AMD CEO talks of long-term turnaround

[AtenRa]

We wouldnt have bulldozer in the first place then would we?

You know very well Bulldozer was faster than Thuban in MT which was the goal in the first time. PileDriver and especially SteamRoller are ways faster than Thuban.

 

[Fjodor2001]

I didn't say that. I said it would be in the same competitive position, meaning it won't beat Intel's 2016 chips.

So what exactly are you saying then? Previously you said Zen would be in the same position towards Intel's CPUs as AMD FX CPUs are today. So do you think Zen will be the same percentage of performance behind Intels mainstream CPUs in both ST and MT performance as AMD FX CPUs are today. Really?

And what exactly do you mean by that Intel CPUs In 2016 will "beat" Zen? In ST performance, MT performance, price, or what?

I think nobody is expecting Zen to beat Intel Skylake in ST performance if that's what you mean.

But do you also think that an 8 core Zen will be slower than a 4 core Intel CPU in MT performance?

 

[cytg111]

Would you use a system that didn't have upgradable memory?

Memory bandwidth isn't an issue with CPUs anyway.

Assuming you mean x86. If not, then here ya go, http://forums.anandtech.com/showthread.php?t=2426221&highlight=

I think that may be the one vector AMD has a chance to edge past Intel, getting stacked memory on the die/sock first.
And yes, most 99.999% people (and I did this survey myself so you can trust it with P<0.0001) would happily use a system with non-upgradeable memory, me included.

But you guys a talking about things so vague, so distant crystall ball bs that putting your money(or mouth) on the outcome right now is just plain.. lesser insulting word... even lesser insulting word.. Silly. Just plain silly.

 

[mrmt]

I think nobody is expecting Zen to beat Intel Skylake in ST performance if that's what you mean.

But do you also think that an 8 core Zen will be slower than a 4 core Intel CPU in MT performance?

Come on, think harder. Why didn't AMD bring their 16 core Visheras to the consumer market?

 

[Phynaz]

You know very well Bulldozer was faster than Thuban in MT which was the goal in the first time. PileDriver and especially SteamRoller are ways faster than Thuban.

That's BS. Don't make me start quoting JFAMD posts from this very forum.

Go ahead and defend AMD all you want, but don't start making up lies to do it.

 

[Phynaz]

So what exactly are you saying then? Previously you said Zen would be in the same position towards Intel's CPUs as AMD FX CPUs are today. So do you think Zen will be the same percentage of performance behind Intels mainstream CPUs in both ST and MT performance as AMD FX CPUs are today. Really?

And what exactly do you mean by that Intel CPUs In 2016 will "beat" Zen? In ST performance, MT performance, price, or what?

I think nobody is expecting Zen to beat Intel Skylake in ST performance if that's what you mean.

But do you also think that an 8 core Zen will be slower than a 4 core Intel CPU in MT performance?

Kind of a low bar to set, don't you think? You're saying AMD will need twice the number of cores to beat Intel's mainstream CPUs.

Do you have any facts that mainstream Zen will be greater than four cores?

 

[Fjodor2001]

Come on, think harder. Why didn't AMD bring their 16 core Visheras to the consumer market?

Huh? What does that have to do with anything. 16 cores would be too much in the mainstream market. Certainly back then when we did not have 14 nm.

Zen is a completely different uArch. It will prioritize ST performance more, which is where AMD has been lacking. And they will still have 8 cores, while Intel will be on 4 cores.

You should already know all of this actually. And I'm really not sure what you were trying to say anyway...

 

[Fjodor2001]

Kind of a low bar to set, don't you think? You're saying AMD will need twice the number of cores to beat Intel's mainstream CPUs.

Do you have any facts that mainstream Zen will be greater than four cores?

No, that's not what I'm saying. But based on info available so far I'd estimate 8 core Zen to demolish Intel 4 core CPUs in MT performance, and come close in ST performance (most are guessing SB to Haswell level of performance).

If they succeed with that, it should be a very good product. Don't you agree?

 

[AtenRa]

That's BS. Don't make me start quoting JFAMD posts from this very forum.

Go ahead and defend AMD all you want, but don't start making up lies to do it.

