimed. Just make a list of the reasons it should be that much better than core2 and be done with it.
"I would say that as soon as Core2 results came out last year AMD realised thay they had been caught by surprise and ould not be able to catch up on the IPC side and were hoping for pure speed to somehow be got with later spins."
As even Scientia said Core2 probably wasn't a big surprise to AMD. Later mobile pentiums already had higher IPC than K8, Core2 was just an evolution of them.
"The main problem I see for AMD is that even if they get 65nm sorted out, Intel has Nehalem next. We know the process is going to deliver, so the only unknown is whether the architecture will."
One thing about the architecture is pretty sure, it cannot be worse than Core2. Now with lower memory latency thanks to IMC AMD has lost one of its main trumps. I'd say AMD has hard times coming.
"So it's hardly likely to come out only 1-5% better than Penryn, it is more likely to be higher, nearer to Core2 type jump."
Intel itself once said it'll actually be even better than Core2 kind of jump. I guess the return of SMT and IMC help quite a bit too. I would certainly like to see as big jump as was Netburst to Core2 but I doubt we would see too big differences in single threaded workloads.
scientia
"The simple fact is that the Integer IPC on K10 should be at least 20% above K8 and the SSE3 performance should be nearly double."
So, can you make a list of what exactly made integer IPC so much better? You've claimed it several times but never could prove it. "Go read a book" isn't really a proof, you know. Is it really that difficult for you? Once you've done it you at least know that I won't bother you with that question again.
"That's not actually true as AMD still has to sell 90nm chips at the higher speeds."
Remind me when was 65nm crossover for AMD? Wasn't it done in record time? Now I wonder how many parts are still on 90nm at the moment and how long will it take to entirely get rid of 90nm. Assuming their 65nm works fine there is no reason to continue making 90nm CPUs as they are more expensive and having three different products instead of two is certainly making logistics a bit harder.
"AMDs big problem right now is that people suck.... up to Intel so Intel can ... force them to take what they get. OR ELSE."
Erm, what? Was it you or Howell who said that? If you then please elaborate as you are not making any sense.
"If they are totally cherry-picking 5000+ BE, then perhaps but they all seem to run above 3GHz without exotic cooling (nearly every reviewer got above 3GHz)."
So the fact that CPUs run at much higher speeds with inbox coolers sais something about them? What about Penryns at 4GHz+ or Core2 at 3.5GHz+ with their inbox coolers? I thought that OC'ing didn't mean a thing and only released parts were that mattered. At least that was your POV when we talked how high clocked Core2 Intel could release if it wanted to. Double standards?
Does anyone know how well do other non-black 65nm K8 OC? Is it much different from the black edition? I also wonder if anyone has seen power measurements of 65nm K8 running at 3GHz+.
ho ho
It's fun to use past experience to predict future events, isn't it?
Curious notion. Should we use Prescott to predict Nehalem?
"Well, I've read it and I still can't find the reason why K10 should be as good as you've claimed."
They say that admitting a problem is the first step to recovery. So, perhaps there is hope. However, I think most people can understand the difference between a throughput of 1/2 (K8) and a throughput of 1 (K10). This is clearly stated in the manual.
"As even Scientia said Core2 probably wasn't a big surprise to AMD. Later mobile pentiums already had higher IPC than K8, Core2 was just an evolution of them."
No, you are misquoting me yet again. I never said that. I'm sure everyone expected the generation after Yonah to be faster but the amount of the speed increase could have been a surprise.
"One thing about the architecture is pretty sure, it cannot be worse than Core2."
Actually, we don't know that. Nehalem could have lower single threaded IPC than Penryn and rely on hyperthreading to increase IPC. Nehalem could be tuned more for multi-core operations. There are a number of variables and I doubt Nehalem's architecture is much like C2D's. I would guess that it is better but that is by no means a certainty.
" Now with lower memory latency thanks to IMC"
The problem is that Intel is switching to IMC with DDR3 which has higher latency than DDR2. This gives Intel the least bang for the buck.
"Intel itself once said it'll actually be even better than Core2 kind of jump."
So, what you've just said is that you assume that anything AMD says is a lie but anything Intel says is the truth? My guess is that reality is a bit more complicated.
"scientia"
You are getting my comments confused with Howell's. I edited his remarks because of derogatory language directed at Intel.
"So, can you make a list of what exactly made integer IPC so much better?"
I thought this was common knowledge:
1.) Sideband stack optimizer removes stack instructions from pipeline
2.) Wider fetch avoids broken instructions
3.) Increased L1 and L2 bandwidth
4.) Greater OoO operation
5.) Improved branch prediction
And, for memory limited operations:
1.) Memory controller prefetch
2.) Increased priority for loads
3.) Prefetch directly to L1 instead of L2
4.) Split controller
"Remind me when was 65nm crossover for AMD? Wasn't it done in record time?"
You are using the wrong word; it's "conversion" not "crossover". Since FAB 30 could not be upgraded to 65nm the term "crossover" is pointless as a metric. FAB 36 was fully converted to 65nm in Q1 which was about one quarter early.
"Now I wonder how many parts are still on 90nm at the moment and how long will it take to entirely get rid of 90nm."
Only the highest clocked K8 parts: 2.8 - 3.2Ghz. I'm certain that AMD has stopped labeling 2.8Ghz 90nm parts but I can't say about 3.0 and 3.2Ghz.
"Assuming their 65nm works fine there is no reason to continue making 90nm CPUs as they are more expensive and having three different products instead of two is certainly making logistics a bit harder."
I'm pretty sure that AMD had expected to be further ahead in phasing out K8. The 90nm Opterons should be phased out first with desktop chips following.
"Erm, what? Was it you or Howell who said that?"
That was Howell.
"I thought that OC'ing didn't mean a thing and only released parts were that mattered."
I wouldn't count anything that used a heatsink larger than stock or any kind of exotic cooling. I'll concede that what you can do with a stock heatsink should be a good indicator of headroom.
"At least that was your POV when we talked how high clocked Core2 Intel could release if it wanted to."
Again, not my comment.
"Does anyone know how well do other non-black 65nm K8 OC?"
From what I've heard, almost all of the 90nm K8's overclock well. That isn't a surprise for such a mature process. I have not however heard the same things about AMD's 65nm process.
christian
The Fudzilla article says nothing about using 3 different cores. Where did you get that from?
mo
"I can understand censorship of things that shouldn't be here like foul language, derogatory remarks, personal attacks.
But when the conversation is nearly on-topic, then I don't see a problem. "
I guess I'll have to agree.
christian
"Also, Brisbane is a shrink actually from 130nm (dual core was 90nm but the design was created at 130) so it has obviously reached the limit for the DESIGN not the process."
That's a legitimate point. K8 probably was designed to be converted to 90nm but not necessarily 65nm. It is possible that Brisbane could be having problems from the shrink. However since K10 is designed for 45nm I'm not sure you could make the same argument with it.
Scientia
...K10 is designed for 45nm...
You have said this before, but can you link to where AMD has made this statement?
Even tier-1 vendors like IBM can't get enough 1.9GHz (!!) parts. When I read about 2.5Ghz, or even 3Ghz, for Barcelona I can't help but think "I'll believe it when I see it shipping in volume!".
Oh, and Penryn, which launched only 2 weeks ago is available from HP now.
Only the highest clocked K8 parts: 2.8 - 3.2Ghz. I'm certain that AMD has stopped labeling 2.8Ghz 90nm parts but I can't say about 3.0 and 3.2Ghz.
Proof that AMD is operating a last generation fab to produce a trickle of high clocked parts?
a) Opteron
b) This has been an assumption that you have constantly perpetuated without proof
c) AMD had record shipments last quarter, and I'm sure that this was not because of 65nm and a trickle of 90nm Black Editions.
giant
"Intel has made certain claims, and has lived up to most of them."
I guess you must mean they have lately. We could start with Intel's withdrawal of it's 1.13Ghz PIII, Intel's multi-year delay of Itanium along with the exagerated claims, Intel's assertions by two top officials in 2003 that they were not working on 64 bit extensions (which had already been included in Prescott), Intel's claims of 5.0Ghz for Tejas and 7Ghz for Nehalem (both of which were canceled), Intel's claims for Whitefield (which was canceled), Intel's continuing delays on Itanium upgrades, and Intel's exagerations about FBDIMM. Yes, Intel is has been doing fairly well on the desktop (or at least better than AMD) in spite of scaling back C2D and having the 45nm schedule slip.
However, in terms of 4-way, I don't see it. Tulsa is now made obsolete after just 12 months by Tigerton which itself will be made obsolete 12 months later by Nehalem.
"AMD has failed to live up to most of their grandiose claims. The '40% faster than Clovertown' one is the most obvious to me."
Then you are blind. As I explained months ago, the 40% claim was true for K8 over Clovertown on SPEC and equally true of K10. Even today it is clear that K10 will take the lead in HPC, database and other areas at the same clock. These are server applications.
"Then AMD goes around talking about 'ramping 45nm in H1'08' and you're not surprised that people doubt these claims Scientia?"
45nm is not a known quantity. There are four possiblities that I can see:
1.) 45nm is on track but any slippage for 65nm pushes it back.
2.) 45nm is harder to implement so slippage is added to what is already there from 65nm
3.) 45nm stays on track timewise but begins at the same or lower clock than 65nm
4.) 45nm stays on track and includes improvements over 65nm
At this point I have no idea which is more likely. I guess we'll find out in Q3.
john
"Proof that AMD is operating a last generation fab to produce a trickle of high clocked parts?"
You need to get your timeline in order. The parts sold last quarter were produced in Q2 after which AMD began reducing FAB 30. The parts being made now would be sold in Q1. What I'm saying is that I doubt AMD is now producing K8's that will be labeled as 2.8Ghz parts for sale in Q1. So, today, I imagine that 90nm parts are down to a trickle. Secondly, you are not taking into account that FAB 36 has nearly doubled in capacity since the beginning of the year. At the beginning of the year, FAB 30 was still outproducing FAB 36.
