Remember when I called out AMD’s John Fruehe for spreading lies regarding the inflated performance expectations of Bulldozer?
AMD is now lowering expectations for Bulldozer’s performance. Originally, Fruehe claimed a figure of 50% in an official AMD blog:
From a performance standpoint, if you compare our 16-core Interlagos to our current 12-core AMD Opteron™ 6100 Series processors (code named “Magny Cours”) we estimate that customers will see up to 50% more performance from 33% more cores. This means we expect the per core performance to go in the right direction — up. That is all I will say until launch.
AMD’s Thomas Seifert is now giving some more realistic figures:
The Interlagos platform is our first server offering, optimized for today’s cloud datacenters and the architecture excels at compute-intensive and HPC workloads, where it will deliver up to 35% performance improvements compared to our current offerings.
So there we have it, I told you John Fruehe was lying about Bulldozer performance, now AMD has officially confirmed this. Sadly it also doesn’t spell that much good for Bulldozer… Up to 35% more performance from 33% more cores is not all that spectacular, for an entirely new architecture.
Update: Some people came up with the ‘explanation’ that faster AMD processors had come out between Fruehe’s and Seifert’s claims. Ofcourse they assume that I could not think of that myself, and try to do the math to prove that Fruehe was not lying.
Problem is, I DID think of that, I just discarded the idea because the CPUs had not improved that much. I have done the REAL math in one of the comments below, but I will repeat it here, because people may overlook it:
Try these benchmarks for example: http://www.cisco.com/web/DK/assets/docs/presentations/vBootcamp_Performance_Benchmark.pdf
They include some 6176SE vs 6180SE numbers… guess what? It doesn’t add up!
The 200 extra MHz and the slightly faster memory don’t add enough performance to support your theory.Basically you have this to solve, if both statements mean the same:
X * 1.5 = Y
(X * Z) * 1.35 = YSo:
X * Z * 1.35 = X * 1.5
Z * 1.35 = 1.5
Z = 1.5/1.35 = 1.11In other words, for Seifert and Fruehe to say the same, assuming one is taking about the 6176SE and the other about the 6180SE, the 6180SE has to be 11% faster than the 6176SE.
Even with a purely synthetic test such as Dhrystone, it cannot get the required ~11% improvement. It is stuck at ~8%
In the still very synthetic SpecINT, it is only ~5% faster. Nowhere near fast enough.
If we take the 5% of SpecINT, and assume this is one of those HPC scenerios they are talking about, and substitute that, we get:
1.05 * 1.35 = 1.42.
So that would still make Fruehe’s estimate off by almost 20%.
Quite simply… if you cannot make accurate predictions about performance more than a year ahead of time, then just DON’T make predictions!
Scali's OpenBlog™ 
Guess it will all depend on price point. It will probably still do fairly well. I just wish we could see some damn legit examples now on how well it performs.
I’m never worried about price point with AMD. They’ve shown that they are willing to ride out their products so long that they practically have to give them away for free, in order to maintain decent price/performance.
But price/performance is a metric I’m only marginally interested in. AMD is suffering performance/watt by milking out old architectures for too long, and the absolute level of performance is further and further away from the high-end as dictated by Intel.
I’d like to see CPUs with better absolute performance and better performance/watt. So I’d like to see what AMD can improve there with their new architecture.
But if even AMD is watering down their own highly synthetic best-case performance figures from 50% to 35%, things are looking bleak.
Aside from that, I’m pretty tired of AMD’s false marketing. They’ve made claims of 40% better performance with Barcelona, when in reality it was nowhere near. They’ve made claims about OpenCL, literally YEARS before they released their first driver, not to mention physics acceleration, which they still don’t have. They’ve made false claims about why their tessellation isn’t as fast as nVidia’s… and Bulldozer’s claims were also too good to be true, I already knew that a year ago. I wouldn’t be surprised if even this 35% figure is too rosy.
Sure, all companies like to present their products in the best possible light, and you should take their words with a grain of salt, but I can’t think of any other company that uses such boldfaced lies in their marketing over and over again. There are countries where such false advertisement is simply illegal. AMD actually got sued once, for airing a commercial that claimed that their 64-bit processors would prevent viruses (nX-bit). They had to take it off the air, and pay a fine.
You conveniently neglected the “compute-intensive and HPC workloads” part of Seifert’s sentence.
Explain how that would be relevant?
(I also didn’t explicitly say that John Fruehe and his blog are about the server/workstation products… which means compute-intensive/HPC workloads are a given. Oh wait, did you conveniently neglect that? Guess that makes us even… Oh no, wait, that means that nothing about my blog post changes, and your response is useless).
JF’s statement was obviously an average about server workloads in general. It’s perfectly possible for this average to be 50% and for the sub-average pertaining to HPC to be around 33%, especially considering that current Magny-Cours Opterons are a tad faster than they were when he made the original statement.
And since HPC is FP-heavy while Bulldozer is somewhat biased towards integer workloads, it’s not even surprising.
‘Obviously’, yes… whatever makes you sleep at night, right?
Logic is what makes me sleep at night, so let me apply logic:
His formulation of ‘up to’ indicates that he is not talking about averages, but rather a rough upper limit of performance. Using the words ‘up to’ when talking about an average makes no sense, because of the nature of averages.
Seifert uses the same ‘up to’ formulation, so they are talking about the same thing: peak performance rather than average.
More logic is that 50% or even 35% more performance on average from one architecture to the next is not something we have ever seen before. Therefore it is highly unlikely that Bulldozer is THAT good. No architecture update in the history of computers was ever THAT good.
Some more logic you have already given yourself: the nature of Bulldozer’s design is not that favourable to FP workloads… so all the more unlikely.
You’re right, I didn’t notice the “up to”, which means he wasn’t talking about averages, but the best-case scenario. My bad. The argument still stands, though: it’s perfectly possible for the best-case scenario to involve a 35% improvement in HPC and 50% when considering all types of server workloads.
Also, I don’t know if Core 2 was 50% faster than Netburst on average, but it couldn’t have been far. At least there were many applications where it was well over 50% faster, even though it was made on the same 65nm process. Bulldozer, on the other hand, will be introduced along with a new process (compared to Magny-Cours, admittedly not Llano) featuring not only smaller transistors but also HKMG and power gating.
Perhaps it’s unreasonable to expect a 50% performance bump on servers; then again, Intel’s best quad-core is about 50% faster than AMD’s. Is it so crazy to think that a 4-module Bulldozer could reach the same level? Of course that doesn’t necessarily mean Interlagos can be 50% faster than Magny-Cours, but that was my misunderstanding, not JF’s statement.
PS: I assure you that John Fruehe’s claims about future Opteron performance have no impact on my sleep. 😉
Well, I have some more logic for you: Why would AMD go from using the 50% figure to using the 35% figure in promoting their products? Doesn’t make sense, as the 50% figure sounds much better.
I don’t buy the theory that 50% was not about HPC. I will even give some extra logic for that: HPC is generally the most favourable when it comes to performance scaling. As things get less computationally intensive, the bottleneck shifts from CPU to other components such as memory and disk I/O. Clearly AMD’s new CPU is not going to give a 50% improvement in memory and disk I/O performance. It is not likely to have any effect at all. So if anything, if one figure is HPC-related, then the other figure must be lower, not higher.
As for Core2 and Netburst… that’s a very long story, which we shouldn’t have to get into as it clearly has no relation to this situation.
The crazy part in Bulldozer performance expectations is in the fact that Bulldozer cuts down on execution units per core, compared to AMD’s current architecture.
It is a lot easier to see why Core2 was such an improvement over Pentium 4 than it is to see where Bulldozer would get any performance gains at all over Magny Cours.
