Computex 2007: AMD's Barcelona and Other Products
by Gary Key on June 11, 2007 4:00 AM EST- Posted in
- Trade Shows
AMD's Barcelona: Why we haven't published benchmarks
A month ago we were able to present to you a rare look at AMD's forthcoming roadmap, detailing everything from new plans for the mobile space to giving a better idea of AMD's reasoning behind the ATI acquisition. We left that article with a positive and hopeful note for AMD:
"For a while we had lost confidence in AMD, like many of you had as well, and although AMD's position in the market hasn't changed we are more confident now that it can actually bounce back from this. Intel seemed to have the perfect roadmap with Conroe, Penryn and Nehalem all lined up back to back, and we saw little room for AMD to compete. Now, coming away from these meetings, we do believe that AMD may have a fighting chance. Over the coming months you'll begin to see why; it won't be an easy battle, but it will be one that will be fought with more than just price."
A strong roadmap alone does not make for a successful company; we need to see near term execution as well. For AMD, that means Barcelona has to be competitive. The interesting part of AMD's disclosures as of late is that as much information as AMD has given us about its roadmap for 2008, 2009 and beyond, we have little to no details about when we can expect Barcelona and how fast it will be.
When we headed out to Taiwan, a country of leaked processors and benchmarking opportunities, for Computex we surely expected to return with some Barcelona performance figures. We were hoping we'd come back with the very data that AMD hadn't allowed us to get ourselves when we visited the company over a month ago. And while there were some performance results reported from Taiwan, there was an eerie silence about AMD's updated micro-architecture.
We were determined not to leave the island without running at least one test on Barcelona. We worked long and hard, and we were finally able to spend some time alone with Barcelona in Taiwan. But the story doesn't end there; it's unfortunately not that simple.
Motherboard Problems
We know that Barcelona works and runs benchmarks, as we saw back at AMD in May. But the demos that AMD ran were on its own motherboards, not on motherboards from its partners. AMD's partners just recently received their first "production quality" Barcelona samples, and as expected, the current boards required some heavy BIOS work before the new chips would even work, much less perform up to the expectations set by AMD.
The motherboard we tested on had minimal HT functionality and wouldn't run at memory speeds faster than DDR2-667; most 3D video cards wouldn't even work in the motherboard. Memory performance was just atrocious on the system, but the motherboard manufacturers we worked with attributed this to BIOS tuning issues that should be fixed in the very near future.
In the end, performance was absolutely terrible. We're beginning to understand why AMD didn't let us test Barcelona last month. It's not that AMD is waiting to surprise Intel; it's that the platform just isn't ready for production yet.
A month ago we were able to present to you a rare look at AMD's forthcoming roadmap, detailing everything from new plans for the mobile space to giving a better idea of AMD's reasoning behind the ATI acquisition. We left that article with a positive and hopeful note for AMD:
"For a while we had lost confidence in AMD, like many of you had as well, and although AMD's position in the market hasn't changed we are more confident now that it can actually bounce back from this. Intel seemed to have the perfect roadmap with Conroe, Penryn and Nehalem all lined up back to back, and we saw little room for AMD to compete. Now, coming away from these meetings, we do believe that AMD may have a fighting chance. Over the coming months you'll begin to see why; it won't be an easy battle, but it will be one that will be fought with more than just price."
A strong roadmap alone does not make for a successful company; we need to see near term execution as well. For AMD, that means Barcelona has to be competitive. The interesting part of AMD's disclosures as of late is that as much information as AMD has given us about its roadmap for 2008, 2009 and beyond, we have little to no details about when we can expect Barcelona and how fast it will be.
When we headed out to Taiwan, a country of leaked processors and benchmarking opportunities, for Computex we surely expected to return with some Barcelona performance figures. We were hoping we'd come back with the very data that AMD hadn't allowed us to get ourselves when we visited the company over a month ago. And while there were some performance results reported from Taiwan, there was an eerie silence about AMD's updated micro-architecture.
We were determined not to leave the island without running at least one test on Barcelona. We worked long and hard, and we were finally able to spend some time alone with Barcelona in Taiwan. But the story doesn't end there; it's unfortunately not that simple.