Scientia
...Yes, Intel is has been doing fairly well on the desktop (or at least better than AMD) in spite of scaling back C2D and having the 45nm schedule slip.
Please explain the scaling back comment.
Also please explain how 45nm slipped, and what you used as a comparison.
Thanks
enumae
"Please explain the scaling back comment."
Being a year late with a chip faster than 3.0Ghz would seem to be scaling back.
"Also please explain how 45nm slipped, and what you used as a comparison."
That would be in comparison to Intel's schedule.
Again, in comparison to AMD with its K10 slippage and low clocks Intel's slips are not much of a factor.
Scientia
Being a year late with a chip faster than 3.0Ghz would seem to be scaling back.
I understand what you are saying, thanks.
In my opinion it shouldn't be considered late when you factor in that Intel has had no competition nor has there been a consequence for not having released them.
That would be in comparison to Intel's schedule.
I thought 65nm was late 2005, if not, when did Intel launch 65nm?
Randy allen
It will be key for AMD to ramp up frequency of the Shanghai product, they'll need at least 3.5GHz IMO
Well if AMD hit 3.2 GHz with 90nm, at normal evolution and scaling it would hit 3.4 GHz easily at 65nm but it seams they can’t. I think AMD surprise was how its design/process didn’t clock higher.
Ho ho
I'd be surprised if R700 was more than twice as fast in games as hd3870. Sure, multichip sounds very nice but when you start thinking about how it can be implemented you can see how much problems it really has.
Well rendering is one of the tasks than can easily be multithread we just don’t seen much advances there because there are even less multi GPU solutions and implementations than dual core CPU’s.
For example Ati and Nvidia could go the MCM route like Intel.
In fact I wonder how hard it is for Ati to pickup the R600 512bit board design and put two RV670 chips (MCM) giving each one its own 256 memory bus and you’re done with it.
If Intel can do MCM designs why can’t no one else do it, and don’t do it?
After all the bug is near L3, BIOS can do little there.
Bios solved lots of critical bugs in Core 2 Duo in case you forgot. Even Microsoft released patches.
Well, the whole K10 performance has been shown to differ quite a bit from what was claimed by AMD. I wouldn't take claims of random people too seriously (superlinear scaling, suddenly gets superfast at >2.4GHz, 3Ghz blows everything out of the water etc). Those claims are just redicilous.
Well lets assume the bugs and the slower L3 (that is also slowing the IMC in case nobody noticed) removed about 10%/15% performance overall, where would AMD be with that added 10%?
"AMD: All Dressed Up, But No Place To Go"
"Whether these problems are due to process, design flaws, or unfixed bugs is hard to say"
I'd say that AMD missed the party and can just pack it up this generation and hope they got the 2009 product and process fixed.
Nothing for AMD is more urgent then getting out a Barcelona and a Phenom that they can sell upwards of 750 bucks to compete with INTEL high end. It matters not that the bin split or volumes would be skinny, but if AMD even offered a paper launch of a handful of units it would give insight to what they are facing. The fact that they have NO offering at the high end says lots.
THis is far different then INTEL's debacle with Prescot a few years back. At that time the issue was that architecture simply ws broken in its ability to use a deep pipeline with highly inefficient IPC against a much more energy efficient wider architecture. Back in those Netbust days prescot could easily scale to 4GHz but the problem was power. Even with Cedar Mill and another couple years to try and optimize the architecture on an even better power efficient technology Cedar Mill never really made much more progress. INTEL during this Prescott disaster had far more resources and money and they worked hard and finally released C2D and fixed it all at 65nm but it still took them a few years to fix their big party miss
Barcelona is late by many months and slow, and Phenom is coming and also slow. It does look very reminescent of INTEL's Prescot days except for the fact AMD continues to teease about higher speeds coming. Now is this any different then the promises INTEL gave in 2003 for Prescot with noise about 4 GHz?
The only way AMD could be here with no high end is really both a process and a design issue. A simple few logic bugs our layout to process marginality should be easily debugged and fixed with one if not two full layer steppings and a couple metal follow-ups. The fact is that we are quarters into the Barcelona and Phenom stepping cycle with no material output smells of something much bigger. I think we can pretty much kiss 65nm K10 architecture as finished. To be 9 months late will miss the landing zone for the original design for Barcelona and Phenom and INTEL continues on its cadence. AMD now must insure that they capture all the fixes both design and process that were missed on the 65nm / K10 and incorporate it on 45nm. We already had eagerly awaited for AMD's response to C2D in 2007. 2007 is over and Barcelona has come and been a wimper. There is no reason to expect that it'll get better in 2008 with the same inferior process and architeture issues on the K10 architecture. Some very fundamental assumptions about design, architecture, and process were missed by AMD and IBM. Only they aren't talking, but hopefully its fixed for their 45nm cycle in 2009.
The only question now is if AMD can afford to pay to survive thru 2008 and pay the high capital expenditures for design, process R&D, and build out the multi-billion dollar factory.
This much is sure, Hector is gone in 2008.
Lex,
Nothing for AMD is more urgent then getting out a Barcelona and a Phenom that they can sell upwards of 750 bucks to compete with INTEL high end. It matters not that the bin split or volumes would be skinny, but if AMD even offered a paper launch of a handful of units it would give insight to what they are facing. The fact that they have NO offering at the high end says lots.
Well lex maybe you want AMD do what Intel did, releasing 1000$ CPUs that performed slower than the lowest end AMD part. Or you forgot that Intel EE (840EE/9xxEE) line cost the same of the AMD FX (FX60) but performed about 30% slower? And the even the cheapest (X2 3800+) AMD part performed faster than those Intel CPUs?
THis is far different then INTEL's debacle with Prescot a few years back.
Yeah, AMD doesn’t have the guts to rob people like Intel.
A 2.4Ghz unlocked phenom FX for 999$ would put many Intel guys talking good things about AMD.
Back in those Netbust days prescot could easily scale to 4GHz but the problem was power. Even with Cedar Mill and another couple years to try and optimize the architecture on an even better power efficient technology Cedar Mill never really made much more progress.
Well if Intel had problems and didn’t solve them like you said, why do you expect AMD to solve then all in a couple of months especially after merging with a company that is has large has them?
INTEL during this Prescott disaster had far more resources and money and they worked hard and finally released C2D and fixed it all at 65nm but it still took them a few years to fix their big party miss
And you call that fixing? The only thing they did was brought theirs mobile CPU into the desktop and server markets. Even with another design ready to take off took them years to do, AMD with only one design can’t do much don’t you think?
The only way AMD could be here with no high end is really both a process and a design issue. A simple few logic bugs our layout to process marginality should be easily debugged and fixed with one if not two full layer steppings and a couple metal follow-ups.
If that’s sooooo easy, why didn’t Intel do it with their huge resources you where talking about, in fact by your own logic the ultimate Intel would never put Presscott on sale before it fixed the design don’t you think?
Do I detect sour grapes.
This much is sure, if AMD could sell anything that people would buy for a thousand bucks, you'd be sure their marketing people would have lined up a product. Only look at that silly quad-core and our the black line to tell you all you need to know about AMD brand value with the consumer. Sure there are a few stupid lappers that will buy anything that AMD puts out even if it is garbage, but believe it or not brand matters and AMD brand is trash. Its no wonder their marketing guys left the sinking ship before the K9 launch.
INTELs sold its prescott space heaters for 1000 bucks because they could.
If you want to talk highway robbery look at AMD's pricing on the Opetron. Once that superior product came out from INTEL what did they do with their 1000 dollar price, droped it to what 200 bucks over 9 months. Any sucker that bought AMD was the one robbed by Hector and his robbers.
They spent that money on ATI, a crappy silicon 65nm technology, and a bunch of lawyers.
INTEL spend the money converting a good mobile design into a killer lower power efficient Opetron killer, building 3 45nm factories and brigning HighK Metal K Penrym on time. So tell me again who robbed the customers. INTEL who invests in a sound business plan or AMD who takes your money and hires lawyers to sue their opponets.
Its sucks to back a losing company with a losing CEO who only knows how to blame the other guy for his failings. That is robbery sucker!
@ lex,
I can't speak for you as a person, but your rant wasn't insightful in the slightest.
You're still playing the game of "pick the winning company," when instead you should be thinking about the consumer. I still maintain that anyone who doesn't realize the consumers need both AMD and Intel to be healthy are completely missing the point.
You reward innovation by buying the best, not crap from some money losing company that uses foreign workes and whines and sues because it doesn't make anythign innovative.
How is that good for the consumer, rewarding bad management?
Yup what an argument, its good for the consumer to keep crappy companies in business. Show me a case when that was the case. Its called free enterprise, best product and business wins. if it doesn't win its not good business
spam said...
You're still playing the game of "pick the winning company," when instead you should be thinking about the consumer. I still maintain that anyone who doesn't realize the consumers need both AMD and Intel to be healthy are completely missing the point.
While evaluating the market situation from the perspective of the consumer is certainly important, especially to the enthusiast community, it is important to remember that these companies are beholden to their shareholders first-and-foremost (of which I am) and customers second.
The issue of climbing prices and reduced competition are certainly significant to the marketplace, but if one company is "winning", that is ultimately their right. Just because you don't like it, doesn't make it any less true. While lex may be short-sighted in some of his ramblings, there is nevertheless some truth behind his diatribe.
Just read Scott Wassons piece here at the TechReport
http://www.techreport.com/discussions.x/13724
The reason i am interested in this article is, of course, that it casts doubts on the TLB issue being the reason for the 2.4GHz Phenom no show so backing up mine and other peoples theory this was a red herring. It still seems to be a speed issue that the 2.4 is not present.
Also of note, Scott seems to indicate that the tests they did didn't have the BIOS fix implemented and also the northbrdige was actually set too high. This means current tests are 10% too high for the northbridge and another 10% too high for the TLB BIOS/microcode fix not being implemnted. A current 9500 then is a lot lower than an Intel Q6600 even thogh they are about the same price.
