Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[igor_kavinski]

20a is already out. It's Meteor Lake. We won't see anything else on 20a.

Just woke up?

 

[Markfw]

It does but it’s really close, you can basically say Specint is a wash but Intel is slightly ahead in Specfp.

View attachment 91409

Is that the 7840u vs something ? Also, I don't put a lot of faith in spec anything, I use real world benches, but laptop chips don't use as many of those. But one chip is close in ONE bench ? I would not say Intel is ahead in IPC, best case scenario, the latest laptop chips are "in the ballpark". Intel is far behind in everything else.

 

[H433x0n]

Is that the 7840u vs something ? Also, I don't put a lot of faith in spec anything, I use real world benches, but laptop chips don't use as many of those. But one chip is close in ONE bench ? I would not say Intel is ahead in IPC, best case scenario, the latest laptop chips are "in the ballpark". Intel is far behind in everything else.

It's results from client desktop comparing 13900K vs 7950X. This is just raw score for Specint & Specfp at a fixed frequency using the same memory, it doesn't take power consumption into account.

 

[Abwx]

It's results from client desktop comparing 13900K vs 7950X. This is just raw score for Specint & Specfp at a fixed frequency using the same memory, it doesn't take power consumption into account.

There s Povray in the FP tests, Zen 4 perform 6% better per clock than RPL and 18% better than in the usual Povray distribution, assuming of course that AVX2 is enabled for AMD, wich is not the case in this Povray test as well, how much more such misses in the Spec FP suite..?..

 

[soresu]

And AMD can't benefit so much from Radeon image than Nvidia with RTX

I don't know that benefit is the word I would use to describe RTX given their PR has essentially conflated it to mean real time ray tracing too which causes people to often say RTX even when they are talking about AMD hardware.

 

[StefanR5R]

you can basically say Specint is a wash but Intel is slightly ahead in Specfp.

View attachment 91409

Are the TLAs meaning the following?
RPC = Raptor Cove, Core i9-13900K P core
GLC = Golden Cove, Core i9-12900K P core
RPH = Raphael, Ryzen 9 7950X Zen 4 core
GCM-12 = Gracemont, Core i9-12900K E core
GCM-13 = Gracemont, Core i9-13900K E core

Are the results yours or taken from a site which you could link to for those who are interested in the fineprint of the report?

 

[H433x0n]

Are the TLAs meaning the following?
RPC = Raptor Cove, Core i9-13900K P core
GLC = Golden Cove, Core i9-12900K P core
RPH = Raphael, Ryzen 9 7950X Zen 4 core
GCM-12 = Gracemont, Core i9-12900K E core
GCM-13 = Gracemont, Core i9-13900K E core

Are the results yours or taken from a site which you could link to for those who are interested in the fineprint of the report?

Yes, you've got that correct. This is the original source... sorry to hijack.

 

[Hans Gruber]

Better IPC in what vs what chip ? desktop, NO, server NO, HEDT NO, laptop (7840u vs several different Intel) NO. Care to qualify that ?

AMD vs Intel: Which CPUs Are Better in 2023?

We put Intel vs AMD in a battle of processor prowess.

www.tomshardware.com

Here you go.

 

[StefanR5R]

Intel has a better IPC,

Better IPC in what vs what chip ?

More importantly, better IPC in which workload? If this is not stated or can't be inferred from context, the term "IPC" is useless bullshit.

AMD vs Intel: Which CPUs Are Better in 2023?

We put Intel vs AMD in a battle of processor prowess.

www.tomshardware.com

Here you go.

There is no valid discussion of "IPC" in this article.

But at least it's a fun read. "For professionals on the hunt for performance in content creation and productivity applications, the winner of AMD vs Intel CPUs goes to Intel on the strength of its x86 hybrid architecture." :-D As if.

 

[Hans Gruber]

More importantly, better IPC in which workload? If this is not stated or can't be inferred from context, the term "IPC" is useless bullshit.


There is no valid discussion of "IPC" in this article.

But at least it's a fun read. "For professionals on the hunt for performance in content creation and productivity applications, the winner of AMD vs Intel CPUs goes to Intel on the strength of its x86 hybrid architecture." :-D As if.

None of this matters when AMD releases Zen5. AMD will be in the lead until at least Arrowlake with the 20A 5nm silicon. We still do not know what the new Intel silicon will do. It will be more efficient going from 10nm to 5nm silicon. That is why for mainstream Zen5, it makes sense to roll back TDP to Zen3 levels. That would be 65w, 105w and 125w at the high end for mainstream processors.

Right now the 5700x (Zen3) is 8 core with a TDP 65w. If Zen5 has an 8 core part with a 65w TDP. That will wipe the floor with Intel. Zen 5 should be setup for efficiency with a bios setting to increase the TDP in their processors. Right now, users have to turn on eco mode. It should be the other way around.

AMD obviously still smokes Intel on the server side, more cores and much lower power consumption.