Motherboard Problems
We know that Barcelona works and runs benchmarks, as we saw back at AMD in May. But the demos that AMD ran were on its own motherboards, not on motherboards from its partners. AMD's partners just recently received their first "production quality" Barcelona samples, and as expected, the current boards required some heavy BIOS work before the new chips would even work, much less perform up to the expectations set by AMD.
The motherboard we tested on had minimal HT functionality and wouldn't run at memory speeds faster than DDR2-667; most 3D video cards wouldn't even work in the motherboard. Memory performance was just atrocious on the system, but the motherboard manufacturers we worked with attributed this to BIOS tuning issues that should be fixed in the very near future.
In the end, performance was absolutely terrible. We're beginning to understand why AMD didn't let us test Barcelona last month. It's not that AMD is waiting to surprise Intel; it's that the platform just isn't ready for production yet.
Facebook
Twitter
RSS
64 Comments
View All Comments
TA152H - Tuesday, June 12, 2007 - link
I have seen other benchmarks, and they are actually very close. Plus, you have to discount other things in the system like higher FSB speeds, etc...Penryn will broadly be 5-10% faster, everything else being equal, on a majority of benchmarks. You might find a few that will, like those that love division, but broadly it's not going to happen. Pre-release benchmarks aren't even close, except for those made to illustrate the differences. Also, most of that comes from cache.
It is entirely relevant where it comes from, because adding additional cache is easy for AMD as well, although not as easy and not quite as effective. If they hadn't created Penryn, they would have just added the same cache, or more, to Conroe with the shrink. Besides, Barcelona is adding a L3 cache, which should offer a big improvement since they have relatively small L2 caches.
If you're expecting the Penryn to maintain the lead the Conroe has over the current K8, you're in for a shock. It'll be much closer.
Who needs Quad core? Not many people. If that's the Penryn's claim to fame, it's kind of limited. In the server space, sure, but that's not exactly where Intel is strong. They have a system architecture that is very poor, although more cache should help. In dual core, Conroe can go much higher, and that's where most processors sell now. We'll see how well they do with Penryn.
CobraT1 - Tuesday, June 12, 2007 - link
Who's getting emotional?Emryse didn't say anything about architecture. He merely stated his opinion that "worst-case scenario would be for (and this is very likely) Intel's '08 lineup to be introduced at the same time Barcelona releases (soft or hard)." He was referring to (correct me if I'm wrong Emryse) public opinion, hence the statement he prefaced with "sustinance often exists in the court of public opinion". He was stating that if Barcy is pushed out from it's pre-Penryn announced launch, until Penryn is launched, it will be a PR nightmare for AMD.
1.) Problems to release on schedule denotes execution issues and reflects poorly on the company and the product.
2.) AMD would loose the benefits of launching solo and have to compete in the limelight with the anticipated Penryn.
Compound these together and this would be very concerning as Barcy is critical for AMD.
Now consider, if Barcy has to launch with Penryn and Penryn outperforms Barcy. AMD will loose the precious pre-Penryn marketing opportunities AMD was shooting for and have no time to tout superiority (if it can best the top C2D Quad). Hence the serious concern. If you don't think AMD's issues are significant, well, that's your opinion. If they are of no concern why doesn't AMD just announce the problems and update the launch dates to stop the speculation? We both know the answer to that. AMD needs the speculation right now. The what if.
As for Penryn. To say the additions\tweaks to Penryn are not for speed is silly. Why would they have done them then? It certainly wasn't cost, or they would have left the design alone, (i.e. left cache the same size). You said it yourself "which allows for the larger cache, and possibly greater clock speeds". Well, what is it? Of course it is to improve the performance of the architecture. And as we have seen from early benches thus far, the Penryn will be a nice improvement. Not only the 5-10% (not made up) derived from the cache and other enhancements, but the very significant enhancements from SSE4 supported applications (see the early alpha version of DivX performance doubling). Of course the higher clock speeds will only be, well, for speed as well. And of course as you stated, Intel's manufacturing prowess and it's current architectural designs have very long legs in regards to clock speed. The combination of the enhancements and increase clock speeds will make for another winner for Intel.
You misunderstood defters comments regarding AND's execution issues. While there are a few unknowns with Penryn, defters comments are in line with what has already been established, and curiously, with assertions you are making. Increased performance from architectural updates and increased clock speeds.