Obviously the real launch will be with the B3 stepping chips but that now seems to be the end of Fenruary at least, or 3 months away. It has got to be hoped that the B3 revision allows the produiction of 9900 and lower versions in large quanities as well as fixing bugs.
LOL ;)
Farewell trolls.
Better to have the site purged from nasty parasites like you all.
Whereever these guys go and post they turn nice blogs into nasty discussions with their flamebaits, insults, name calling and the typical "X sucks, Y rules".
Andy:
Don't listen to them.
If you indeed want a thorough discussion about AMD's past, current and future offerings, this is the place.
Roborat's blog is worse than Sharikou because:
->Intel critters gather there to talk crap about AMD.
-> All threads become flamebaits.
-> You won't find a single piece of truth on whatever they say about AMD.
We don't want that here. That's why there are blogs like Roborat's and Sharikou's to do that.
Whenever I feel bored, I just go and read all the crap posted at Roborat's blog. It's comic relief. :D
Even, it's more probable to find reasonable facts and thorough discussion about intel in Scientias blog than at Roborat's fansite (it's more a fansite than ablog). Just visiting his frontpage tells you all about what to expect from it. :)
Long time lurker here... For all you belly aching whiners here I'm going to post my 2 cents to you chumps who are so obsessed over Scientia's censorship here, so here it is plain, simple and in black and white for you:
THIS IS HIS BLOG.
HE MAKES THE RULES AND GUIDELINES.
WHAT? YOU DON'T LIKE THEM?
FEEL FREE TO IMMEDIATELY CARRY YOUR POSTERIOR ELSEWHERE, THANKS IN ADVANCE.
Personally, the less of you rabid Intel cheerleaders I have to wade through your mouth-spewing to actually read relevant and useful information regarding AMD and the semi-conductor industry, the better.
Thanks again Scientia for all your analysis and thoughts, keep up the hard work.
Hear hear Lou! I am tired of wading throught the Intel fanboy crap as well. Good riddance!
"HE MAKES THE RULES AND GUIDELINES."
Problem is he has stated several times to be unbiased but he showed people his true face with this blog entry and censoring that followed.
Erlindo, you do know that your post would suite well as an entry into Sharikous blog, right? At least it contains as much truth as his ramblings.
The funny thing is, as soon as the "pro-intel" people left, this blog became a ghost town. No one else has anything intelligent to say...
Scientia,
Here is your chance. Apologize for your inappropriate censorship, take a vow not to do it ever again, and maybe the knowledgeable people will return...
LOL. If all the knowledgeable people have left, and you're still posting here, that kinda marks you as a know-nothing. Why should anyone listen to you?
Its actually very boring to post the truth to a forum of stupid cheerleaders who don't know technology and more fun to post to a board of stupid fanbois even if you get censored. It amazes me that the cheerleaders don't really understand why INTEL is winning or the fact that INTEL will win the war and even more amazing how the fanbois can't appreciate why AMD will lose the war regardless of their one trick poney and FUD they continue to spout
Tick Tock Tick Tock
AMD if it is dressed for any party its the fools party!
Barcelona has come and gone and has proven itself to have all sorts of problem
Next the AMD fanbois have put all their hopes on Phenom. The first of many reviews are in
Here are some good lines from Extreme Tech
"Final Thoughts: Phenom not so Phenomenal"
"not much AMD could have done at this stage to pump up performance of Phenom. The processor is what it is"
"Performance is another matter, and we may have to wait until the 45nm die shrink to see improvements. The real question is whether or not AMD has the luxury of time to get to 45nm"
Its no wonder that Phenom is being sold for a few hundred bucks. Leading edge bottom dweller from AMD, is it no wonder they are sucking loss, losing marketshare when their leading edge products perform like this.
Lets look at the some numbers
Sysmark 2004SE
INTEL Q9650 536 WINNER
INTEL Q6600 455
Phenom 9600 324
PCMark05
INTEL Q9650 9701 WINNER
INTEL Q6600 7773
Phenom 9600 6787
Video Encoding DiVX6.7
INTEL Q9650 97 WINNER
INTEL Q6600 117
Phenom 9600 145
3DMark06 F/sec
INTEL Q9650 2.4 WINNER
INTEL Q6600 1.86
Phenom 9600 1.67
3DMark06 Score
INTEL Q9650 10218 WINNER
INTEL Q6600 9787
Phenom 9600 9831
To sum it up again "Phenom not so Phenomenal"
Want the best most reliable and stable platform buy INTEL
Want the best value and stable platfrom buy INTEL
Want to support a company producing crappy products, lose money, using foreign labor to make everything and taking handouts from Arabs then BUY AMD!
Back to the boring ranting you guys hate. The fact of the matter to produce leading edge CPU you need a good design running on a fast and capable silicon technology. To ship a leading edge CPU you need not only great designers, architects and a competent desing team but a HUGE test and validation team behind it. You think a company losing billions puts the necessary exhaustive validation behind its new products or cuts corners as the best leave the company?
You think a company that is billions in debt and trying to integrate multiple design teams, exeuctives and architects with varying personalities might have lost focus on design and validation execution? You think this might be similar to the early 2000's when someone by the name of Barrett went off at INTEL and spent billions buying companies and took the ball off the core business.
The different is INTEL has 100K employees and billions in the bank to recover. INTEL also has a laser focus with huge resourse on the silicon side. AMD has none of that. So how can AMD win? DEsign and validaiton in total dissary, no resources or money to recover with huge pressure to put a quick fix. INTEL had the luxury of billions in the bank if not a tomato on the face for a couple years. AMD is cuting corners now in the one place it can't afford which is technology. THey are very late on 45nm development and knowing how technology short cuts show up, well be seing this in late 2008 and ealry 2009 as the 45nm wheels fall off the car. DOn't be fooled by AMD/IBM claiming that their 45nm will be out in 2008. Unlike 65nm where you could argue that SOI / DSL was close to bulk/Strain but if you look at 45nm INTEL has a far larger performance lead with HighK/Metal gate compared to AMD/IBMs likely SOI / DSL offering when it come in 2008. It won't even be close unless Nehalem looks more like Prescott vs. Core2. If AMD was to pursue a asset lite approach to manufacturing then AMD really is finished as for high volume leading edge CPUs you absolutely need the IDM approach. Foundrys simply can't compete on the core silicon nor provide enough capacity and ramp capability for the hundred million unit volume required to recover the investments and get the profits required. Don't like the rant but that is the fact of the business and AMD can't escape the fact of the business.
AMD is finished, Hector is gone in 2008, Dirk will be gone in 2009 and AMD will be spun out in 2010.
Don't like my ranting but the camel's back has been broken with the ATI purchase and the lost of focus on execution by Hector in 2005-2006 as they sat fat dumb and they WERE CAUGHT WITH THEIR PANT DOWN! No different then when INTEL was caught with Prescott.
lex,
Many flaws in your post, has always.
Like I said here before, the leather in this Industry is the small AMD not Intel, so it seems.
If AMD is doing bad execution or having dumb ideas, then why after AMD bought Ati and announced fusion Intel (the copycat like sharikou says) announced that is going to do it too?
Why Intel (the copycat like sharikou says) will bring nehelem with all the goodies the first Opteron has brought to us?
Wait you are going to say that it’s a coincidence or that Intel was going to do it with or without AMD or Intel already done before the transistor was invented or you are going to say that Intel version is the real thing the AMD version was just a beta experience.
ho ho,
Tell me what is your suggestion then?
Go to roborat blog, where you invited me to post there and as soon as I saw those logos it was pretty clear to me that it is an anti-AMD blog which is even worse than a pro Intel blog.
Besides some of the guys that post there seems to be more interested in the stock market than to discuss technology or like I said bash AMD or even insult Scientia, which is funny they don’t like to be insulted by him but like to insult him which is very politely correct don’t you think?
Lex, why can't you write a single sentence without trolling? I think Scientia has been patient with you and let you get away with your trolling a lot. You know Intel is ten times bigger than AMD and you talk about Intel winning Intel blah blah blah. Just post something interesting to read and you will be respected, even by the most rabid AMD-fan. Most AMD supporters know that Intel is leading in the enthusiastic platform, no question about that. SO you don´t have to school us about that. And most of us buy AMD cause of other values than just pure performance. Concider that.
They need some product to sell for 2008, since Phenom was shipped with serious bugs. It will be Q2'08 before AMD can ship large quantities of Phenom CPU with B3 stepping that has this bug fixed while not causing the performance slowdown.
Wow, you ARE stupid. Did you think AMD would keep Fab 30 producing 90nm chips forever?
If there were no problem with K10 then what would you say? That AMD is moving OFF of 90nm? That's what I would say.
Everyone admits that the engr samples were B3 stepping, but there were too many B2\BA chips in the oven when the problem was found.
By discontinuing 90nm, they can be 100% 300mm wafers, which by itself gives you 2X the chips. The shrink to 65nm gives you another 30%.
Since ArsTechnica is reporting that they are foregoing 65nm for Fab38 (which I said MONTHS AGO on this blog) and go to 45nm, then they won't have to share space between Shanghai and Barcelona and K8.
I had no doubt that AMD would have problems with K10 when they lost the amount of money they did. That's also why I thought they should have ended the "price war" sooner.
But they are still gaining share. If they were being underhanded they wouldn't have recalled Barcelona for VM usage.
They're doign what they have to do against a company that is 10x larger and tends to play "dirty."
The one and only thing capable of killing AMD is a serious crisis of world economy, such big trouble that not a single investor can be found.
Maybe we face such a crisis in the near future...
Intel has no discrete graphics, Nvidia has no CPU - maybe they will buy VIA - AMD has both and this will matter in the end.
EU has Intel on the agenda for bad business - we have to wait for the outcome...
Maybe Nvidia will buy AMD as soon as Intel delivers competetive discrete graphic solutions?