Going from 3 ALUs per core to 2 ALUs, sharing an FPU between 2 cores rather than each core having a dedicated FPU? Clearly these actions decrease performance. While AMD may be able to significantly improve the performance of their ALU and FPU in a generation… they are going to have to compensate for the lower amount of units first, before they even break even, performance-wise.
In fact, there are certain parallels with the Pentium 4: AMD is going to try for higher clockspeeds to compensate for this performance loss, which also means a longer pipeline (another source of inefficiency: larger penalties for cache misses and branch mispredictions).
AMD is also going to offer more cores per CPU, so like the Pentium 4, they will opt for a larger, less efficient chip compared to the competition.
This makes it all the more unlikely to see Core2-like performance boosts.
At least with Core2 it was obvious where the better performance came from, compared to Pentium 4. With Bulldozer, every sign points the other way.
There’s no theory to buy or not buy, JF was talking about servers in general, TS was talking about HPC.
Just because many server workloads are dominated by system and I/O performance doesn’t mean you couldn’t find a few heavily integer-dependent ones, where Bulldozer would be expected to shine.
Architectures are much more than just a number of ALUs and FPUs, but I’m sure you know that very well. I’m a bit puzzled as to why you feel the need to call someone a liar based on conjecture and speculation, especially when launch is only about 6 weeks away and by then we’ll know for sure.
I don’t think you get my point: Whether JF was talking about general performance or HPC is irrelevant, since HPC will likely deliver the best cases for performance in general.
I call John Fruehe a liar because he is… not 6 weeks before launch, a year before launch. I did that because I stuck my neck out for what I think is right (just like I did with Randy Allen years ago, prior to Barcelona’s launch). You think I did that just on conjecture and speculation? You’ll know that in 6 weeks.
“I don’t think you get my point: Whether JF was talking about general performance or HPC is irrelevant, since HPC will likely deliver the best cases for performance in general.”
Yes, in general. But some server workloads may exhibit higher performance improvements with Bulldozer, provided they’re not too dependent on I/Os, storage, etc., given that Bulldozer is (expected to be) better at integer workloads.
Either I don’t get your point, or you didn’t get mine.
Let me try again: If we take the whole set of workloads (‘general’), then there will be some subset that are the highest performing.
I am saying that the HPC workloads are among this subset (which means they are also in the ‘general’ set).
Server workloads tend to be relatively I/O and storage-bound, given their nature.
Unless you are saying that HPC can not have integer workloads? So fields like cryptography don’t count?…
Alexko, Seifert also says right before that that that’s where BD excels. He does make it sound like like compute-intensive and HPC (I’m reading: FP heavy) is BD’s strong suit.
(LOL at triple that, I hope that that was readable.)
That said, IIUC, those workloads are likely to run only one thread per module to harness all of the FPU, making the comparo “8 BD threads” vs “12 MC threads”. So his statement might boil down to “the FPU is 35% faster with only 66% of the cores/threads used”. So IMHO it remains partially ambiguous and open to interpretation what he means there.
I guess it is difficult to see the different quotes in their original timeline. Fruehe’s comment was made when the fastest MC was running at 2.3GHz. The statement from Seifert was made when the fastest MC was/is running at 2.5GHz. If the target speed of Interlagos did not change in the mean time (do you have any information/proof it went up?), it is no more than logical that the expected increase in performance lowers, compared to the first statement. In that light, calling Fruehe a liar for that could be seen as nothing else but spinning the actual truth for personal or corporation (which one would that be?) benefits.
Not so fast… first of all, my original complaint about Fruehe went much further than just his 50% claim (he also claims significantly increased IPC etc… anyway, read my earlier blog about it).
Secondly, I didn’t say that I find the 33% claim trustworthy. While less outrageous than Fruehe’s original claims, it’s probably still a highly inflated number, which can either not be attained at all in practice, or only in isolated synthetic benchmarks that don’t amount to much in the bigger picture (such as for example Intel’s claims about Penryn at the time. Yes, *some* operations related to MPEG4 encoding were over twice as fast with Penryn’s new SSE4, but encoding an entire MPEG4 movie was nowhere near as fast… and for most other applications, SSE4 did not do much at all. Then again, in Intel’s defense: they were always very clear that they were strictly talking about one pass of the MPEG4 encoding process, and were in no way trying to generalize or imply the results to mean anything more than that. And what they claimed was true).
Lastly, did you bother to do the math, even if we assume that Fruehe was talking about then-current Magny Cours, rather than the ones that would be available at introduction of Bulldozer (which we don’t know as he wasn’t very specific, nor was Seifert for that matter)?
Let’s take the 2.3 GHz MC as index 1.0.
Assuming linear scaling with clockspeed, the 2.5 GHz MC would roughly have an index of 2.5/2.3 = 1.087 (where in reality scaling will be sublinear, so it will be lower).
Now, if we take 50% performance of the 2.3 GHz CPU, we get an index of 1.5.
If we take 35% performance of the 2.5 GHz CPU, we get an index of 1.47. And again, if we take the sublinearity into account, we will at least know the quoted figure is below 1.47 times a 2.3 GHz MC.
So we still see AMD lowering its claim.
As for personal or company benefits? Really? What do you think Fruehe and Seifert’s objectives are?
“As for personal or company benefits? Really? What do you think Fruehe and Seifert’s objectives are?”
Umm, wild guess: to benefit their company ? since, in the case that this hasn’t been covered, they’re EMPLOYED by AMD ?
Heaven forbid they say positive things about their companys product……
I realize I may have been a bit subtle. I will not make that mistake again:
Clearly Fruehe and Seifert are employed by AMD, and they are trying to market a product.
My point is: Why is this apparently NOT a problem, while on the other hand my motives are being questioned in order to discredit what I say, or at least to take what I say with a grain of salt?
Whatever any AMD employee says should *always* be taken with a healthy helping of salt, by default (or employees of any other company when talking about their products, for that matter).
But apparently people have forgotten this already.
As for my motives… clearly I have no ads on my blog, so I have no monetary benefit from getting hits.
I am not affiliated with any company, and am not trying to promote any company or its products. Likewise, I have no interest in discrediting any company or its products… other than just getting information out there, as an independent developer and industry insider.
But sure, try to discredit *me*, rather than the message, or the company involved. That’s smart.
Not so fast there.
Freuhe clearly stated that the ‘up to 50%’ increase was compared to their ‘current (at that time 2.3GHz MC)’ fastest 12-core offering.
Seifert clearly stated that the ‘up to 35%’ increase was compared to their ‘current (at that time 2.5GHz MC) fastest 12-core offering.
Handily you already made the calculation what the difference means, going from linear scaling. You find it to be a 47% increase, which is close enough to Fruehe’s original blog entry, as most often you don’t see the ‘47%’ perfomance increase, when ‘up to’s’ are in play. It is an understandable margin.
So, I still fail to see how AMD is really decreasing it’s performance expectation, based on those two seperate statements. The only thing I see is that you more or less try to claim that Fruehe claimed a ~30% IPC increase per core over MC. Unless you can offer factual claims of such, I can’t agree with your statements. For the sake of accuracy, I’d advise you to read the linked and quoted texts again yourself, as you will see that statements are made as the situation was at the moment of statement.
Right, so don’t read my other blog entry when I tell you to.
Just make some more false assumptions so you can put me in a bad light.
See, if you DID read my other blog, you would know that it is not 30% IPC, but 17% IPC that Fruehe is claiming, and you’d also see that I have provided a direct link to where he said it.
Which, as you will see in 6 weeks time, is incorrect.
Also, apparently you missed what I said about non-linear scaling: The 2.5 GHz MC is not actually 1.087 as fast as the 2.3 GHz MC. It is less. Hence, the actual figure will not be 47%, but less. How much less depends on the exact mix of benchmarks you base it on, but the result would be something not close enough to 50% to just ignore it (even 47% would already be closer to 45% than to 50%, so if we are talking about rounding…).