TA152H - Tuesday, June 12, 2007 - link
Are you sure that's the worst case for AMD, or Intel, if they release at the same time?Problems with releasing a product Intel can't release isn't that big a deal. Intel conceded it was too difficult to release a native quad core. AMD didn't, maybe they should have :P.
Penryn is a nothing release, it's only discussed by people that are clueless. The average person will not even know about it, Intel isn't changing the name like AMD is.
Barcelona will do extremely well in the quad server space, and potentially in the eight socket space if they change the socket and someone decides to sell them.
I never said the tweaks to Penryn were not for speed, so stop acting like you can't comprehend English and misrepresenting me. They are very minor changes, and will not afford a 5-10% IPC on most applications. SSE4 is not important for most applications, and the cache has nothing to do with it being Penryn, it's about the die shrink. Also, if the cache is bigger, it might have more latency, so, it's not a free lunch.
Intel already has a winner with Conroe, and an increase in clock speed is speculative at this point, not a certainty. We'll see soon enough. A shrink makes it a lot cheaper to make, so, it's not necessarily about clock speed. It was, but lately, it's been dicey.
The problem with those assertions are they are overblown and misrepresented, like they are a big change like Barcelona is. It's not even close.
And the whining about AMD is what gets on my nerves. Oh, this is so bad, etc... It's weak minded and completely overblown. The Barcelona will be fine, people just need to be patient and accept AMD just bit off more than they, or Intel, could chew. The difference is, Intel knew it.
CobraT1 - Tuesday, June 12, 2007 - link
Are you sure that's the worst case for AMD, or Intel, if they release at the same time?Ummm, comparing what is known thus far, worse for AMD. Intel is happy with their product and have made it available for preview 5-8 months prior to launch. AMD is not and keeping their product under lock and key a month or so prior to it's supposed launch.
Problems with releasing a product Intel can't release isn't that big a deal. Intel conceded it was too difficult to release a native quad core. AMD didn't, maybe they should have :P.
Intel can't? What? To difficult? That's drama queen baloney and you (or you should) know that. Intel CHOSE not to release a native at .65 for 4 reasons. 1.)Cost 2.)Cost 3.)Cost 2.)It makes little difference performance wise (please not, Intel, AMD nor I ever said it has no performance difference, only little). AMD has stated that they probably made a MISTAKE by choosing to go straight to native. And everyone agrees with them.
Penryn is a nothing release, it's only discussed by people that are clueless. The average person will not even know about it, Intel isn't changing the name like AMD is.
Your opinions not shared by all. It is mainly a die shrink with some enhancements to boost performance, scale speed and increase manufacturing efficiencies. It's not a new architecture like Core or Barcy, but far from a nothing release. No one said Penryn was a new architecture, a revolutionary release or anything resembling as such. You are the one who brought that topic up.
I never said the tweaks to Penryn were not for speed, so stop acting like you can't comprehend English and misrepresenting me. They are very minor changes, and will not afford a 5-10% IPC on most applications. SSE4 is not important for most applications, and the cache has nothing to do with it being Penryn, it's about the die shrink. Also, if the cache is bigger, it might have more latency, so, it's not a free lunch.
Blah blah blah I comprehend just fine. You are the one who misunderstood defter. I just explained what he was saying to you. What I do question is your ability to access preview hardware or comprehend micro-architecture design if you can't extrapolate what is known about C2D, couple that with data garnered from very early Penryn previews and form at least a few decent assumptions about what Intel with have to offer. Hell, 2 people have already told you. Yet, everyone is an idiot in your view. Oh, well.
Intel already has a winner with Conroe, and an increase in clock speed is speculative at this point, not a certainty. We'll see soon enough. A shrink makes it a lot cheaper to make, so, it's not necessarily about clock speed. It was, but lately, it's been dicey.
Yes they do and yes, it is all speculative, and you have been doing it. As you stated earlier, C2D has long legs and can overclock like crazy and this coupled with a die shrink we should all expect Penryn to scale in stock clock speed. And we have said as much.
The problem with those assertions are they are overblown and misrepresented, like they are a big change like Barcelona is. It's not even close.