This would be a vision of a dark future!
Phenom fragged by itself!
http://techreport.com/articles.x/13741/1
Abinstein are you still here? What was that about Phenom having higher IPC than Yorkfield?! At this rate K10 will be luck to match the IPC of K8!
AMD BK by Q2'08.
Andy, if they fix the problem in the TLB without micro-code (as in, on the processor) wouldn't that not have a performance penalty similar to the microcode and in fact result in an increase in performance? It seems to me like some of TechReports analysis is off in this sense.
Also, I'm fairly certain the 9500 and the 9600 are spec'd as having a 2ghz HT speed. It sounds to me like their 1.8ghz bus was simply a motherboard issue or something similar.
I'd really like to see benchmarks of the chip with the HT clocks turned way up and the core clock turned up a little as well. Not really saying I'd expect anything spectacular, but just as an intellectual stimulant:)
hyc, of course core2dude can't count himself as knowledgeable. His blog is pretty effective proof of that. It's a pretty good read if you're looking for some comedy.
"HE MAKES THE RULES AND GUIDELINES."
Hoho blathered...
"Problem is he has stated several times to be unbiased but he showed people his true face with this blog entry and censoring that followed."
Oh please. Like you're not biased about Intel Hoho?!?! HAHAHAHAHAHA! Get real man, here's a shocking revelation for you since you haven't figured this out yet:
EVERYONE is biased.
To what degree is more the question at hand. At least Scientia keeps an open mind about Intel unlike someone like Sharikou who thinks Intel is nothing but the spawn of Satan in the semi-conductor industry. Granted my viewpoint of them is anything but rosy, as they are a corporation that has built itself into a massive, fierce and aggressive force in the microprocessor world.(thank IBM everyone) However, at least I am not calling for something as rediculous as Otellini to be "burned at the stake in a public execution" for "crimes against AMD". If anything, AMD will have "it's day in court", in the not so distant future, then they will get any restitution from Intel they rightly deserve from the US Supreme Court in the anti-trust case.
Now, whether you or anyone else likes it or not, Scientia keeps a fair view of Intel when they screw up or do well and the same goes for AMD. Now while he may root for AMD to do well(as do I and anyone with half a brain in keeping a competitive, progressing and driving x86 processor market, cuz a monopoly sucks) but you cannot say he doesn't give Intel praise when they actually have earned it.(and they have lately with the intro of the C2D)
That's about as fair a shake as you're gonna get from anybody, be he a Intel loyalist or AMD supporter. So if you can't deal with that, that's YOUR problem and I sure hate it for ya.
Again, if you don't like the way Scientia runs this blog, no one is forcing anyone to stay here, you know how you got here, you know where to leave. Don't let the door bite you in the @$$ on the way out bub.
If anything, AMD will have "it's day in court", in the not so distant future, then they will get any restitution from Intel they rightly deserve from the US Supreme Court in the anti-trust case.
AMD's lawsuit is so bogus it's funny. One moment they complaint that Intel prices CPUs too low, so AMD can't make profit. The next they're saying that Intel is a blight to consumers, just out to make a fortune in profit by charing huge prices for CPUs.
Or how about when AMD complains that Intel gives rebates and incentives to OEMs to do business with Intel? They freely admit to doing the same thing in their own legal document. (e.g. Giving Compaq a million free CPUs to do more business with AMD in favor of Intel)
Or (my favorite) when AMD complains of making only limited market share gains in their 'window of opportunity' from 2004 -> Mid 2006. They try to blame this on Intel's "illegal" tactics, when they sold all the CPUs they could make. How is Intel responsible if AMD only has one 200mm fab online?
In other news, AMD has been deceptive about their power ratings:-
http://messages.finance.yahoo.com/Business_%26_Finance/Investments/Stocks_%28A_to_Z%29/Stocks_A/threadview?bn=893%20&tid=1630612&mid=1630612
In the NEW document, version B, AMD has had to increase these drastically:
http://www.amd.com/us-en/assets/content_...
79W, 115W (!!) and 137W(!!)
All these are for Barcelona CPUS that you cannot buy, with a maximum speed of only 2.5Ghz.
By comparison, Intel has a 3Ghz, 80 watt CPU in the market today. In Q1'08, a 2.66Ghz 50 watt CPU is due.
Super-linear scaling is not impossible. For example, with Phenom cache size increases from 2.5MB to 4MB when number of utilized cores goes from 1 to 4.
I'm not saying this (cache size) is the cause of super-linear scaling, but the latter is really possible, and occasionally observed.
Giant spewed...
AMD's lawsuit is so bogus it's funny. One moment they complaint that Intel prices CPUs too low, so AMD can't make profit. The next they're saying that Intel is a blight to consumers, just out to make a fortune in profit by charing huge prices for CPUs."
That dog won't hunt chump. Keep saying it long enough and you'll believe your own lies I guess...
Tell your "facts" to the courts in South Korea, Japan and Europe. Here's the simple fact YOU CANNOT GRASP Giant:
Paying computer makers to push out your competitor is ANTI-TRUST. Please get that through your thick skull.
By comparison, Intel has a 3Ghz, 80 watt CPU in the market today. In Q1'08, a 2.66Ghz 50 watt CPU is due.
Too bad those are cherry picked and are not all the models.
http://techreport.com/discussions.x/13577
Yorkfield is bugged
...and delayed (lol) :D
Paying computer makers to push out your competitor is ANTI-TRUST. Please get that through your thick skull.
Learn to read:
http://www.intel.com/pressroom/archive/releases/20050901corpamdanswer.pdf
Too bad those are cherry picked and are not all the models.
So what? Power consumption isn't all that important on the desktop. In servers, when you're dealing with many, many CPUs in the data center it becomes much more important. It makes perfect sense for Intel to use the lower power CPUs in servers and notebooks where the power consumption truly matters.
Yorkfield is bugged
...and delayed (lol) :D
If this is true, it is serious. But I would treat this as a rumor for now, because it's only been posted on a single site, and has not been confirmed by Intel.
If Intel admits there is a problem, and there's stories all over tech sites like TechReport, Tomshardware and Ananadtech etc THEN I would take it as a fact.
If you assume for a moment that this is true, it's not as serious as AMD's TLB bug. Why? Because Intel has it's 65nm Clovertown and Kentsfield products to fall back on. All AMD has is K8 now. That's a disaster for them. AMD desperately needs to get a competitive parts to have any hope of returning to profitability. Even without it's 45nm products, Intel has a wide performance lead across servers, desktop and mobile.
It also doesn't mean the CPU would be 'officially' late. Intel has stated a Q1'08 introduction for the non 'Extreme' Yorkfields. They may arrive later than the rumors predicted, but still within Intel's Q1'08 timeframe.
Conflicting information:
http://www.digitimes.com/mobos/a20071206PD221.html
Intel will release its first desktop dual-core Celeron series, E1000, January 20, according to sources at motherboard makers.
...
On the same day, Intel will also launch three quad-core (Yorkfield) CPUs, the Q9300, Q9450 and Q9550 and another four dual-core (Wolfdale) CPUs, the E8190, E8200, E8400 and E8500 for desktops, the motherboard makers added.
Giant,
In case you forgot AMD first launched the CPU then talked about the bugs.
I can’t figure out how what you posted changes that?
And you posted:
January (according to sources at motherboard makers)
(the motherboard makers added)
The Celeron E1000 is still at 65nm. I also wonder if dualcore 1.6Ghz with so little cache is better than one single core Sempron 2.4Ghz with 256Kb/512Kb cache.
I also wonder if pushing new cmos processes are really cost effective, since all the Intel 45nm CPUs cost a lot more than the 65nm and the supposed cheap ones like the Celeron are still manufactured at 65nm and not 45nm. If the all point of new cmos process is to reduce cost, why the cheapest versions aren’t the first ones to use the new processes.
Also doesn’t the K10 TBL bug only affect the 2.4Ghz versions and higher?
enumae
Looking over Intel's original 65nm delivery in 2005 may show that you are correct. I was under the impression that Intel delivered a good volume in Q4 but with more checking it looks like this was only a token delivery. So, a token delivery of 65nm in Q4 2005 with real delivery early in 2006 would mean that 45nm would indeed be on track with only a token volume in Q4.
lex
"THis is far different then INTEL's debacle with Prescot a few years back."
I think the most similar time for Intel would be late 2001. Intel was behind with copper interconnects, PIII wasn't keeping up with K7 and P4 Williamette had serious design problems. Intel didn't get back into the race until it released P4 Northwood on 130nm.
K7 is clearly not keeping pace with C2D, Barcelona seems to be having problems, and it looks like AMD is behind with 65nm. I think this is the best comparison, not Prescott.
lex
"You reward innovation by buying the best"
Now you know this isn't really true. I don't drive a Rolls Royce or BMW; I drive what I can afford. The vast majority of people do not buy top end products. However, I do not believe that 2.3Ghz is covering enough of the range. If AMD's top quad core was 2.8Ghz this might be true but there is no doubt that AMD is leaving to much of the mainstream market open by not having 2.6 and 2.8Ghz quads.
andyw35
I have to agree with you. It doesn't sound like AMD is being honest about what problems Barcelona may or may not have. And, I have to say that 1.8Ghz for the memory controller and L3 is a big disappointment to me as well. If AMD is truly calling this a feature instead of a patch then you could argue that Penryn's current slower FSB speed is also a feature. A little more honest information from AMD would be nice.
trolls
I've now seen people claim several times that I've gone into hiding when K10 looked bad.
Can you really not read? The article for this thread is fairly critical of AMD's current processor offerings. There are a few benchmarks at Anandtech that show some potential but not enough to write a new article. I don't always post here or at AMDZone everyday.
core2dude
"Apologize for your inappropriate censorship, take a vow not to do it ever again, and maybe the knowledgeable people will return..."