Really, if you want to discuss things, try to put some effort into them. I can’t stand slackers.
Try these benchmarks for example: http://www.cisco.com/web/DK/assets/docs/presentations/vBootcamp_Performance_Benchmark.pdf
They include some 6176SE vs 6180SE numbers… guess what? It doesn’t add up!
The 200 extra MHz and the slightly faster memory don’t add enough performance to support your theory.
Basically you have this to solve, if both statements mean the same:
X * 1.5 = Y
(X * Z) * 1.35 = Y
So:
X * Z * 1.35 = X * 1.5
Z * 1.35 = 1.5
Z = 1.5/1.35 = 1.11
In other words, for Seifert and Fruehe to say the same, assuming one is taking about the 6176SE and the other about the 6180SE, the 6180SE has to be 11% faster than the 6176SE.
Even with a purely synthetic test such as Dhrystone, it cannot get the required ~11% improvement. It is stuck at ~8%
In the still very synthetic SpecINT, it is only ~5% faster. Nowhere near fast enough.
If we take the 5% of SpecINT, and assume this is one of those HPC scenerios they are talking about, and substitute that, we get:
1.05 * 1.35 = 1.42.
So that would still make Fruehe’s estimate off by almost 20%.
You are making no sense. All of the big clients who would potentially purchase these chips by the thousands, will of course, see and test hard silicon before purchase. What is JF going to do, Lie and hope that no one notices in the benchmarks ? The only spin here is yours, and it’s sad really. Of all the pretend battles to pick, CPU speeds ? really ?
I should just trash replies like these, but oh well… You fanboys can’t help it, you just cannot think.
First of all, spin? Baseless accusations, nice going!
Secondly, indeed, what is JF going to do? Why does JF have this blog on servers and workstations anyway? Why is JF so actively participating in online forums, which have nothing to do at all with his potential clients?
We don’t see Intel putting anywhere near as much work into marketing their server products, let alone that Intel employees actively participate in (consumer/enthusiast-)forums to try and convince them how good their server products are.
What *is* JF doing?
I suppose you never heard of the Halo effect.
Besides, AMD has a history of such spin. Randy Allen did almost the same thing with Barcelona (hey again, a server product!), he even posted YouTube movies explaining how their chip was going to be 40% faster than the best x86 server chips out there. Why would he do that? (YouTube, really? What potential big server client would go to YouTube to get information on server products? And even if they did go to YouTube, what client would be dumb enough to buy Randy Allen’s pathetic marketing dribble? Only consumers/enthusiasts are dumb enough to buy it. Not people like me)
So yes, he did lie…. I’m not sure if he hoped that no one noticed, but obviously everyone did. I think he didn’t care. Apparently you can get away with stuff like this. Because people like you are dumb enough for it.
By the way, try again, CPU speed is not the issue here (I could care less which company has the faster CPU). Try to figure out what the REAL point is that I’m making here.
You dumb ass
bulldozer will have around 50% better multitasking
with around 33% better IPC
this is what it means throughoutput means more about multitasking
HPC is all about IPC
Scali, if this is a legit site you need to be taken off of it, this is a load of crap JF said before you cant compare throughoutput with speed!!!! i’ll say it again throughoutput does not mean faster just means you can do MORE.
You can’t say a truck is faster then a sports car just because it has more Towing Capacity,This is what you just did!!
you will say other wise but then again your not a CPU engineer now are you?
Wow, talk about getting everything mixed up!
The ‘throughput’ thing is something I’ve seen a lot… but did you actually READ the quote above? He says 50% performance, not 50% throughput. There are some AMD slides that say 50% throughput… so either Fruehe says that performance == throughput, or he is wrong about saying 50% performance (he has been inconsistent with AMD slides on more than one occasion…).
He also didn’t say 33% better IPC. He said 17% better IPC.
The 33% was about the extra number of cores (Magny Cours having 12 cores and Bulldozer having 16).
Again, read: “..we estimate that customers will see up to 50% more performance from 33% more cores.”
That is what he said.
Haha, how do you feel about that post now? 🙂
How are the boys at AMDZone doing? Lol!
“Why is JF so actively participating in online forums”
Because unlike 99% of Intel employers he gives a shit. then people like you come in and talk crap about it.
We will know who was right at launch no? i have never been on this site before but i thank i’ll stay on here until bulldozers launch some time in September!
Just made a new friend Scali, please don’t cancel this pathetic attempt at a blog! i’ll post this site on many different forums and we will all be waiting with you until launch!
I don’t think you quite got my point: John Fruehe is the Director of Product Marketing for Server/Workstation products at AMD. Why would anyone try to market server/workstation products on consumer forums? That’s not their target market. You think these forums are where the people who buy server/workstation hardware hang out? I don’t, not in a professional capacity anyway. Forums may be a reasonable source for buyer information on gaming systems and such, but servers and workstations are a very specific market, where each client has their own specific needs.
What Fruehe and other AMD marketing people are doing on online forums is biasing people towards their products/brand. Intel cares just as much about making money and selling products as AMD, they just don’t stoop as low as joining forum discussions (which is safe, because 99% of forum users won’t be able to spot the misinformation spread by Fruehe et al anyway). Same for nVidia, you don’t see their employees on forums, where the AMD graphics guys do participate on forums.
In my opinion it is ‘not done’ for employees of any company to participate in forum discussions, promoting products. Forums are for regular consumers and enthusiasts. They are not a place to conduct business or do any kind of marketing. What would you feel like if you were having a discussion on PCs in a bar, and some AMD employee would butt in and explain how ‘great’ their products are? Bars are not the place for that sort of thing, same with forums.
What’s even worse: most people actually believe everything they say… which is ironic, since they tend to scream ‘bias’ and ‘shill’ to any non-employee.
The way you compare the Phenom architecture vs the Bulldozer architecture is just stupid. Going from 3 ALU’s to 2 ALU’s is obviously gonna decrease performance boo boo boo…
SHUT UP!
3 AGU’s and 3 ALU’s on Phenom were kept to maintain parity between execution units, they never worked as 3 ALU’s and 3 AGU’s, but rather as 1.5 ALU and 1.5 AGU.
Bulldozer with 2 ALU’s and 2 AGU’s will be definately more efficient than previous micro-architecture. That’s 33% more vs Phenom as it also has 4 integer pipes, vs 3 on the Phenom. Again 33% more integer execution units.
Wow, you really believe that?
I didn’t realize my blog would attract so many clueless fanboys.
I am an assembly programmer, and my blog is mainly aimed at people with half a clue about software design, CPU architectures and low-level optimization. If you want to try and snipe, at least make sure you know your stuff, because I do (which is probably something fanboys such as yourself are completely unfamiliar with… you have to be pretty clueless about CPU design in the first place to be an AMD fanboy, or any kind of x86 fanboy for that matter).
Anyone can tell you that ever since the K7, AMD has had 3 ALUs and 3 AGUs (this was actually one of the strong points over the Pentium Pro/II/III architecture, which had only 2 ALUs… pretty much the reason for the higher IPC… something you should know if you are pro-AMD).
Heck, even a Pentium Classic had two ALU pipelines, capable of 2 int operations per cycle. Fruehe is basically saying that K7/K8/K10 are worse than a Pentium Classic, and you believe that?
Read Agner Fog’s documents, if you don’t believe me: http://www.agner.org/optimize/
His microarchitecture.pdf will explain it:
Sure, you couldn’t execute 3 ALU instructions *and* 3 AGU instructions every cycle, since the CPU could only dispatch 3 micro-ops per cycle. But that’s not the point. Of these 3 micro-ops, 3 of them could be ALU, or 3 of them could be AGU.