Everyone knows that CMT is all about Throughput and Bulldozer architecture was made for the Server first. Now if you have an agenda and want to make anyone here believe that AMD was aiming for High IPC and High Single Thread performance for the Bulldozer architecture go ahead.

 

[Sweepr]

Everybody knows that JF-AMD kept telling BD's IPC would improve. Anyone who didn't believe the absurd hype about better than Phenom II IPC, high clocks and 8-cores was being made fun of by fanboys (including creative forum signatures). Turns out the joke was on them.

 

[mrmt]

Zen is a completely different uArch. It will prioritize ST performance more, which is where AMD has been lacking. And they will still have 8 cores, while Intel will be on 4 cores.

Zen will have higher ST performance and will have a high core count, what do you expect to happen with the clocks of these 8 cores, and consequently what happens with performance per core?

No, that's not what I'm saying. But based on info available so far I'd estimate 8 core Zen to demolish Intel 4 core CPUs in MT performance, and come close in ST performance (most are guessing SB to Haswell level of performance).

Ed: So basically you think they might have slightly lower ST performance but a lot of MT performance because of core count. I assume that you think Zen cores will clock as high as Intel core and Samsung 14nm will be as efficient as Intel 14nm (because it's a server product, it has to be efficient, right), and not only that, you think AMD will be able to clock the cores as high as Intel, but twice the quantity.

Pie-in-the-sky stuff is going to happen if AMD is with all these goals in mind.

 

[AtenRa]

From the original Bulldozer launch conference call(EMEA).

Aten-Ra:What IPC improvements will we see against Deneb or Thuban in single Thread and Multi thread applications?

AMD - Bulldozer is a new and radically different core architecture making it impossible to draw IPC parallels to any previous design. With the design of the AMD FX, we focused on offering the highest possible instructions per watt. AMD FX is designed with current and future workloads in mind. At the same time, we have added the new instruction sets and as improvements are realized in applications and operating systems (like the Windows 8 scheduler improvements), we believe performance will increase.
Note: In older apps, either compiled with an older compiler, or those using older instructions will see FX at a disadvantage. Again this was a design decision. Design for today’s apps and the future was the engineering mantra.

That was the problem from the start, the Bulldozer architecture was completely different than Thuban and needed software specifically written for it to be able to perform at its peak. Something that was lacking in the BD launch back in 2011.
IPC is related to the software, if the software is written with Bulldozer guidelines then there are lots and lots of situations that the BD has higher IPC than Thuban. The problem is that people always see legacy software/applications and there BD is slower in IPC than Thuban/Deneb.

 

[mrmt]

Everybody knows that JF-AMD kept telling BD's IPC would improve. Anyone who didn't believe the absurd hype about better than Phenom II IPC, high clocks and 8-cores was being made fun of by fanboys (including creative forum signatures). Turns out the joke was on them.

Actually I said 3 things:
1. Interlagos will perform 50% better than MC in the same thermals
2. IPC would be higher
3. Increase for single-threaded workloads will be "a lot" more than 17%

I have made no statements that I am aware of about IPC with a percentage implied because I don't know what the IPC is, all I was told is that it would be higher. If I did say IPC would be higher by a percentage, it was a mistake; ocasionally those things happen to us humans.


(...)

This is a nice summary by Frueher himself.

 

[AtenRa]

Zen will have higher ST performance and will have a high core count, what do you expect to happen with the clocks of these 8 cores, and consequently what happens with performance per core?

You are forgetting that ZEN will be manufactured at 14nm FF (Gate Last) and not 32/28 (Gate First).
So, clocks may stay the same at 4GHz or even increase. We dont know yet but a 4 Core 8 Thread ZEN could be able to maintain the 4GHz clocks of current Bulldozer SKUs.

 

[Spartak]

I've said this a few times already, and I'll say it again: IPC alone tells you absolutely nothing about performance, and AMD absolutely can hit any IPC target they want, assuming they are willing to relax clocks enough.

What matters is performance. Performance is IPC * clock speed. How difficult it is to get high IPC depends entirely on how high you set your clock targets, which determines the maximum length of the critical paths in your CPU, and thus how complex things you can do per each clock cycle.