You are a little behind with your comments. See my reply to Mo a number of comments back. Mo pointed out that I was being a bit too heavy handed; I agreed and reposted ho ho's comments. I don't claim to be perfect but if I make a mistake I can try to fix it. It is a sad indication of Mo's character (or perhaps lack of) that he began calling for a boycott after I agreed with him. I can only conclude that he truly does want a very onesided pro-Intel discussion. There are indeed good posts and some knowledgable people on Roborat's blog however for every one of these posts you will read two from pro-Intel trolls. Roborat apparently wants it that way and I suppose that is his choice. If someone here feels more at home in that environment then that is where you should be.
Would I make a promise to never censor or delete again? Absolutely if everyone who posts here will also make a promise to avoid trolling, personal attacks, and improper language. Since that is not going to happen, I don't see how I can. I will say again that the comments here will not go the way of Sharikou's no matter how shrill the protests about censorship are.
You can get an idea of who the trolls are because everytime they get in a huff they always compare me to Sharikou. However, he is currently grasping at straws predicting a huge comeback for AMD and bankruptcy for Intel in 2009. This is a joke without merit and everyone here knows it.
My current view is wondering whether AMD will be able to gain any additional share in 2009. That is a good question because without more share I don't see how AMD can reach profitability.
lex
"Want to support a company producing crappy products, lose money, using foreign labor to make everything and taking handouts from Arabs then BUY AMD!"
This is trolling. It is high on insults and accusations but low on facts. Sure, AMD uses "foreign labor" but so does Intel. And, AMD did take a handout from Germany but the money from "Arabs" was a routine investment just as they did with several other companies like Citigroup.
"AMD is finished, Hector is gone in 2008, Dirk will be gone in 2009 and AMD will be spun out in 2010."
More trolling.
"Don't like my ranting but the camel's back has been broken with the ATI purchase"
You obviously are not understanding the facts. The ATI products are currently doing much better than the processor products. By any normal measure this would make the ATI purchase a good idea for AMD, not a bad one.
"and the lost of focus on execution by Hector in 2005-2006 as they sat fat dumb and they WERE CAUGHT WITH THEIR PANT DOWN!"
What loss of focus are you referring to? The 65nm delay was unavoidable since it had to wait on FAB 36. I think it is clear that AMD would have liked K10 to be ready earlier but it is obvious that it was not possible. I don't see a loss of focus; K10 seems to be very much what AMD needs if they can get it to work. And, as I've already stated, there simply is no better process partner than IBM.
"No different then when INTEL was caught with Prescott."
Not true at all. You seem to forget that Prescott moved to 90nm ahead of AMD rather than trailing as AMD is doing now with 65nm and 45nm. This is one of those areas where you completely twist up the facts. Prescott on 90nm was worse than K8 on 130nm. However, what is typically being compared right now is Penryn on 45nm with K10 on 65nm. That is backwards of Prescott.
giant
Concerning the lawsuit against Intel. AMD's biggest complaint is not the 2004-2006 period as you stated but the period from early 2002 when AMD lost about 70% of its market share after hitting over 23% in late 2001. AMD alleges that Intel had agreements that excluded AMD products. This would clearly be illegal. However, when I ran the numbers and considered legitimate factors like Intel's quicker move to 130nm I only came up with a figure of about $150 Million in damages. I suppose you could argue that $150 Million would have gotten FAB 36 built a little sooner and therefore put AMD in a better position in 2006. So, compounded damages might hit half a billion but I don't see this as a multi-billion claim as some do.
I get annoyed when people either misquote me (which they do at least half the time on Roborat's blog) or they repeat what I've already said and then pretend they are disagreeing with me (which happens here frequently with Intel enthusiast posters).
I have no idea what the actual problem or problems might be with K10. There are several possible things including design, bugs, and process but nothing I've been able to nail down. Now, I'm curious which of these things the pro-Intel crowd disagree with:
1.) If the problem is process then AMD needs to move to 45nm or perhaps even 45nm high-K as fast as humanly possible.
2.) If the problem is bugs then AMD needs to get them fixed. Preferably by Q2 at the latest.
3.) If the problem is a small design flaw then AMD had better be hammering out a repair for Shanghai. Bulldozer is too long to wait for a proper design.
4.) It is possible that AMD is aware of and keeping quiet about additional problems. If this is true then I don't expect to see anything until AMD comes up with a fix (because AMD would not want to sabotage sales any further). If AMD is not able to fix the problems then we may never know what they were.
5.) AMD really needs 2.8Ghz now and at least 2.9Ghz in Q1. Less than this brings up the question of whether AMD's share will start sliding again and that would bring up the question of profitability.
AMD: Survival and the Future
Yes, everything in Dresden is proceeding as planned.
If this is true then let's hope the plans include a big improvement soon.
Fab 36 is producing 65-nm parts, while the pilot batches of 45-nm parts are also right on schedule.
Again, 65nm doesn't look on schedule so I wonder about 45nm.
"We're currently ramping down 200-mm production in Fab 30. The first 300-mm tools have already been installed in the new Fab 38, which will begin production alongside Fab 36 in the course of the next year. Fab 38 will become a self-contained production site in 2009.
This statement does however completely refute the posters here who have claimed repeatedly that AMD will shutter or sell FAB 30.
I would feel more confident if the above statements had included a remark about 45nm being a lot better than 65nm.
I see I was brought into the discussion.
Sci, I was sorry to see that it took some criticism for you to post back what Ho Ho had said. That's what needs to be stopped.
I remember where I said that If the post is on topic, it should not be censored and you agreed... and I still saw posts being censored and thats when people started complained on robo's blog.
I'm glad that you've eased off a bit we can have meaningful ON TOPIC discussions.
Now you know this isn't really true. I don't drive a Rolls Royce or BMW; I drive what I can afford.
What if the BMW cost the same as the Honda?
The cost of 9600 Phenom vs. the cost of Q6600.
on a side note: I will admit that since the launch of K10 and the blunder that it was, you have realized that something is indeed not up to par (something you failed to do prior to launch)...
Your bias is starting to shift more towards a neutral zone now. (please don't deny that you were never biased towards AMD).
You did post back my comment but didn't answer the questions.
mo,
And what if the Honda cost more than the BMW, what’s the problem there?
Are you saying that for example don’t exist pens that are more expensive and worst than the cheaper ones?
There are also cars that are much worst and cost more than others, I don’t see a problem with that.
Maybe the differentiation is the key. Why not have lots of the same products?
Wasn’t this the all point that made Intel release the 20 x86 licenses to other manufactures?
Servers for example AMD users DDR2 Intel FBDIMMS, AMD cpu cost more but the DDR2 memory cost less. I can build a similar system with less money with AMD, is this not a differential factor and with price differentiation too?
SSD disc cost more than regular plates disc and have much less store capacity, so removing the price equation, which is the differential factor between them?
mo
"What if the BMW cost the same as the Honda?"
Well, that wasn't the point lex was making. There is no doubt that Intel's top chips are faster than AMD's but that is also not the mainstream market. I feel the same way about processors as graphic cards and motherboards; if something is better at the same price that is what you go with.
"you have realized that something is indeed not up to par (something you failed to do prior to launch)..."
Well, I'm not sure how I could form a solid opinion before testing was done.
"Your bias is starting to shift more towards a neutral zone now. (please don't deny that you were never biased towards AMD)."
It isn't a matter of bias; it is a matter of credibility. AMD's credibility is a bit low right now. So, I'm a bit more skeptical when they say that everything is on track. We had a similar situation back in 2003 when Intel had lots of credibility (after its Northwood success) so everyone believed that Prescott was going to be a K8 killer.
Today, AMD's chipsets are quite good and they are much closer to parity with nVidia in discreet graphics. Without seeing something solid in processors however I am less optimistic.
ho ho
"You did post back my comment but didn't answer the questions."
As far as I can tell, I made comments about your posts. Maybe you could be more specific.
lex
Although I deleted your original post your comments are quoted in their entirety. I realized that you have a lot of trouble with facts. I'll see if I can assist you.
"Troll posting controversial messages or is it more fact you don't like."
If you look at the parts I called trolls you will see that the first one contained no facts at all. I explained why each of your points was incorrect. The second part only contained one point that might be true. If AMD is doing worse by end of 2008, Ruiz could be forced out. However, this is a long way from your assertion that Ruiz will be forced out.
"INTEL has 3 fabs in Oregon, one in California, two in Arizona, and one in New Mexio, and another near Boston. It has a fab overseas in Ireland and another in Isreal and one going into China. If you look there are pretty good reason to put fabs in China, the EU. I often wonder why INTEL put one one truck bomb or scud missle away from disaster but you can't say INTEL runs to put fabs in cheap places with tax heavens and goverment handouts.
AMD on the other hand is something else there in Dresden and the deal they got from Saxony. And they got a pretty sweet deal there in NY for a fab they will never build.
Please its clear why AMD fab is overseas. AMD can't afford to do it in the US. INTEL can and choses to put a portion over seas for multiple benifits."
This entire section is garbage. The truth is that both AMD and Intel receive subsidies from places where they build facilities. Perhaps your memory is failing. I'll help you:
jewish news weekly, october 1995
As an incentive, Israel awarded Intel a controversial tax-subsidy package worth a third of its overall investment -- 38 percent -- to be paid over 10 years, according to Reuters.
International Herald Tribune 2005
The Israeli government provided substantial subsidies for Intel to build its first plant, which opened here in 1999. This time, the government's benefits package is even larger, with incentives totaling $525 million.
case study of Intel
In 1993 the company set off a bidding war among half a dozen states when it announced plans for the largest plant expansion in the history of the semiconductor industry, one that involved the hiring of up to 1,000 new workers and investment of $1 billion. In the end, it chose Rio Rancho again, thanks to the offer of a new $2 billion IRB deal, investment credits and training funds.
Intel went with Chandler only after getting the state legislature to approve various tax exemptions and tax credits. The company also received foreign trade zone status from the U.S. Department of Commerce, which triggered additional tax breaks.