Bulldozer will not be able to execute 3 ALUs or 3 AGUs per cycle in any case. While current AMD architectures (and Intel for that matter) can.
My personal experience is that you generally have more ALU instructions than AGU instructions in the mix, so having 2+2 is less ideal than having 3 for either. Which is probably why AMD designed the K7 with 3 ALUs in the first place, and why Intel adopted the 3-ALU approach in their later architectures as well (since Core2… and Pentium 4 is a bit of a special case, since although it has 2 ALUs, they can execute 2 instructions per cycle for the most common ALU instructions, so it is virtually a 4-ALU CPU). Now that 3-ALU CPUs have been around for quite a while, and software is optimized for that, it doesn’t seem like a good idea to go back to 2 ALUs… unless you want to create smaller cores, which is probably the underlying reason for Bulldozer: cut down core logic, so you can fit more cores on a die. It’s not the MHz-myth, it’s the core-myth.
And the bigger problem is one FPU shared by 2 cores… which you conveniently decided to ignore.
Ah, I see where you got the 1.5 number: http://www.xtremesystems.org/forums/showthread.php?257927-AMD-s-Bobcat-and-Bulldozer&p=4523917&viewfull=1#post4523917
It’s another one of Fruehe’s lies! And you fell for it! Trying to prove that Fruehe is not lying by using Fruehe’s own lies doesn’t work, obviously.
Read Agner Fog yet?
(It is irrelevant that the ALU/AGU units may be shared, since the architecture can only dispatch 3 ops per clk at most. However, all 3 ops can be ALU, or all 3 can be AGU. That is not possible with BD).
Guess who was right? Indeed, not John Fruehe! You guys look so dumb now!
Pingback: Yes, AMD fanboys *are* idiots | Scali's blog
“He said 17% better IPC.”
He never said that and he even said you can do the math that way.
just because there’s 33% more cores does not mean they scale perfectly or did you even thank that?
He *did* say that (actually he said “a lot more than 17%”), I linked to a post of his in my earlier blog on Fruehe’s lies (you know, the one I linked to in this one, try clicking the links so that I don’t have to bother with useless comments such as this): http://forums.anandtech.com/showpost.php?p=30364913&postcount=280
Satisfied?
“What Fruehe and other AMD marketing people are doing on online forums is biasing people towards their products/brand.”
Funny i never seen that, i only herd from JF though. He has never seemed bias towards anything. He never says buy this or that NEVER!!! he is on the forums to give people the RIGHT idea about bulldozer design. he is on their with his OWN TIME he does not get paid for doing it. he even said he goes on bike forums is that because he is trying to sell AMD products?
Well, since you are already convinced that the idea that Fruehe spreads on Bulldozer’s design is the ‘right’ one, there’s no point in trying to argue, is there?
I’m convinced that his information is incorrect, and I have pointed out flaws in his reasoning time and time again, both on the forums and on this blog.
“Forums are for regular consumers and enthusiasts. They are not a place to conduct business or do any kind of marketing.”
Funny i never knew a server guy cant be a enthusiast?
He goes on their to help people understand bulldozer its just a thing he enjoys doing he works all the f**king time and then assholes like you come on and put him down.
would you put down the president of Nintendo if he came on and talk about technical areas of their knew systems?.
You are a NVIDIA+INTEL lover and you hate on all Amd employers even when their nice people
I like nvidia a lot and i find all their business practices to be good and i like Amd a lot and i find their business practices to be good but Intel’s NO!
Did you know a lot of Intel employers do go on forum sites but the don’t say their from Intel their is even some that are on tomshardware. you know a legit site unlike your blog.
We all know AnandTech is full of Intel employers their the only site that said llano is not good unless you do gaming. when the don’t thank logical that even websites are more graphical today then 5 years ago and then they don’t have common sense to see where the next 5 years well be like in terms of graphics.
whats up with people on IPC why is that so important. i know its important but its not the future why make a whole new design that well give you at best 30% more IPC when the can give you more cores instead then add performance later?
Do you know the Average person does more then 1 thing and the things they do are so light they don’t matter! did you also know that the enthusiast market is less then 10% why spend all the company’s money on that market? Really bulldozer screams server not client. servers like facebook needs more cores not IPC. (http://www.eweek.com/c/a/IT-Infrastructure/Facebook-Tilera-Chips-More-Efficent-Than-Intel-AMD-Products-469783/)
maybe then you will understand why they made the bulldozer!
Even if a server guy is an enthusiast, I doubt he would go on a forum to do server business (since it is not a good place, as I said).
And me an Intel + nVidia lover? You have no idea, do you?
I’ve always hated Intel’s x86 architecture (I’m a Motorola man myself, if anything), and nVidia’s business practices in the past… not so good. I was totally on nVidia’s case in the past, when they were cheating in drivers, and trying to make their GeForce FX look good etc. Pretty much the same kind of thing that AMD is doing these days. See, I don’t care about the company or the products, I care about end-users getting the right information.
Oh, and AnandTech is biased because they say LLano is only good for gaming? Yea right… I guess all those other sites are afraid to tell the truth then: LLano is just a warmed-up X4 CPU with a low-end GPU stuck to it. Yes, the GPU-part is nice… *if you are gaming*… but for pretty much everything else you don’t need the extra GPU power that LLano provides over an IGP, and the CPU is less than spectacular. It’s outdated.
And IPC *is* important… It’s why Intel is currently the performance leader, and also has a much healthier business: Intel needs less cores than AMD to get the same performance, meaning they can sell smaller, cheaper, more power-efficient chips at the same prices, since performance determines price, not manufacturing costs.
” Increase for single-threaded workloads will be “a lot” more than 17%”
Do you know why because 33% more cores does not mean 33% more performance right of the bat! they don’t scale perfectly. who the hell hired you to be a software designer anyways are you still?
IF your link has don’t anything is prove you wrong why would you even post that? makes you look worse.
If bulldozer fails it wont be because of a price/performance ratio it will be because of marketing. If anything Amd needs to advertise more they have no commercialism what so ever.
What are you on about? I never said that 33% more cores means 33% more performance.
The problem with Bulldozer is mainly that not all cores are equal (similar to Core2 Quad in a way: two cores on one die, sharing cache, but the third and fourth core need to sync via FSB). We can’t say much about IPC based on overall multithreaded performance and core-count.
Aside from that: you seem to forget that other factor for performance: clockspeed.
We already know that Bulldozer is going to be clocked higher than current CPUs (top model will be 3.6 GHz with a turbo of 4.2 GHz). So AMD’s 35% performance increase is roughly dependent on these changes:
– More cores
– Synchronization logic between cores
– Higher IPC
– Higher clockspeed
So no, you don’t necessarily need more than 35% better IPC to get 35% better performance. Oh dear, did I actually show more insight in CPU architectures right in this one comment than you have in all your comments? Funny, considering your arrogant tone.
“I’m convinced that his information is incorrect, and I have pointed out flaws in his reasoning time and time again, both on the forums and on this blog.”
Do you really thank you know more then a person who worked at Compaq being a server guy for over 20 years and do you think you know more then a person who works on the future products every day?
Fruehe is a marketing guy, not an engineer. Marketing guys often get the technology mixed up.
Especially marketing guys at AMD… I’ve caught Richard Huddy on numerous mistakes over the years. Dave Baumann as well.
Yes, apparently I did know more than John Fruehe.
i really don’t thank Amd single threaded performance will match or beat Intel’s, i don’t really thank that will ever happen again. but i just don’t thank that matters i believe in the next 10 years the cpu will be used much less and the Gpu world will kick in. people will use more cores not 1 core heavily.
Back in 2005 no one ever thought they need a dual-core for another 3 years those people are the same who said i’ll never need a quad for 2 more years then now their saying i’ll never need more then a quad so i’ll wait until i need a 8 core.