No, big or little change with Barcy is irrelevant. If it doesn't perform, a big change in it's architecture is all moot. Who will care about the design? AMD would not be the first company to design a nice product yet fail to execute and deliver to the customer. Oh, and it would not be AMD's first failure to deliver either. Old Sanders had a few blems grace his resume during his tenure that almost cost the company.
And the whining about AMD is what gets on my nerves. Oh, this is so bad, etc... It's weak minded and completely overblown. The Barcelona will be fine, people just need to be patient and accept AMD just bit off more than they, or Intel, could chew. The difference is, Intel knew it.
It doesn't bother me at all, one way or the other. Barcy may very well end up being fine, "WE" just don't know. If the news reports and the people commenting on the news get on your nerves, then ignore them (who's week minded?). That is the mature thing to do, as until you have evidence to prove otherwise, AMD's current press does not bode well in instilling confidence in it's ability to hit it own committed benchmarks. Remember, no one claimed AMD's roadmap but AMD. Only AMD can fix it's issues, no one else.
Ok, I give, what does "AMD just bit off more than they, or Intel, could chew. The difference is, Intel knew it." mean?
TA152H - Tuesday, June 12, 2007 - link
You missed the point about Barcelona and Penryn. Barcelona is a big release, and it will make Penryn look small if they release the same time. The only problem is, can AMD compete so long with the antiquated K8. If they release at the same though, Penryn will get buried.Actually, Intel stated in an interview that they found the native quad core extremely difficult. I forget the exact wording, but they were asked to comment on the Barcelona, and the guy said he thought it was a really difficult thing to do, even for us". So, no, they couldn't do it well either. But, they were smarter and realized it. It wasn't about cost either, it was time to market, and possibly performance. Performance, clock normalized, is going to be lower, of course, but you can generally clock higher because not all four cores have to hit the same clock speed. As far as performance clock normalized, it's not a small penalty though, or were you unaware that even on the same processor, each pair has to communicate with each other over the FSB. That's bad, really, but maybe the clock speed makes up for it? I don't know. It's not trivial though.
Actually, people keep talking about Penryn like it's so significant, but it isn't. Oh boy, Barcelona better come out befor Penryn, or they're doomed!!!!! AGGGGGGGH. It's absurd, like once the big bad Penryn comes out all is lost for AMD, so they better beat it to market. Nonsense. Intel is a different company, they don't improve manufacturing on the same process the way AMD does. So, sure, they'll have Penryn, and AMD will continue to improve their manufacturing on .65 and probably add cache (which is why they added latency to the L2 after all). So, Penryn isn't going to present an insurmountable problem for AMD. Unless AMD seriously screws up, they'll own Intel on servers, Penryn won't change that at all.
Most of the people that tell me stuff, like you, have no background in what you're talking about. Everyone has opinions, but not everyone has valid ones. 5-10% clock normalized is not trivial at all, and the changes to Penryn are trivial in most applications. It doesn't add up. Barcelona does. But you can't tell the difference, huh?
You're suffering from a logical problem. If pigs have wings, they would fly (maybe not, chickens don't), but you don't know how Barcelona performs. So, you're starting from an unknown and extrapolating from it. What we do know is the K8 performs reasonably well, and we know how many of the changes on the Barcelona have worked on other processors, and can approximate what the performance should be. It's not all smoke and mirrors, it's like adding 50 horsepower to a car. You have an idea how fast it will go, but maybe not precisely. If you knew more about their design you'd realize it is very conservative and very predictable, and it will deliver. It's not radical like the Itanium or P7 where you're scratching your head wondering what it will do because it's so unique. Sanders had some problems, mainly with the K5 which was a radically new thing for AMD, but he was smart enough to buy NexGen and thus the company succeeded. Also, keep in mind, AMD is still alive. Can you name any other Intel competitor that is?
clairvoyant129 - Wednesday, June 13, 2007 - link
"<i>You missed the point about Barcelona and Penryn. Barcelona is a big release, and it will make Penryn look small if they release the same time. The only problem is, can AMD compete so long with the antiquated K8. <b>If they release at the same though, Penryn will get buried.</b></i>"How do you know this? Did Henri show you some benchmarks at Computex? Oh wait, there was none except that pathetic Cinebench result that showed an aging Kentsfield fragging Barcelona.
Pathetic, I love how you're just making **** up.