Intel built a fabrication plant in Washington County, Oregon in 1996 and proceeded to seek a series of tax breaks under the auspices of the state's Strategic Investment Program, which was reportedly drafted with the help of the company.
The truth is that both AMD and Intel make decisions about where to build faciities based on exactly the same thing: saving money.
"You talk about assertions.. Tell me what data do you have to back that AMD will get 45nm on time, have High K, will fix their bugs."
I'm wondering whether you have been reading another blog since you've completely misquoted me. I've already expressed doubt that 45nm is on track. AMD has said nothing about high-K any earlier than the 2nd half of 2009, and I've only said that AMD might be able to fix any major problems with K10 in Shanghai.
" I can debate and win any aspect about those 3 and why AMD won't come in on time with technology and why they won't fix their bug. My conclusion is based on understanding what it takes not hope of what I want to happen."
Yes, when you make up strawman arguments it bolsters your debating ability considerably. However, since you don't seem to be aware that Intel gets subsidies I wonder what else you are not aware of.
"As to my opinion of the executives. Why is that trolling. AMD executives have made horrible decisions and its natural to expect the top guy to take the ax. Its no different then what happend at all the financial companies when the sub-prime blew up. The reason Dirk will be gone in 2009 is that he too can't fix it."
Your opinion is insane. Even if Ruiz were forced out in, say, late 2008 there is no indication that Meyer would only be given a year to fix things nor is there any reason to think he would or would not fail. You would need to wait at least a year for your opinion to have any connection with reality. Your notion that ATI will be split off again makes no sense because ATI is currently doing well. Why would AMD sell off a demonstrated asset?
" AMD's problem is it can't compete across all business segments with INTEL."
This view was correct in early 2002 but is years out of date. There is again no reason to believe that AMD will withdraw from servers, the desktop, or mobile. What do you base this on other than fantasy?
"That isn't trolling thats the fact! I'm all eyes for you to tell me how they will compete effectively."
It's fact because you believe it? AMD's current graphics and chipset position seems solid. The current mobile position is adequate so we'll see how griffin does. Obviously K10 needs fixed. I think anyone who is not delusional though would agree that it would take massive continued failure for AMD to actually withdraw from any of these segments. Do you honestly expect AMD to fail forever?
"As to Prescott vs Opeteron... you still fail to see what I'm talking about. I won't try anyomre as its beyond your ability."
Actually, I understand perfectly what you are talking about but I've already said that your analogy is incorrect. Since you still don't understand why your analogy is incorrect I'll explain it further:
2004: Intel's 90nm process good
2007: AMD's 65nm process still unproven
2004: Prescott launches first as Celeron D due to overheating
2007: K10 launches first as Opteron
2004: Prescott's extremely long pipeline causes problems with power consumption.
2007: K10's pipeline is similar to K8's
2007: After 4 full years of development, Tulsa tops out with 65nm at speeds barely above the Northwood cores on 130nm.
2007: K10 is brand new and has had no development.
I'm sorry but I don't see how these two are related. As far as I can tell your analogy consists entirely of a lame attempt to equate K10 with Intel's biggest failure.
There is no doubt that Intel's top chips are faster than AMD's but that is also not the mainstream market.
Wrong. Intel's Slowest Quad Core is Faster than AMD's fastest Quad offering. (Q6600 vs Phenom 9600).
Not sure why you keep going to the top end...when it is across the board.
joe
"Wrong. Intel's Slowest Quad Core is Faster than AMD's fastest Quad offering. (Q6600 vs Phenom 9600)."
That's fine but quad cores are not exactly mainstream at this point.
Joe,
Wrong. Intel's Slowest Quad Core is Faster than AMD's fastest Quad offering. (Q6600 vs Phenom 9600).
The situation is basically the same when AMD X2 3800+ was also faster than all Intel offerings and also overclocked like crazy.
And I don’t think when the X2 3800+ with the cost of 300$ at that time could be considered mainstream.
Mainstream Intel products are the E6xxx line.
Value/mainstream the E4xxx
Value the E2xxx
And at the entry the Celeron line.
With some products costing a few bucks like motherboards, memory, disks, DVD drives, … 300$ is very far for being considered mainstream.
Aguia
The situation is basically the same when AMD X2 3800+ was also faster than all Intel offerings and also overclocked like crazy.
Is your name Sharikou?
enumae
I think what he is referring to is late 2005 when AMD's X2 offerings were well ahead of Intel's Smithfield dual core products. This situation didn't really change until C2D was introduced mid 2006.
Scientia
I think what he is referring to is late 2005 when AMD's X2 offerings were well ahead of Intel's Smithfield dual core products...
His comment stating all Intel offerings was very Sharikou like :)
Yes, late August 2005.
Thanks
enumae,
I wonder why I sounded like sharikou and joe didn’t.
Maybe you have a problem with the AMD name.
Aguia
I wonder why I sounded like sharikou and joe didn’t.
Sharikou used to talk about the X2 3800 beating all Intel chips.
Maybe you have a problem with the AMD name.
Well, maybe your right, like our discussion about ATI's 3870...
enumae,
Well, maybe your right, like our discussion about ATI's 3870...
Well at least you are a correct person.
Scientia, what do you make of this article on EETimes.com:
AMD exec discusses Barcelona debacle
Both the Opteron and the desktop Phenom chips suffer 5 to 20 percent performance hits as a result of the BIOS fix. But Rivas clarified a point about the quad-core Phenoms, which were launched in November and have been shipped with the workaround in place at the ramp-up AMD planned. Those processors are listed at clock speeds that account for the degradation from the BIOS fix, he said, explaining why the first available Phenoms have listed speeds and prices below those AMD initially projected for fourth-quarter shipments. AMD's quad-core Opteron, commonly known by its codename Barcelona, will get a design correction before it is made available in volume to the channel and tier ones, Rivas said. The next batch of Phenoms will also be partially redesigned.
At least this Rivas fellow is honest. That's more than I could say for Randy Allen, Hector Ruiz or (formerly) Henri Richards!
The Rivas article puzzled me at times, for instance he said they had factored in the loss of performance when selling Phenom with the TLB workaround, however all the testing at Tahoe did not have the TLB fix in place so would be 5-20% too high. Maybe he means in repsect to selling price.
One concern was that he indicated that the B3 fix for the TLB issue is not actually tested yet and they will have the results in January. I was under the impression they had it fixed and were just doing validation tests, but it seems they have not even got to seeing whether it is a good fix or not. This leads to a level of uncertainty still.
Really AMD need to make sure it is fixed and then concentrate on getting good yields if these chips are going to be value priced. Yeild might be better than chasing ultimate speed currently as they are not going to get that away from Intel, they'd need 3.3Ghz or so.
As the 3870 is a much better gpu than the 2900XT then they need to replicate that again with Phenom. Whether this, in similar fashion, requires a process shrink though is a good question.
Yo Scientia
Want to do a article now about the ATI purchase
Since AMD's buy how many executives have left? How many products are out on schedule, and how have they done against nvidia?
Want to take a stab about how many executive man years have the top guys at AMD been spending on ATI instead of trying to fix the broken CPU which was once their bread and butter operation?
Now they admit they way over paid and they don't even have a number of the goodwill amount that has evaporated since they purchased?
Tell me again what did AMD get beside a ton of debt and some extra heads that they couldn't have gotten without spending billions?
AMD products are slow and again only selling as "value" product making veyr little money on that brand new expensive 65nm factory that will be a huge depreciation anchor next year with no high end cash flow.
45nm is coming wiht huge additional capital spends on immersion lithography but won't have HighK/Metal gate so expect AMD 45nm to have 3x more gate leakage and inferior transistors then INTEL. That delta is a very heavy penality to carry on an already burdened design team.
Lastly its interesting IBM is going gate first. The worlds leading expert told the world on Sunday that gate first can't work it simply isn't thermodynamically possible given the physics of how Metals react to HfO.
Lastly its interesting IBM is going gate first. The worlds leading expert told the world on Sunday that gate first can't work it simply isn't thermodynamically possible given the physics of how Metals react to HfO.
I think Intel said storing 2 bits in one transistor was impossible to do and AMD and Fujitsu have done with their flash.
Just because some dumb guy goes out and says it’s impossible don’t take his word as certain.
I prefer the "dumb" guys that say it’s possible and do it, then the dumb guys that go out and say it’s impossible without really knowing what they are talking about.
Sorry, I haven't been keeping up lately. Had a fever of 102. It's down to 100.7 so I'll see if I can catch up a bit. Why are you reading the chopped up EE Times version instead of the original at ChannelWeb? I guess if you read the EE Times version you do leave out all the positive comments if that is your preference. For example:
Nobody's more frustrated than I am, because there's demand in the market.
So, there is demand for Barcelona.
We still are shipping Phenoms in quantities. We are taking good care of the channel, as far as I know. If you're hearing negative feedback on the Phenom, I'd like to hear about it.
No, it hasn't been about Phenom, it's been about the Opterons.
Right. I think this particular errata really affects server work . . . But it doesn't affect the client in the same manner. We were able to identify a workaround for the desktop product that would not have been suitable for several of the server workloads, of the quality and performance we need for the server-workstation market.
So, Phenom is shipping. Also, the 45nm comments are very positive:
I've even seen that people are speculating that your 45nm process isn't on target. Can you address that a little bit?
We will have initial samples also in January. I'm fairly confident that those puppies are going to boot, and then we can have a follow-up conference call and I'll tell you, 'The sucker is booting.' The 45nm, we consider it Rev C of the device. So all the learning, all the hard knocks that we had on Barcelona, we're going to apply it to Shanghai. I just read on Friday, I have to admit to my delight because misery loves company, that our competitor's 45nm part seems to be delayed too.
lex
"Want to do a article now about the ATI purchase
Since AMD's buy how many executives have left? How many products are out on schedule, and how have they done against nvidia?