Why make a design that will last you 2 years when the future keeps changing like this? i bet with my heart and soul that programs and games will use more then 4 threads. hell i though i read BF3 will use up to 12. also RE5 can use 6 threads.
My athlon II x4 has enough performance per/core to play any game out their but what about 4 years from now. i guess you think games will never use more then 4 threads.
What about when the PS4 and Xbox 3 comes out they will have 4 or more cores hell people are saying the PS4 will have the 8 core bulldozer in it.
the only 2 things i do that use my processor a decent amount is video encoding(which needs more cores ask handbrake) and gaming which like i said my Athlon II x4 runs GTA4 fine with my AMD Radeon 6950
Why the hate for Amd anyways. i am a little pissed they delayed bulldozer but i feel confident it will be out in September anyways.
The irony of it all is that the GPU is driven by the CPU, and this is mostly a single-threaded task. So fast GPUs tend to benefit from good single-threaded performance, even with GPGPU tasks.
Intel seems to see where nVidia is going, and is working on their own GPGPU (Larrabee). I think this will be the future: a CPU with a few fast cores, and a GPGPU with lots of not-so-fast cores.
Eventually they will probably merge into a single architecture again in some way.
And I don’t hate AMD. I don’t think you understand where I’m coming from. I buy AMD’s products when they’re worth buying (if you bothered to read my blog, you’d see that I bought a Radeon 5770 not too long ago… I’ve owned plenty of Radeons in the past anyway, although many of them date from the ATi era. I was an early adopter of the Radeon 8500, back when most people didn’t take ATi as a serious competitor to nVidia).
I just don’t like their marketing. They rarely tell a straight story. Always pointing to their competitors, or twisting the facts… or just outright lying about their own products… saying they have OpenCL support when they don’t… saying that tessellation is bottlenecked by triangle size, while their hardware is better at handling small triangles than nVidia’s… they are bottlenecked by their single tessellator unit’s throughput… And they claimed that Barcelona would be 40% faster than anything out there… while it was actually slower (even disregarding the TLB bug)… to name but a few things.
Sure, I can dig up some dirt from Intel and nVidia as well… Difference is: most of that is in the distant past, they both seem to behave much nicer now (and no, I’m not in the ‘wood screw’ camp, I’ve actually been to product preview shows before, so I know what a mock-up is, don’t see any evil in that). AMD has been doing it for many years, and continues doing it.
” more power-efficient chips at the same prices, since performance determines price, not manufacturing costs.”
ok then why is nvidia chips more power hungry then Amd’s. its because they thank more performance per core is important over more cores that are not as power hungry as nvidia’s
“It’s why Intel is currently the performance leader”
On benchmarks that really mean nothing in the real world. on benchmarks they always do them on a clean system with nothing running in the background.
Intel is faster then AMD right now and they have been sense 2006. but does that mean much when are phenom II are the same basic design from the K6 just tweak to hell.
IPC matters but its not the future. Intel will make more cores in the future just wait and i’ll say it again IPC matters most to gamers which are 10% of the market anyways. Servers need more cores and bulldozer allows this. did you even look at the post i gave you proves more cores are more important to servers then per/core performance
http://www.eweek.com/c/a/IT-Infrastructure/Facebook-Tilera-Chips-More-Efficent-Than-Intel-AMD-Products-469783/
Bulldozer was made with server in mind
I know who you are and everyone at every forum i posted this site to told me to just ignore people like you.I would like to just see your point of view, i’m not a bully and i don’t want to be,bully’s are pathetic and bully’s on internet forums are even more pathetic.
Ouch… you are comparing CPUs with GPUs? Don’t do that, ever!
nVidia’s GPU design is massively different from AMD’s (although AMD’s next design will be much more similar to nVidia’s current).
You don’t have this freedom in architecture design with CPUs… Each core still needs to be fully x86-compatible.
You also seem to forget HyperThreading: in a way Intel has its cake and eats it too: HT turns a single fast core into two cores. Great for servers (ever looked at Sun’s Niagara CPU? At a place where I used to work, we had them for web servers).
“And they claimed that Barcelona would be 40% faster than anything out there… while it was actually slower (even disregarding the TLB bug)… to name but a few things.”
You much understand that this is the first new design from the AMD Team sense the K5
“K6 microarchitecture – The K6 was not based on the K5 and was instead based on the Nx686 processor that was being designed by NexGen when that company was bought by AMD.”
This launch can NOT be compared to any other launch Amd has done sense its performance will not be a jump but most likely a leap forward. It is a completely new design and Amd has not done this in a long long time.
Really the K5 was “AMD’s first original microarchitecture. The K5 was based on the AMD Am29000 microarchitecture with the addition of an x86 decoder.” the first 29000 was released in 1988. and i’m sure that design was from amd older design that was from the 70’s.
Also i notice you never said that the FPU unit in bulldozer is more like two 128bit units that can combined into 1 256bit when needed. or when AVX comes into play
The FPU in bulldozer is much better then the sandy as it can do more per cycle then intel’s design.
Firstly, the K5 was eons ago, I doubt that the team from those days is still around today.
Secondly, you’ll need to back up your claim that the FPU in Bulldozer is so good.
It has the problem that it is shared by two threads in a module. SB has a dedicated unit per core.
Also, what’s so special about two 128-bit units? Even K10 has two 128-bit units per core.
Sandy Bridge’s FPU is perhaps the most impressive improvement over Nehalem, and Nehalem already had a very good FPU. It’s a pretty hard act to follow.
“You also seem to forget HyperThreading: in a way Intel has its cake and eats it too: HT turns a single fast core into two cores. Great for servers (ever looked at Sun’s Niagara CPU? At a place where I used to work, we had them for web servers”
Did you also know that most IT departments turn that setting off because it does not give constant performance.
“For applications like Contributor that already maximize the CPU processes, when you attempt to push two 100% CPU utilization threads through each processor you do not receive any improvement in performance. You actually get a decrease in performance since there is some overhead to the hyper-threading processing that gets added to the 200% utilization. If we assume an additional 2% for overhead, you end up attempting to utilize 202% of the CPU cycles available. Thus, Contributor processes will run faster with hyper-threading turned off.”
From IBM
Sure, not all applications may work well with HT, but things like SQL queries and HTTP requests tend to work very nicely. As I said: very nice for servers.
Probably a reason why IBM also supports SMT on their POWER architecture.
Did you look at Sun’s Niagara yet? They have 8-way SMT. An 8-core chip can run 64 threads simultaneously (and it only has ONE FPU for all of them, imagine that).
Surely Sun designed that because it’s not going to work…
“HP recommends that you start with Hyper-Threading disabled and experiment later”
even HP does not seem satisfied with this. HT is not a good way to get more performance when it comes to servers such as Facebook or some other big site.
Actually, it is.
See here for example: http://jrothmanshore.posterous.com/performance-hyper-threading-actually-does-boo
Or here: http://venublog.com/2010/03/21/hyper-threading-performance/
HT adds 20% performance most of the time and only when the cores are not being used at 100%. also adding more cores adds more performance then Fake cores.
The Amd Flex Fpu can do the same amount as Intel and more.
Modern 128-bit FPUs can execute four single precision commands or two double precision commands in parallel per cycle. The yet-to-come AVX technology allows to execute eight 32-bit commands or four 64-bit commands per cycle. However, once a program does not support AVX then: “that flashy new 256-bit FPU only executes in 128-bit mode”. This is naturally a blow for Intel’s 256-bit FPU of Sandy Bridge processor.
The beauty of the Flex FP is that it is a single 256-bit FPU that is shared by two integer cores. With each cycle, either core can operate on 256 bits of parallel data via two 128-bit instructions or one 256-bit instruction, or each of the integer cores can execute 128-bit commands simultaneously.