Saying "our next chip will have 40% more ipc than the last" is kind of bullshit marketing speech. It sounds good, "40% better", but it tells you practically nothing except that they decided to aim for a more brainiac, probably wider design. Zen could have 40% higher ipc, and 25% lower clock targets, for a net gain of 5%.

Bulldozers IPC problem isnt related to its pipeline length like Prescott, so what you're saying is incorrect. It was a 'castrated' dual processor (with only one FPU unit) running on shared front and back-end resources. A wider design doesnt impact scalability in itself, but there could be related factors when you increase the complexity that do have an impact. So while there could be a frequency penalty to Zen attributed to the new design, it's not to be expected in a significant way if at all. The new 14nm FF process is the wildcard here.

 

[Spartak]

Kind of a low bar to set, don't you think? You're saying AMD will need twice the number of cores to beat Intel's mainstream CPUs.

Do you have any facts that mainstream Zen will be greater than four cores?

The "fact" is the official 40% IPC gain claim from AMD which would put it in Nehalem, maybe even Haswell territory, performance per core. It's fair to make that assumption based on that claim.

It's not a crazy claim when you understand _why_ Bulldozer performed so bad and _why_ AMD decided to go with the design in spite of that. It's not like they were caught with their pants down once the Bulldozer benchmarks hit the internet. The process generation gap with Intel was spiralling out of control and multicore was the new megahertz myth. A desperate attempt from a company that lacked the resources to iterate on Phenom (making it more complex in a useful way is the hard part) but definately one they can recover from once they make the jump to 14nm with a proper design that isnt starved from execution resources.

Intel did the same with Conroe and on; combine what whas great about Pentium III with some of the great innovations from Netburst in branch prediction, SMT etc.. Conroe wasnt designed from scratch, and I'd wager a bet Zen isnt either.

 

[dark zero]

Seeing AMD well and how they are retarded and seeing how nVIDIA is about to commit seppuku with risky moves, I am conviced that both needs to die in order to make Intel an absolute monopoly.

 

[ShintaiDK]

You are forgetting that ZEN will be manufactured at 14nm FF (Gate Last) and not 32/28 (Gate First).
So, clocks may stay the same at 4GHz or even increase. We dont know yet but a 4 Core 8 Thread ZEN could be able to maintain the 4GHz clocks of current Bulldozer SKUs.

Lets see how the fabled Samsung 14nm will perform first as well. On the SoC side it havent exactly turned any heads yet.

Not to mention the extremely optimistic view on the rest.

One would think AMDs history had taught you something by now.

 

[Phynaz]

Everyone knows that CMT is all about Throughput and Bulldozer architecture was made for the Server first. Now if you have an agenda and want to make anyone here believe that AMD was aiming for High IPC and High Single Thread performance for the Bulldozer architecture go ahead.

You of all people saying I have an agenda??

"Throughput computing" is marketing BS. If it was true then you really have to admit how badly AMD screwed up, since they have lost about 90% of their server market share since BD launched.

Here's everything to prove you wrong, strait from AMD's mouth.

JF's excuse (after which his credibility was shot and he never posted again)

The original design goal was higher IPC.

http://forums.anandtech.com/showpost.php?p=32421412&postcount=106

it's about performance, power consumption and price.

http://forums.anandtech.com/showpost.php?p=31639192&postcount=1703

BD will be plenty competitive.

Q2 for client, Q3 for server
http://forums.anandtech.com/showpost.php?p=31279774&postcount=853

1. Interlagos will perform 50% better than MC in the same thermals
2. IPC would be higher
3. Increase for single-threaded workloads will be "a lot" more than 17%

Adam Kozak-

The architecture focuses on high-frequency and resource sharing to achieve optimal throughput and speed in next generation applications and high-resolution gaming.

Analyst day slide:

Targeted at Server and Client
Continued scaling for single thread performance

http://cdn.mos.techradar.com/classi...eripherals/images/amd-bulldozerjpg-970-80.jpg


Would you please stop now?

Edit:
mrmt beat me to it.

 

[Fjodor2001]

Zen will have higher ST performance and will have a high core count, what do you expect to happen with the clocks of these 8 cores, and consequently what happens with performance per core?