Want to take a stab about how many executive man years have the top guys at AMD been spending on ATI instead of trying to fix the broken CPU which was once their bread and butter operation?"
Lex, why do you keep trying to dredge through the muck to put a negative spin on AMD? Some executives have left but I notice you didn't mention the ones that were hired.
The ATI products appear to be on schedule now, much better than April, earlier this year. Also, we know that AMD's troubles were caused by losing $2 Billion in 4 quarters, not by purchasing ATI. I'm baffled that you would actually believe that AMD has pulled engineers off Barcelona to "fix" ATI's products. In reality, AMD still had some chipset designers in Dresden, so if anything, these may have been assigned to help ATI.
Now they admit they way over paid and they don't even have a number of the goodwill amount that has evaporated since they purchased?
Quite true. I'm still wondering how much though. For example, overpaying by $3 Billion is a whole lot worse than overpaying by, say, $400 Million.
Still, the ATI products are one of the successes at AMD and their outlook would be worse without them (although they would currently have more cash). So, you have a choice between being worse now but with a better outlook or better now with a worse outlook.
There is no need to try to make up things since AMD admits the problem, Forbes:
President and Chief Operating Officer Dirk Meyer said AMD has "done a lot of things very well" since the first quarter. But, he added, "we have done one thing very poorly. Namely, we haven't delivered our quad-core product."
I'm also not sure where you get the idea that AMD has been dragging its feet fixing Barcelona.
AMD, Meyer added, knows what the problems are and is "hell-bent on fixing them."
I would say that is very true. The January timeframe for a possible fix matches well with the timeframe for discovery, redesign, and 14 weeks to run samples.
The company said the problems were not related to manufacturing, but were design-specific that, once found, were pretty staightforward to fix.
We can only hope this is true and that 65nm will deliver some higher clocks soon.
Chief Executive Hector Ruiz said AMD, the world's No. 2 microprocessor maker behind Intel Corp. (nasdaq: INTC - news - people ), is committed to finishing the second quarter at breakeven, with a return to a profit in the third quarter.
Chief Financial Officer Robert Rivet said AMD expects sales to grow faster than the market in 2008. The company, which has been growing its revenue even as it posted quarterly losses, also expects its gross margins to continue to expand.
Have to take these statements with a grain of salt since AMD didn't meet its target of breaking even in Q4. However, if the only reason is the problems with Barcelona then it could be true I guess.
Blah blah blah blah..
Bottom line that Scientia continues to miss is even if AMD fixes its design problems it still has a tiny problem
AMD is behind a good year on 45nm. So for most of 2008 they will compete with a product that has esenttial double the die size for the same # of transistors. In addition their 65nm and 45nm stil use traditional SiON gateoxide. In IBM/AMD's own words that cost them considerable more in leakage and performance, about 30%. So how does AMD compete when its already losing when you handicap them with a process that is half the density and between 30-50% slower. You think AMD designers are that clever.
Sorry, architecture can't make it up.
AMD is at a gun fight but instead of showing up with a six shotter its got a crossbow. I don't care how good they are they can't compete with what their technologist give them especially when they are trying to cheat thermodynamics.
Dirk and company could be smart but they can't shrink transistors, stop leakage nor change physics.
I'm all eyes whats the strategy for AMD to compete across the board and make money in a sustained fashion?
lex
What is the most intelligent thing you can add to this discussion?
"Blah blah blah blah.."
Hmmm, ok. Well, did you read the part where he says 45nm samples from AMD will be out in January?
"AMD is behind a good year on 45nm."
I guess not. You are aware aren't you that Intel is not yet fully converted to 45nm?
" So for most of 2008 they will compete with a product that has esenttial double the die size for the same # of transistors."
I guess not. Lex can you add anything of value to this discussion?
"In addition their 65nm and 45nm stil use traditional SiON gateoxide. In IBM/AMD's own words that cost them considerable more in leakage and performance, about 30%. So how does AMD compete when its already losing when you handicap them with a process that is half the density and between 30-50% slower."
Well, that is halfway intelligent anyway. First of all, you are again making assumptions about 45nm that don't match what either company is saying. Intel is not claiming that it will have 100% 45nm before year's end. AMD is saying that they hope to have bootable samples in January. We'll see. AMD could slip from this but there is always the possibility that they could gain back some lost time.
I've seen no indication that Intel's 45nm would be 30-50% faster than AMD's. However, I agree that you can't escape the gate leakage problem. That should give Intel some kind of edge but I simply don't know how much.
"they are trying to cheat thermodynamics."
AMD has again indicated that it is not going to move to high-K any sooner than 2009 so I would not expect this before Bulldozer. There are some suggestions that Nehalem could be having problems but it could slip a long way, six months, and still arrive in time to compete with Bulldozer. Also, since Penryn is still likely to be ahead in Q4 2008, Intel has little to be concerned about.
"I'm all eyes whats the strategy for AMD to compete across the board and make money in a sustained fashion?"
Exactly.
Chief Financial Officer Robert Rivet said AMD expects sales to grow faster than the market in 2008.
I wonder how they plan to do this as well. I have no idea if this is realistic or just another rosy projection. I'll wait and see.
They apparently plan on doing it by cutting R&D and CAPEX considerably.
Consider this:
http://www.xbitlabs.com/discussion/4525.html
No new high end GPU (aside from the 3870 dual GPU card) from Ati until 2009. There were earlier reports that Nvidia will release the Geforce 9 as early as next February.
Cutting R&D doesn't have much impact in the short term, but long term I am seriously worried about DAAMIT's capability to compete on the high end with Nvidia and Intel.
Scientia,
Whats you opinion on the why sites like Techreport, Tomshardware, Xbitlabs, ...
Are all writing articles bashing AMD for their "bad" products.
I didn’t see that when Intel had the ridiculous Pentium 4 and PentiumD. Why do you think they are taking on AMD?
For example they make “fun” of the SB600 has being outdated but forget that Intel and Nvidia SB are also outdated. And how Nvidia didn’t even release their PCIe 2.0 chipsets nor Intel (except X38), ...
Even the Phenom not being exceptional it’s a much better adversary for the Core 2 Duo than the Pentium4/PentiumD was to the Athlon64/Athlon64X2.
What’s your take on this entire bash?
Scientia
What do you make of this statement from the Rivas article:
"Those processors are listed at clock speeds that account for the degradation from the BIOS fix, he said, explaining why the first available Phenoms have listed speeds and prices below those AMD initially projected for fourth-quarter shipments."
What are the actual clock speeds of the chips? Is the 2.3 GHz really a 2.5 chip with a bios fix?
"I didn’t see that when Intel had the ridiculous Pentium 4 and PentiumD. Why do you think they are taking on AMD?"
...
When AMD was at the top of their game, no one recommended Pentium 4/Ds. Stop trying to put AMD in the "I'm innocent" spotlight. What are you suggesting? Intel paid these sites off to discredit AMD?
Take example from Anand after Core 2s release,
"But make no mistake, what you see before you is not the power hungry, poor performing, non-competitive garbage (sorry guys, it's the truth) that Intel has been shoving down our throats for the greater part of the past 5 years."
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2795
You're pathetic.
Okay, for those among us who just can't stop shouting about the supposed loss of expertise at AMD here is an article that should knock this idea off its pedestal.
Former MIPS Chief Technology Officer joins AMD:
Today, the company says it has hired industry veteran Mike Uhler as its first ever Vice President of Accelerated Computing. Uhler has 34 computer architecture and design patents to his name, and his 30-year career spans positions as Senior Consulting Engineer for Digital Equipment Corp, Director of Engineering for Silicon Graphics, and most recently, Chief Technology Officer of MIPS Technologies.
As AMD's VP of Accelerated Computing, Uhler will contribute to the company's Accelerated Computing R&D program, whose focus is to accelerate specific tasks through the use of both discrete co-processors on-chip accelerator cores. AMD CTO and Senior VP Phil Hester explains, "Customers are asking for design innovations that apply hardware and software more directly toward a set of workloads, rather than a one-size-fits-all approach. Mike furthers AMD's design leadership by applying that philosophy to our Accelerated Computing vision and I welcome him aboard."
*****
As I've already mentioned the myth of an expertise drain only has some weight when you talk about the people leaving but conveniently leave out the people hired. There is no general loss of expertise at AMD.
giant
"They apparently plan on doing it by cutting R&D and CAPEX considerably."
The chart at Tech Report, AMD Targets only seems to show a small reduction in the percentage spent for R&D.
The CapEx target listed in the 2008 Guidance is $1.1 Billion.
Okay, we know (by simply checking AMD's public records) that AMD spent close to $1.9 Billion in 2006 and should be about the same in 2007. That would be a reduction of about 40% which of course is large.
But now comes the big question. Is a 40% reduction somehow gutting AMD's ability to compete? My guess is no. Almost all of the money spent in 2006 and 2007 was for tooling for FAB 36. We also know that at least some tooling has been purchased for FAB 38. So, the $1.1 Billion is probably enough to top out FAB 36 with 45nm tooling and buy some small set of 45nm tooling for FAB 38. What has probably been cut is money to aggressive ramp FAB 38 as AMD did FAB 36 in 2006. However, this probably matches with AMD's projected sales.
Fudzilla:
Demand for 38x0 is tremendous and no delay of the next GPU-generation.
Poke,
Take example from Anand after Core 2s release,
"But make no mistake, what you see before you is not the power hungry, poor performing, non-competitive garbage (sorry guys, it's the truth) that Intel has been shoving down our throats for the greater part of the past 5 years."
Yeah that’s much easy to say, after 5 years selling that crap. Why did they not say it before?
It’s like I sell bad cars for years but keep the mouth shut, then I make a good one and say this one is fine compared to the others that where complete garbage, how do you fill with that?
pathetic.
Then you know better than me ...
I noticed, that when Intel started hinting about conroe performance, and a when they admited that netburst arch was a bad move, everyone started repeating same stuff: conroe is brilliant, pentiums are shit.