This would be a whole different story if Amd was selling the 8 core bulldozer at a price of the future (if Intel can even make a 8 core that clocks high) 8 core ivy. But their not at 300$ Amd is basically giving you twice the FPU power! Yeah?
But it does not matter much anyways as most things care more about integer performance.
If it doesn’t support AVX, then that flashy new 256-bit FPU only executes in 128-bit mode (half the throughput). That is, unless you have a Flex FP.
Code doesn’t have to be rewritten to support AVX in order to take advantage of the Flex FP, however. The Flex FP is comprised of two 128-bit FMAC units capable of performing FMAC, FADD, or FMUL instructions per cycle, either combined as a 256-bit instruction or split as two 128-bit instructions, which apparently provides significantly higher performance than ‘competing solutions.’
If one area bulldozer will excel on it will be its FPU unit. one area that does not matter a bit to me.
“HT adds 20% performance most of the time and only when the cores are not being used at 100%. also adding more cores adds more performance then Fake cores.”
Yea, but you haven’t done the maths yet. Some things to consider:
– HT adds 20% performance to cores that are already faster than what AMD has.
– When cores aren’t used 100%, such as for example with server duties that are mostly limited by network or disk I/O (then again, they are never used anywhere near 100%, since x86 sucks).
– Adding more cores takes more die-space than adding HT to a core. A quadcore with HT has only a slightly higher transistor count than a regular quadcore. A six-core CPU already has a much higher transistor count, let alone 8 real cores. In other words: HT is a very efficient way, transistor-wise, to get better multithreading performance out of your CPU. On a quadcore you get 4 extra logical cores from HT… They may be ‘only’ 20% of a real core, but there’s 4 of them, which in most cases beats 2 extra ‘real’ cores (smaller, cheaper chip, lower power consumption, higher clockspeeds on the cores etc).
“However, once a program does not support AVX then: “that flashy new 256-bit FPU only executes in 128-bit mode”. This is naturally a blow for Intel’s 256-bit FPU of Sandy Bridge processor.”
No it’s not, since SB still has one 256-bit FPU per core, where Bulldozer needs to share it between both cores in the module (oh dear, there goes your “twice the FPU power” argument!).
Aside from that, SB’s FPU is built up from multiple execution units, spread over multiple execution ports. So it can indeed perform multiple 128-bit or 256-bit operations per cycle (current AMD architectures can also do multiple 128-bit operations per cycle by the way). See Agner Fog: http://www.agner.org/optimize/microarchitecture.pdf
So it’s not quite as rosy as you thought.
And that’s not even considering the fact that AMD’s current FPU design is not quite as efficient as Intel’s implementations…
If there’s one area where Bulldozer DOESN’T excel, it will be the FPU unit. That seems to be where AMD cut most, in order to fit more cores onto a die.
This one more thing to add for know
What the two statements say to me is that the 17% above the 6174SE (2.3GHz) JF quoted is throughput meaning that if you run SPEC_FP_Rate on MagnyCours(12) and Interlagos(16) at the same clock, the SPEC_FP_Rate result will be UP TO 50% higher due to interconnect and bandwidth constraints.
Then if you run SPEC FP or SPEC INT then the result (1 core) will be 35% higher.
This simply says IPC increases by 35% which is about my estimates for server while I see closer to 50% on the desktop because the Front End should hide latency much better.
Wow, here we go again. Another person who know more intimate details than AMDs own employees. It’s understandable that you may mistakenly think the forecast has changed but you have a choice of taking a positive view or a negative view. You chose to take a negative, which is you prerogative, but to say the word LIAR is disingenuous at best and libelous at worst.
Did you check the date on the earlier post I linked to? It’s almost a year old. To me the forecast didn’t change, I’ve always said Fruehe is full of it, and given plenty of reasons why his stories don’t fit.
Funny you should mention libel… I think Fruehe is quite liable if he indeed spread deliberate lies about AMD’s upcoming products on various internet forums over the course of a year.
There are probably laws against that sort of thing, seeing as it is unfair competition, false advertisement and all that (and I doubt he can use the excuse “I did it in my spare time”, since he very obviously made himself known as an AMD employee, and linked to official blogs and press releases on AMD’s website… which I can only assume he has written during work hours).
I give up dude. It’s obvious no one will change your mind. Hopefully you won’t end up apologizing.
If you want me to change my mind, you need to make a compelling argument. I’m not even sure why you think my mind needs changing in the first place, let alone that I would think it is a good reason.
In fact, I’m not even sure if you even read my previous post on Fruehe, so I’m not sure if you even understand my position.
Yes, I was so wrong in calling Fruehe a liar, shame on me!
“I think Fruehe is quite liable if he indeed spread deliberate lies about AMD’s upcoming products on various internet forums over the course of a year.”
Find 1 JUST 1 Thing that JF lied about and did not make a simple mistake. Really i’ll talk about the bulldozer design with you but this whole thing about JF is a liar Needs to stop.
Just drop it you are wrong about this at least the part were JF is a liar and your wrong about him being bias he has never done that and if he did he has FACTS to back it up. Its like your pissed about him calling you out or something. Who are you anyways what gives you the right to call anyone a liar.
You are nothing more then a bully on the internet.
It will all come out when Bulldozer is released, won’t it?
What is your stance on Randy Allen’s presentation a few months before Barcelona was released? He had been talking about how a native quadcore is more efficient, and made an outright claim that it would be 40% faster than anything on the market.
He must have known this was not correct when he said it, they had engineering samples.
It seems that Fruehe is doing the same thing. Just like with Allen at the time, what Fruehe says doesn’t add up.
Back with Allen, I said that a native design alone wouldn’t improve performance much (since I had actually bothered to measure the overhead on Intel’s MCM designs, and concluded that their latency and bandwidth were actually pretty damn good, even through the FSB). I said they would need some magic ‘secret sauce’ for the design to really be 40% faster, and nothing that AMD had shown would indicate anything like that. What they had shown was basically 4 slightly tweaked Athlon64 cores on a single die, with a shared L3 cache attached. And that was exactly what it turned out to be. There *was* no secret sauce.
So I will just quote what I said a year ago, on Bulldozer:
“What we know so far is this:
– They are removing one ALU and one AGU per core
– They are removing one decoder per core (effectively, as you get one 4-wide decoder per module, so effectively 2-wide decoders per core, as opposed to three decoders).
– They are sharing one FPU/SIMD unit per 2 cores
These are all actions that DECREASE the execution resources in some way.
We have not heard about them even COMPENSATING for the removal of these resources yet… so even being on par with previous generation IPC in single threads would already be quite an improvement in efficiency (roughly 33%, which is arguably more than what AMD or Intel ever achieved in a microarchitecture update, with the obvious exception of Netburst->Conroe, although this was skewed by the drop of about 1 GHz in clockspeed).
So basically, with this level of reduction and sharing of resources, and STILL increasing IPC over the previous gen, that would be one INCREDIBLE feat of engineering… And they’d need some pretty nifty ‘secret sauce’ to make these ‘anemic’ cores run that fast. Have you seen it? I haven’t.”
Fruehe isn’t so much lying about Bulldozer directly, but mostly misrepresenting K7/K8 to make it look less good than it is (to make Bulldozer look better).
Two things have come up in the comments here again:
The 1.5 ALU/AGU part and the 128-bit SSE unit parts can be traced back to Fruehe.
What I am missing is the secret sauce that makes the cores perform better. Sure, Bulldozer will have good performance based on its higher clockspeed and its increased core count… but neither of those are IPC.
Amd is giving you the same amonut of FPU units as of intel. 2600K has 4 256bit units that alwasys work as 128bit units while the rest is just eating power.