Ed: So basically you think they might have slightly lower ST performance but a lot of MT performance because of core count. I assume that you think Zen cores will clock as high as Intel core and Samsung 14nm will be as efficient as Intel 14nm (because it's a server product, it has to be efficient, right), and not only that, you think AMD will be able to clock the cores as high as Intel, but twice the quantity.

Pie-in-the-sky stuff is going to happen if AMD is with all these goals in mind.

Well, just so we know where you yourself are standing on this, what are your own expectations on Zen?

Let's say we take Intel 4770 (or Skylake 6700 which Zen will be up against if you prefer that, but its performance is not 100% known yet) as reference. Then normalize its performance at index 100. What performance value would you in that case estimate 8 core Zen to have, in ST and MT performance respectively (e.g. 80 in ST and 150 in MT perf)?

I'm just looking for a rough guesstimate from your side here...

 

[Phynaz]

No, that's not what I'm saying. But based on info available so far I'd estimate 8 core Zen to demolish Intel 4 core CPUs in MT performance, and come close in ST performance (most are guessing SB to Haswell level of performance).

If they succeed with that, it should be a very good product. Don't you agree?

No. let me know when 4 core vs 4 core is competitive.

How about we switch it around and compare 8 core Intel CPUs to 4 core AMD CPUs. Talk about demolishing the competition.

Again, do you have any info that mainstream Zen will be 8 cores? If not, then you are just making up fake scenarios so your team wins.

 

[mrmt]

Well, just so we know where you yourself are standing on this, what are your own expectations on Zen?

What I expect is the same trade off as of Bulldozer, albeit in a less unbalanced fashion, e.g., less anemic cores but still a high quantity of them available, which means that the consumers will face the same trade off they face today with Bulldozer, getting stomped in most common workloads, gaming included, and performing better in niche applications that only a few niche users and certain dishonest AMD resellers care about. I think Zen will be a mediocre product for the consumer market, and only an improvement because the CMT chips were an outright disaster that imploded the company in the 5 years they were sold.

I think it will fail badly on the server market. Intel is raising core count like crazy and if AMD plans to stick with that old crossbar + MCM of them they will be toast, and given that they plan to reuse server Zen on the consumer market, they are well on its way to hit that wall.

 

[Fjodor2001]

What I expect is the same trade off as of Bulldozer, albeit in a less unbalanced fashion, e.g., less anemic cores but still a high quantity of them available, which means that the consumers will face the same trade off they face today with Bulldozer, getting stomped in most common workloads, gaming included, and performing better in niche applications that only a few niche users and certain dishonest AMD resellers care about. I think Zen will be a mediocre product for the consumer market, and only an improvement because the CMT chips were an outright disaster that imploded the company in the 5 years they were sold.

I think it will fail badly on the server market. Intel is raising core count like crazy and if AMD plans to stick with that old crossbar + MCM of them they will be toast, and given that they plan to reuse server Zen on the consumer market, they are well on its way to hit that wall.

It's very hard to judge what your expectations are by that alone. Care to put some actual numbers on estimates for ST and MT performance for 8 core Zen vs some Intel reference CPU as I requested to make things more clear?

 

[mrmt]

It's very hard to judge what your expectations are by that alone. Care to put some actual numbers on estimates for ST and MT performance for 8 core Zen vs some Intel reference CPU as I requested to make things more clear?

Nehalen performance with Nehalen clocks.

 

[Spartak]

No. let me know when 4 core vs 4 core is competitive.

How about we switch it around and compare 8 core Intel CPUs to 4 core AMD CPUs. Talk about demolishing the competition.

Again, do you have any info that mainstream Zen will be 8 cores? If not, then you are just making up fake scenarios so your team wins.

Jesus christ, he won't be able to let you know until it's actually released, wont he? But AMD's 40% claim would imply it's in the same ballpark.

So an 8-core desktop version of Zen would be competitive to Intels offerings in the high performance segment if it indeed stacks up to AMD's own claims. It doesnt need to beat Core i7's IPC performance for it to be a good value proposition in the high-performance segment.

Here's one source for your 'fake scenario': http://www.kitguru.net/components/c...en-processors-is-our-largest-rd-spending-now/