CobraT1 - Wednesday, June 13, 2007 - link
You missed the point about Barcelona and Penryn. Barcelona is a big release, and it will make Penryn look small if they release the same time. The only problem is, can AMD compete so long with the antiquated K8. If they release at the same though, Penryn will get buried.Oh, I got your point. You are excited. It is a big release for AMD and Penryn will only look small if AMD is able to deliver and deliver C2D like domination over the Penryn. Per the reports we are seeing and AMD’s actions it is doubtful that this will be the case. I get it, you believe that Barcy will rock the PC world. Not likely at this point.
Actually, Intel stated in an interview that they found the native quad core extremely difficult. I forget the exact wording, but they were asked to comment on the Barcelona, and the guy said he thought it was a really difficult thing to do, even for us". So, no, they couldn't do it well either. But, they were smarter and realized it. It wasn't about cost either, it was time to market, and possibly performance. Performance, clock normalized, is going to be lower, of course, but you can generally clock higher because not all four cores have to hit the same clock speed. As far as performance clock normalized, it's not a small penalty though, or were you unaware that even on the same processor, each pair has to communicate with each other over the FSB. That's bad, really, but maybe the clock speed makes up for it? I don't know. It's not trivial though.
This tells me you have no idea what you are talking about and makes me want to lean towards ending this ridiculous waste of time. I will just say that wherever you read that it was false. AMD is NOT doing something Intel could not. Intel made a business decision plain and simple. If you have any understanding of processor design and manufacturing you would understand the benefits to using a dual die 4 core setup, especially at .65. Think about it, it is Intel your talking about. Unless you think Intel is a group of incompetent boobs (yes they made a mistake with netburst) who has been just plain lucky for 35 years, what you thought you heard makes zero sense. Hence, AMD admitting making the wrong choice. It was about cost, it’s called efficiency in manufacturing. Time to market was a non-issue as AMD had no quad core plans until well into 07’ and everyone knew it. Performance was a non issue with C2D yields at speed very high. The very the worst core would perform at least at the lower end of C2D’d line up in quantity. The rest of the drivel in this paragraph just makes me tired. Bottom line, none of your assertions hold water.
Actually, people keep talking about Penryn like it's so significant, but it isn't. Oh boy, Barcelona better come out befor Penryn, or they're doomed!!!!! AGGGGGGGH. It's absurd, like once the big bad Penryn comes out all is lost for AMD, so they better beat it to market. Nonsense. Intel is a different company, they don't improve manufacturing on the same process the way AMD does. So, sure, they'll have Penryn, and AMD will continue to improve their manufacturing on .65 and probably add cache (which is why they added latency to the L2 after all). So, Penryn isn't going to present an insurmountable problem for AMD. Unless AMD seriously screws up, they'll own Intel on servers, Penryn won't change that at all.
Penryn is not “so significant” yet it is significant. It should boost IPC and scale clock speed which will increase it’s overall performance. C2D is already the performance leader by a large margin and it’s release will extend it performance envelope. Oh and lets not forget, it will be much cheaper to manufacture. Penryn is 25% smaller even with it’s additions. Barcy better come out and be everything AMD has touted it to be or even without Penryn, AMD will be in deep, deep trouble. This not about Barcy vs Penryn, Intel products are a known quantity. No, this is about AMD and it’s ability to deliver. They are bleeding money big time and need a success. This is why people are concerned.
Most of the people that tell me stuff, like you, have no background in what you're talking about. Everyone has opinions, but not everyone has valid ones. 5-10% clock normalized is not trivial at all, and the changes to Penryn are trivial in most applications. It doesn't add up. Barcelona does. But you can't tell the difference, huh?
I will do you a favor and ignore this one.
You're suffering from a logical problem. If pigs have wings, they would fly (maybe not, chickens don't), but you don't know how Barcelona performs. So, you're starting from an unknown and extrapolating from it. What we do know is the K8 performs reasonably well, and we know how many of the changes on the Barcelona have worked on other processors, and can approximate what the performance should be. It's not all smoke and mirrors, it's like adding 50 horsepower to a car. You have an idea how fast it will go, but maybe not precisely. If you knew more about their design you'd realize it is very conservative and very predictable, and it will deliver. It's not radical like the Itanium or P7 where you're scratching your head wondering what it will do because it's so unique. Sanders had some problems, mainly with the K5 which was a radically new thing for AMD, but he was smart enough to buy NexGen and thus the company succeeded. Also, keep in mind, AMD is still alive. Can you name any other Intel competitor that is?