With the flex fpu unit you get 4 256bit units also with the 8 core bulldozer. when both cores need to use the FPU its shared allowing them to split it up to 2 128bit units the same as intel all the time except when AVX is needed.
When AVX is needed it can do 1 256bit per cycle, and per module.
“Two 128 bit FMACs shared per module,allowing for dedicated 128-bit execution per core or shared 256bit per module.”
stuff intel cant do is
Execute Two 128 bit AVX in 1 cycle
Execute 128-bit AVX & SSE in 1 cycle
Execute FMA Operations (A*B*C*D)
Intel does not support XOP
Bulldozer FPU can do 64 Flops per cycle (128-bit FP) while intel can ONLY do 32
Bulldozer FPU can do 64 FLOPS per cycle (128-bit AVX) while intel can only do 64
Intel and Amd can both have the same amount of FLOPS per cycle in terms of a 256-bit AVX
I think you don’t understand the difference between an ‘FPU’ and execution units or ports. As I already said, Intel *can* do multiple 128-bit operations in 1 cycle (as can K7/8/10 for that matter).
The comparison with 2600K is not entirely fair, as it is a quadcore that you compare with an 8-core Bulldozer (which will have a considerably larger die to get the same amount of FPU units).
Intel Design has a single scheduler for both integer and floating point, which means that both integer and floating point commands are issued by a single shared scheduler vs. having dedicated schedulers for both integer and floating point executions unlike AMD’s Flex FPU with its OWN scheduler!!!
AMD needs two separate schedulers as int are per-core and float are per-module (which still need to be kept in-sync, as float operations will depend on int operations and vice-versa, so there is no advantage).
Having one scheduler is not a deficiency. It doesn’t matter how many schedulers you have, as long as they keep your execution units fed.
Really, you need to come up with better arguments, or else we should just end it here. I’m getting tired of this.
“In today’s typical data center workloads, the bulk of the processing is integer and a smaller portion is floating point. So, in most cases you don’t want one massive 256-bit floating point unit per core consuming all of that die space and all of that power just to sit around watching the integer cores do all of the heavy lifting. By sharing one 256-bit floating point unit per every 2 cores, we can keep die size and power consumption down, helping hold down both the acquisition cost and long-term management costs.”
Yes, ‘data center workloads’. Sun figured as much, and designed the Niagara for that (as I already mentioned).
Problem is all those other workloads that DO need a lot of FPU processing (photo/video processing, 3D offline rendering, physics simulation etc).
“The comparison with 2600K is not entirely fair, as it is a quadcore that you compare with an 8-core Bulldozer (which will have a considerably larger die to get the same amount of FPU units).”
Its a 100% fair because Amd is pricing it the same. if they priced it higher that would make it unfair.
Really its going to come down to where they price the 4 core bulldozer if its priced twords the I3 then Amd is going to have a WIN. the closer they price it twoards the 2500K the more dangers intel positon on amd will be.
Did you also note that 4 module bulldozer will be able to pump out 4 beastly threads and when you start to use more the bulldozer will start to share more.
So with AMD you get 4 beastly threads close or slightly below to Intel’s when the user starts to use more threads the bulldozer will start to share more. On average you can get 20% more performance of HT while Amd will give you 80% more a bulldozer module has 180% the performance of 2 BULLDOZER HYPOTHETICAL CORES.
I wish AMD just called the bulldozer a 4 core with super threading. When one core is being used heavily on just 1 module it will perform much better anyways. When bulldozer starts to use more then 4 cores on its 8 core that’s when the bulldozer will start to lose performance per/core.
A 8 core bulldozer will not have the same power of a 8 module bulldozer,even at programs that just use 8 cores.
By the way i watched the video with JF and he said throughput not performance Again throughput with the current magny cores at that time, You really are posting nonsense. Change the title of this blog to Scali did not understand what John Fruehe was saying.
“Its a 100% fair because Amd is pricing it the same. if they priced it higher that would make it unfair.”
That’s nonsense. Price is only a result of performance. What a company like Intel or AMD wants is to have as much performance as possible from as small a die as possible, since this means they can charge high prices while having low manufacturing costs and high volume production.
The current situation is that Intel’s CPUs are more expensive for the consumer than AMD’s, while they are cheaper for Intel to manufacture. AMD practically has to give away 6-core CPUs for free, in order to compete with mainstream Intel quadcores, not even talking about Intel’s high-end quadcores or 6/8-cores. Clearly AMD needs to improve on this situation with Bulldozer, else their long-term viability as a company is severely compromised: less profit means less money to invest in future products and manufacturing.
Price is about as irrelevant a metric as can be, when it comes to silicon.
“Did you also note that 4 module bulldozer will be able to pump out 4 beastly threads and when you start to use more the bulldozer will start to share more.”
We will have to see exactly how ‘beastly’ those threads will be. Sandy Bridge isn’t exactly a slouch.
“By the way i watched the video with JF and he said throughput not performance Again throughput with the current magny cores at that time, You really are posting nonsense.”
What video? I don’t link to a video. Whatever he says in whatever video is completely irrelevant to this blog. I link to a blog that he made last year, and I quote him directly: http://blogs.amd.com/work/2010/08/02/what-is-bulldozer/
The blog has exactly 0 occurances of the word ‘throughput’. It does mention ‘performance’ however. And I’m tired of repeating that. Next post mentioning it will just be deleted.
i was going by the video on youtube
For the consumer buying the processor price/performance is all that matters and that is how products should be reviewed by price.
Don’t review a celeron with a I7 next to it. review it with a semperon next to it.
For the company though your 100% correct the less die space the better and we all need to know where Amd is going to price the 4 core bulldozer. if its close to the I5 2500K and performs just as good i say that’s a victory indeed! if they price it towards the I3 and it gets the same performance on average then i say Amd is making less money off those chips then intel.
Really i rate a bulldozer module a beastly core. With support for 2 threads simultaneity. The module is AMD answer to Intel hyper-threading this is why we might see the 4 core bulldozer being priced towards the I3 while the 6 core will be priced at the I5 and the 8 core will be priced at the I7.
Also this is why Amd has better price/performance then Nvidia sense it takes less die area to beat them thanks to AMD smarter design. Not only does less die space matter in terms of cost but it also matters in terms of heat and power efficiency as well.
The 6970 and the Gtx 570 have the same performance but the GTX 570 has 360M more Transistors.
Again, stop the comparisons with GPUs.
I certainly wouldn’t call AMD’s GPU design smarter. Yes, for a vanilla DX10 card, their approach was efficient…
nVidia never tried for vanilla DX10 though, they went for GPGPU as well.
And now that we have DX11, GPGPU is a standard part of the API.
Another standard feature is tessellation.
AMD doesn’t do well in either of these things, and price/performance quickly swaps to nVidia’s favour when you use these features (then a GTX560 will match or outperform a 6970, where the GTX560 has roughly the same die size, and is actually cheaper).
nVidia’s design is more efficient for GPGPU, and they also have more features than AMD. Which is why we will see AMD abandoning their approach for the upcoming generation of GPUs, and adopting nVidia’s approach instead. So clearly nVidia’s approach is the smarter one.
Also, despite nVidia’s larger die size, power consumption was never far off from AMD’s much smaller chips.
CPUs tend to be much more similar (mostly because they have to support the same ISA in hardware, GPUs are free to implement any ISA they like, drivers provide compatibility, not hardware). You won’t see such differences in die size or power consumption.
“AMD doesn’t do well in either of these things, and price/performance quickly swaps to nVidia’s favour when you use these features”
These features you talk about are used in 5% of the total games on the market today. Also i am more likely to do a 3 moniter setup then use Physx. Amd has better visual quality then nvidia or intel on Videos which more like 90+% of all PC users care about versus the Physx 5% or less.
http://www.tomshardware.com/reviews/radeon-hd-6570-radeon-hd-6670-turks,2925-14.html
Example..