Before that only few brave ones were telling the truth
core2dude spewed...
"This blog is dead!
Good showing Scientia!! This is what you get for censoring sensible posters!!!"
Gee, that's funny... Why then are chumps like you still spammin' his blog like a child with tourettes?
In the most obvious case you don't figure out the real reason why there isn't many people posting here lately, here's one word for you:
CHRISTMAS.
Well Scientia, it is due time for a new blog post methinks, perhaps a year end roundup of the past year and comments on recent AMD developments when you get some vacation time? Keep up the good work and Merry Christmas dewd.
It's easy as 1 2 3 :)
1. There is no way that 65nm is slower than 90nm. AMD is releasing 90nm because there is no other way they would sell, of course right, when both processes have the same thermal stats and Dell really cares about the die size right?
2. AMD is pretending. 20% faster on int_rate and 40% faster on fp_rate is just a simulation of quad k8 and quad k10 will be even more amazing.
3. DTX, small form DESKTOPS (aka becoming extinct) are absolutely revolutionary, will take over, and a holy grail to AMD's profitability!
By the way, AMD used to say they might break even Q407. Now they're saying Q308. Oh no.
john
"1. There is no way that 65nm is slower than 90nm. AMD is releasing 90nm because there is no other way they would sell, of course right, when both processes have the same thermal stats and Dell really cares about the die size right?"
AMD is no longer making 90nm chips. Only 65nm is in production.
"2. AMD is pretending. 20% faster on int_rate and 40% faster on fp_rate is just a simulation of quad k8 and quad k10 will be even more amazing."
Here. Try looking at this Neoseeker review of Phenom 9900. At 2.6 Ghz it runs almost equal to a 2.66 Ghz QX6700. This suggests that AMD should be back in the game by Q2 or at least by Shanghai.
"3. DTX, small form DESKTOPS (aka becoming extinct) are absolutely revolutionary, will take over, and a holy grail to AMD's profitability!"
There are several companies doing DTX. We'll have to wait and see how this develops over Q1 and Q2. BTW, I've never, ever suggested that it was the key to AMD's profitability, just another item that could help AMD. I think that is still true.
"By the way, AMD used to say they might break even Q407. Now they're saying Q308."
This is quite true and is reason to maintain some skepticism. However, I don't see it as a reason to troll. Remember the posters who swore up and down that FAB 30 would either be shuttered all 2008 or would be sold? I noticed that you didn't mention that.
lou ceifer
I was working on one post but now we've gotten the news that Bulldozer has been pushed back. This makes it a lot more difficult to evaluate AMD's competitiveness in 2009 against Nehalem.
muziqaz
"Before that only few brave ones were telling the truth"
I was saying that Prescott wouldn't be very good before it was released in 2003. I don't know anyone else who was saying that at the time. I even had people telling me that I was spreading FUD when I said it would have a 30 stage pipeline. Then, of course, it arrived with a 31 stage pipeline.
John
"That's not the issue. The issue is your insistence that there is no possible way that AMD's 65nm has worse performance characteristics compared to 90nm"
No, I've never said that. AMD has yet to produce a 65nm chip as fast as their 90nm chips. That should be obvious. You can stop trying to strengthen your arguments by misquoting me.
"and that 90nm is released at faster speeds because there would be no other way they would sell."
No. I believe that AMD cannot yet make a 3.2Ghz capable 65nm chip. It is possible that they cannot make a 3.0Ghz 65nm chip yet. Strike two on misquotes.
"You didn't address my point. My point was that AMD presented something pretty clear, K10 results, which you hyped to be merely K8."
Now adding lies to your misquotes. The AMD charts clearly stated that they were based on K8 projections. Anyone who can read could see that. Your post was a waste of time.
Scientia
The AMD charts clearly stated that they were based on K8 projections.
I am sorry Scientia, they did not say anything, let alone clearly, that they were based on K8.
If you recall me and you had a rather long debate about this.
I found a few of them I had saved, they were from the Powerpoint presentation that was floating around, and also from George Ou.
Link - George Ou
Link - Power Point
Anybody got a pair of Opteron 2212's to loan me? It looks like my 2xOpteron 2347s at 1.9GHz are 26% faster than my 4xOpteron 875s at 2.2GHz on my server workload. Dunno how much of that difference is due to DDR333 on the 875s vs DDR2 667 on the 2347s. I guess the only way to really compare would be to swap some Opteron 2210's or 2212's in place of the 2347s. Would be an interesting test.
AMD made some very bold claims. '40% faster than Clovertown across a wide variety of workloads!' 'Barcelona will clearly be a leading performance product. It's going to completely blow away this existing Clovertown product in every dimension!' Just a small sample of AMD FUD.
The results, plain and simple.
http://www.hexus.net/content/item.php?item=10427&page=1
Phenom 9600 is blown away by Q6600 time and time again. These results are WITHOUT the TLB fix that lowers performance by 13.9% according to TechReport.
Yorkfield increases performance further over the current 65nm parts.
AMD has nothing to compete with Intel's current products. By the time AMD's Shanghai is out, it will have to fight with Nehalem. AMD won't have Shanghai samples until January, that's four months behind Intel with Nehalem.
With Bulldozer off current roadmaps, AMD has clearly given up all hope of competing with Intel in the high end segment.
AMD Phenom 2.4 and 2.6Ghz delayed until Q2'08:
http://www.digitimes.com/mobos/a20071224PD200.html
AMD has recently notified its partners that the launch of higher-end quad-core Phenom processors, including the 9700 and 9900, will be postponed to the second quarter of 2008 from the original schedule of early 2008, according to sources at motherboard makers.
AMD talks about a 3Ghz Phenom, but will not have such products until Intel launches Nehalem in Q4'08.
john
Okay, now I understand why you are confused.
"Howell said before The fact that Brisbane has not clocked above 2.7GHz 11 months since it's (very soft) launch at 2.6GHz ought to be indication enough. of course implying that 65nm's performance is lagging."
No, Howell didn't say that. That was actually said by Periander.
"Your responded with
That's not actually true as AMD still has to sell 90nm chips at the higher speeds. "
No, I didn't say that. That was actually said by Howell. I can understand why you were confused though. I had edited Howell's post because of a vulgar line about jamming a cattle prod up someone's butt. So, you thought the post was mine. But now you know it was Howell's.
"So I'm not seeing where AMD explicitly said that these are K8 results and we just adjusted the scores for clock and cores."
Then John you are either blind or are pretty bad at research. Two months after I mentioned that the charts were based on simulations it was mentioned both by Ou and over at Dailytech. You ought to be able to find one of those articles.
enumae
Thanks for the links. It wasn't much of a debate. The charts clearly say that they are based on internal simulations and therefore not real K10 hardware. However, if you run the numbers you'll see that they are identical to the K8 numbers if you adjust for clock and double the number of cores. So, either the numbers are K8 numbers or they are identical to K8 numbers. Whichever way you prefer to say it.
giant
"AMD made some very bold claims. '40% faster than Clovertown across a wide variety of workloads!'"
That statement is outdated. That was AMD's statement from when the fastest Intel quad was 2.6Ghz. When Intel released faster quads AMD changed that to "50% faster at the same clock". Now, even assuming that AMD was referring to just server applications we still aren't seeing that.
The best I've currently seen for non-server applications is 8% slower comparing Phenom to Kentsfield. That is for a 2.6Ghz part. Essentially, a 2.6 Ghz Phenom seems to be running about equal to a 2.4 Ghz Kentsfield. Penryn, however, is a little faster than Kentsfield.
Scientia
The AMD charts clearly stated that they were based on K8 projections.
Again, I was commenting directly about your claim that the slides said that the scores were based on K8.
Here is what you said before...
Scientia from AMDZone said...
The notion that the statements made by AMD are based on K8 projections and not K10 is just my best guess. You are hardly the only one to disagree with this and obviously AMD has never stated this directly.
May 08, 2007 7:38 AM
That is from our discussion.
So no, it was not said by AMD that they were based on K8, it was your best guess.
enumae
Apparently you didn't bother reading what I just wrote. Since you seem to be having trouble I'll summarize:
1.) AMD specifically said that the scores were based on internal simulations.
Therefore, these scores were not based on real hardware. In other words, these scores were not based on K10.
2.) The scores listed are identical to scaled K8 scores.
So, you can either say that the K10 simulations were identical to K8. Or, you can say that the scores were simulations of scaled K8. That K10 simulation would just happen to match K8 is unlikely so it is more likely that the scores are in fact scaled K8. However, if you are bound and determined to beat your semantic arguments to death you could honestly say that:
The scores may not be based on K8 but by sheer coincidence just happen to match perfectly with scaled K8 scores.
If that makes you happy, knock yourself out.
Scientia
However, if you are bound and determined to beat your semantic arguments to death you could honestly say that...
If that makes you happy, knock yourself out.
This is not semantics, and is there any reason to be rude or sarcastic?
I didn't make a false statement, you did and were wrong in making the statement.
"The AMD charts clearly stated that they were based on K8 projections."
The chart does not, it was your conclusion and correct interpretation that makes it clear that the information on the chart is K8 based.
I made it clear that you were correct in the other thread/post.
The scores may not be based on K8 but by sheer coincidence just happen to match perfectly with scaled K8 scores.
Well looks like you have finally found some senses. At the time, K8 had been pretty optimized for SPEC. Yet still then, it didn't scale 100% with added cores and certainly not with clock speed. The mere coincidence of K10 spec scores matching for adjusted K8 cores was a hint that K10 was not so amazingly revolutionary...which has been shown in tests besides SPEC CPU.
1.) AMD specifically said that the scores were based on internal simulations.
spec.org does not allow publication of results based on nonexistent hardware without labeling as such and AMD labeled so to comply. Further more, it used to be that the L3 was supposed to be faster than the core or something but that hasn't panned out. Perhaps at the time, the got K10 ES results with the same slow L3 and estimated what it could have been with fast l3.
Post a Comment