You are suffering from not knowing what you are talking about. I will just end by commenting on your analogy since you also feel you are an expert on motor design. You cannot add 50hp to a car. You have to modify or redesign different areas to improve said motor to create more power. (don’t even get me started on cheats like NOS, doesn’t apply) The performance profile can change at certain load\speed intervals or across the entire range depending how you modify the motor. You can also increase power at certain load\speeds while reducing power output at other load\speeds. Of course, you can modify the motor with a bad combination of the best designed parts on the planet that do not work well together and your performance profile will drop. That is why there are dyno’s, to track changes and optimize the power profile you are trying to achieve. Engines, just like processors, have to have components designed that make the hole that work well. That is what we are all waiting to find out with Barcy. Thus far, the news has been less than stellar so close to AMD’s scheduled launch.
CFKane - Wednesday, June 13, 2007 - link
He wasn't saying that Intel couldn't do it, but there actually was an interview on digitimes where the Intel representative indeed admitted that "even they" would have difficulties to bring native quad core to market:
http://www.digitimes.com/mobos/a20070608PD206.html">http://www.digitimes.com/mobos/a20070608PD206.html
Penryn won't boost IPC since you would need architectural changes for that. It will come with increased clock speeds at the same or lower TDP, which is mainly due to the 45nm shrink. Some of that potential also goes into the larger L2 caches, so it remains to be seen how far they can push the clock speed without hitting the power consumption wall (again).
CobraT1 - Thursday, June 14, 2007 - link
Actually he said "Intel can't release" then he backed off to "difficult". So, difficult how? What did the Intel rep say? "The technology is much easier, the product has higher yields and performance is almost the same as the native quad-core processor." The technonlogy he is refering to is manufacturing technology. Native is difficult to manufacture in quantity to make it economically advantageious at .65. That doesn't mean they can't it means they chose not to for business reasons. That was my contention with his "Intel can't" comments.As far as Penryn IPC and ability to scale, you do not need a new architecture to increae IPC. You can improve the architecture and improve IPC. You can also made modifications that decrease IPC of an architecture. I won't go into all of the reasons or rehash above discussions, I will just go this route. As I don't actually have a Penryn myself, I will defer to this article (which will answer your questions) and you can go there and debate it.
http://www.anandtech.com/cpuchipsets/intel/showdoc...">http://www.anandtech.com/cpuchipsets/intel/showdoc...
"It has also been confirmed that Penryn will deliver higher IPC and higher clock speeds. Intel wouldn't say more than "more than 3 GHz", but considering that the FSB is bumped up to 1600 MHz, 3.2 GHz is likely. However, several Intel people confirmed that if necessary ("depending on what the competition does"), the 45nm CPUs can go quite a bit higher (3.6 GHz is probably a safe estimate, considering how far current Core 2 CPUs are able to overclock)."
CFKane - Thursday, June 14, 2007 - link
Nobody with the most basic sense of logic would make a claim in that simplicity regarding the technological challenges. There is no wizardry involved in chip making and Intel can do everything AMD can. Business considerations are just as important as technological viability for such large scale enterprises, though, and Intel has touted the native quad-core approach on the 65nm node "suicidal" very early. More realistically they should have said "this wouldn't allow the gross margins our shareholders expect". But in the end it just means the same for them: They can't.
The question of if AMD can or should is a different one. I, as a customer interested in the technology, am glad they're doing it. I just wouldn't buy any stock right now.
Okay, I was perhaps a little vague there, but I said you need architectural changes for that, not a new architecture. Penryn brings a few changes and some (like the Radix 16 divider) do increase IPC in certain limited areas. Higher FSB clocks may also improve IPC slightly. Both should hardly be noticable though, except in benchmarks that exploit those improvements heavily - Intel hasn't included the DivX alpha benchmark without reason in their presentations. But this doesn't deserve to be called a "boost in IPC" in my opinion. You can call it that when you add more execution units or other improvements that increase IPC across the board. Conroe surely deserved that label. And from what we know, Barcelona also deserves it.