When i had a nvidia 9500gt video card i sold it because i was upgrading to a 4770 which only cost 105$ and perfourmed like or better then a 9800GT that cost 180$ at the time, i had to use onboard video which was my Ati 4200 i imedictly noticed better visual Quality.
“nVidia’s design is more efficient for GPGPU”
Most people as of today don’t use their gpu’s for GPGPU computing. When Amd feels this starts to change they will change to oh wait that’s what the Radeon HD 7000 series is all about sorry.
“and they also have more features than AMD”
Any people care about? such as better Quality in video’s?
“So clearly nVidia’s approach is the smarter one.”
How when a 6990 is faster then the GTX590 and uses less power? The Amd card is only loud becasue of the shity cooler.
“When Amd feels this starts to change they will change to oh wait that’s what the Radeon HD 7000 series is all about sorry.”
…
“How when a 6990 is faster then the GTX590 and uses less power?”
You are contradicting yourself.
Also the rumor is Amd will make it to 28nm die before Nvidia giving Amd the edge again.
http://www.xbitlabs.com/news/graphics/display/20110726103803_AMD_Expects_to_Be_First_GPU_Designer_with_28nm_Graphics_Chips.html
http://vr-zone.com/articles/rumour-nvidia-s-next-gen-kepler-taped-out-amd-southern-islands-in-september/12799.html
After reading all of these exchanges between tech sites article-based second-hand knowledge versus real world IT hands-on experience, it only reinforces me that AMD fanboys like to shoot themselves on their feet but would never admit it’s their own doing.
In any case, for 7 series, AMD follow NVIDIA’s GPU design.
At least, the basic principle of it, as in no more VLIW-based GPUs.
If, like JDWII has implicitly said numerous times, AMD as the patron saint of efficient designs have started to follow NVIDIA’s way of designing chips, then NVIDIA must’ve done something good, if not right.
Anyway, while I tend to side with Scali’s (negative) view on Fruehe’s statement way back then, which ultimately being reinforced by Seifert’s statement not too long ago, I really hope AMD can put out at least a Sandy Bridge equalizer.
Having it perform x% better than the last generation of its own chips doesn’t really mean anything if it can’t beat a price-equivalent Sandy Bridge.
If Bulldozers aren’t the ‘dozers it’s meant to be, I don’t know how long they’d stand up to their ‘Bulldozer’ moniker when Ivy Bridge, the 22nm Sandy Bridge variant, is out.
All the while still having to compete with the (expected to be) lower-priced Sandy Bridge models, especially the 2500K & 2600K, the prince & king of the CPU hinterland right now (from cost, performance & overclockability point of view).
Remember the price drop of i7-920 back then?
Btw, Scali, I think you missed replying to one of JDWII replies, the one before g2-somerandomhashlikeshit’s reply.
The one that mentions ‘Front End’ or some “smart-looking” random terms.
I think everyone hopes for competitive products from AMD. I think the difference between me and your average AMD fanboy is that I don’t think it will happen just because I hope it’s going to.
And yes, I saw that comment, but didn’t think it was even worth replying to. Just some wild speculation based on… nothing.
In my opinion people are too distracted about bulldozer’s speculations.
i like read more about rumors than news and i try to take all with a grain of salt and to understand what’s going to happen.
So i like also jf’s leaks. But i know what i read…
I could think he has a great passion… sure is true
I could think he likes his bike… why not? sure
I could think he wants help my bulldozer understanding… umh yes yes.
And i can’t ignore the fact that he talks everytime everywhere about his products.
The amazing thing is his indirect use of his nda disclaimer “While I work for AMD, my posts are my own opinions”.
In people mind he is an authority, he knows things from inside, but seems he goes in forums for fun, seems he wants to post his opinions like everyone else.
He talks in this way and his disclaimer indirectly increases this perception.
If i said he makes marketing one could respond: marketing? absolutely not you are a badass
The point is that spreading a marketing content (and i use the world content) through personal opinions is by definition VIRAL MARKETING and a man who makes his job can’t ignore it; it’s impossible. So he isn’t (from an objectively point of view) in the position to make free considerations. he isn’t in the position to talk like everyone else: in fact he isn’t like everyone else. He has and use knowledge’s autority. He has and use his position’s autority to witness his statement’s correctness.
obviously he can make what he wants but people should have a healthy criticism instead of drink everything and defend the system like people in Matrix.
And really you can’t sell contracts for server on youtube, playing with your voice tone like in a message for babies.
Theese stupid things are for us, to grab our attention; and i hope we aren’t babies.
My point of view does not apply to intel and amd, it applies to marketing and the babies. This kind of marketing.
He has lied many times, like others. And many fan know it. I think is stupid defend the company in the religion war becouse in the end babies are you.
i apologize for my comment in regard of jf. i expressed my personal opinion about how cryticism should be always be encouraged in debates like this. I think internet is a place of free expression… My english is really poor and as i confused the term “word” (think about how word is known) with “world” for assonance just to make an example about my errors i used the term lied that is improper and i’m sorry. I’m not used to write in english and while i write i use my little knowledge, google translator or what i see around. hope is clear jf can talk liberally and my critics are addressed to peolpe that don’t use a healty criticism on what they read ariund. I know jf is not here but i’m sorry and i said it in every place i talked about him. In reality few places. Is evident that everyone must be taken with the grain of salt in ambients so poisoned and this regards me too. I’m an average Joe and considered how is hard for me don’t make mistakes i shouldn’t talk at all.
Sorry 😦
Well, I don’t have a problem with talking liberally in general. But now that Bulldozer is out, I think you will agree that John Fruehe went a bit too far in what he said on forums.
I think you made some good points in your first post. He really did abuse his credibility as an AMD employee, which was very deliberate, having AMD in his nickname, and the disclaimer in his signature.
Yes, people should always be critical, and take things with a grain of salt… but John Fruehe should not have done what he did. Marketing should not be about lying like this. It should be about emphasizing your product’s strong points, and directing attention away from its weak points. You don’t have to lie to do that, merely manipulate.
If you think like an engineer then;
Seifert might be the liar, one or both both might be in genuine error, mistaken or have been misquoted. Both could also be correct depending on the specific product referenced in each case.
If you think like a blogger then;
Fruehe is a liar
The internet makes it quite easy to jump to pointless conclusions with a lack of evidence. If you are later proven correct by hard facts, it still won’t mean you think clearly, just that you are lucky and should enter the lottery. Good luck to you sir in your gambling enterprise.
If you think like a commenter, then you don’t bother to read properly, and look for pathetic excuses everywhere.
1) The situation is not ‘Seifert or Fruehe’. If you bothered to read the earlier post, you’d see that I already called Fruehe out a year ago, long before Seifert made any comments.
2) Fruehe can not have been misquoted, since I point directly to his own blog at AMD, and to some forum posts he has made. It is literally what he wrote. If it was wrong, he could have edited or amended.
3) I don’t say Seifert is right, I think they’re probably both being overly positive about Bulldozer. Seifert’s claims are just a tad less positive, which indicates that Fruehe was wrong.
4) I have given plenty of technical reasons why Frueue’s claims are inaccurate in my previous blog. You should read it. It’s not just a gamble. Plenty of information doesn’t match with what Fruehe says. Both about Bulldozer, and about comparisons with K10 (he misrepresents the FPU and ALU/AGU capabilities of the previous generation to make his Bulldozer claims more believable. Why would he do that?).
5) If you’re an engineer like me, you will argue the points I have made in my two blog posts and the comments, rather than attacking me personally, like you did.
Hey look, I won!
Pingback: Hate fail | Scali's blog
Pingback: Details zu AMDs Vishera Quad-Cores - Seite 4
Pingback: AMD Steamroller | Scali's OpenBlog™