Has AMD said if Zen5 will be exclusively 3nm silicon? I have head they may be using TSMC N4 (4nm) which is really 5nm enhanced silicon.

 

[StefanR5R]

Has AMD said if Zen5 will be exclusively 3nm silicon? I have head they may be using TSMC N4 (4nm) which is really 5nm enhanced silicon.

AFAIK AMD's latest public statement on that was on their last Financial Analyst Day, June 9, 2022. They said then that there are going to be Zen 5 products on 4nm and ones on 3nm.

(Due to TSMC's schedule, the first Zen 5 products will be on a 4nm process.)

 

[Markfw]

None of this matters when AMD releases Zen5. AMD will be in the lead until at least Arrowlake with the 20A 5nm silicon. We still do not know what the new Intel silicon will do. It will be more efficient going from 10nm to 5nm silicon. That is why for mainstream Zen5, it makes sense to roll back TDP to Zen3 levels. That would be 65w, 105w and 125w at the high end for mainstream processors.

Right now the 5700x (Zen3) is 8 core with a TDP 65w. If Zen5 has an 8 core part with a 65w TDP. That will wipe the floor with Intel. Zen 5 should be setup for efficiency with a bios setting to increase the TDP in their processors. Right now, users have to turn on eco mode. It should be the other way around.

AMD obviously still smokes Intel on the server side, more cores and much lower power consumption.

Has AMD said if Zen5 will be exclusively 3nm silicon? I have head they may be using TSMC N4 (4nm) which is really 5nm enhanced silicon.

That is still a narrow win. avx-512 ? AMD wipes the floor with Intel (see DC primegrid apps as one example), and others. And if you use 100% 24 hours a day like DC community does ?power use makes Intel not even in consideration. My all AMD DC fleet takes 1/2 the total power a Intel fleet of the same strength. And there are other apps as well where AMD wins. Their benchmarks are not close for many people.

 

[Kolifloro]

My first message here ... hi everybody !!!


I've been reading you silently for several months ... and ... I admit it : I am an enthusiast ignorant ... BUT , I would like to know ...

My question is :

I would like to know whether AMD were able to 'roll-out' Strix Point at ... let's say May ... why would they do it on October instead ??

I can think of 'commercial reasons' ... to give some time Hawk Point to get sell ...

On the other hand ... from my lack of knowledge, I think the soonest Strix Point (Zen 5) hits retailers' shelves ... the more market share AMD will take away from Intel ... at least as far as the laptop market is concerned.


Please ... shed some 'light' ...


Regards.

 

[ikjadoon]

Is that the 7840u vs something ? Also, I don't put a lot of faith in spec anything, I use real world benches, but laptop chips don't use as many of those. But one chip is close in ONE bench ? I would not say Intel is ahead in IPC, best case scenario, the latest laptop chips are "in the ballpark". Intel is far behind in everything else.

Defining real-world benches might be helpful for this discussion?

//

TweakTown also measured a +9% IPC advantage for Raptor Lake versus Zen4:

Tweakers.net found about the same:



Overclockers tested a few more:

 

[DrMrLordX]

Eh? MTL is Intel 4. . .

Oh my bad, I had a brain fart there.

 

[soresu]

Oh my bad, I had a brain fart there.

Don't sweat it.

I'm only 38 and seem to average at least one a week these days 😅

 

[misuspita]

Soo... how's AMD gonna feed so many hungry Strix Halo CPU, GPU and AI cores?

 

[eek2121]

My first message here ... hi everybody !!!


I've been reading you silently for several months ... and ... I admit it : I am an enthusiast ignorant ... BUT , I would like to know ...

My question is :

I would like to know whether AMD were able to 'roll-out' Strix Point at ... let's say May ... why would they do it on October instead ??

I can think of 'commercial reasons' ... to give some time Hawk Point to get sell ...

On the other hand ... from my lack of knowledge, I think the soonest Strix Point (Zen 5) hits retailers' shelves ... the more market share AMD will take away from Intel ... at least as far as the laptop market is concerned.


Please ... shed some 'light' ...


Regards.

Any number of reasons. Software not being ready, Market timing, competitive pressures, regulatory environment, old products still being sold in volume, etc.

 

[Kepler_L2]

Soo... how's AMD gonna feed so many hungry Strix Halo CPU, GPU and AI cores?

256-bit LPDDR5X + 32MB Infinity Cache.

 

[FlameTail]

256-bit LPDDR5X + 32MB Infinity Cache.

Is the Infinity Cache exclusive to the GPU or shared with the CPU like an SLC?

 

[FlameTail]

256 bit + LPDDR5X 8533.

= 273 GB/s

 

[Kepler_L2]

Is the Infinity Cache exclusive to the GPU or shared with the CPU like an SLC?

It's an SLC.

 

[FlameTail]

Phoenix : 12 CU RDNA3 : 100%

Strix Point : 16 CU RDNA3.5 : 150%

Strix Halo : 40 CU RDNA3.5: 375%

Theoretical performance estimation

 

[StefanR5R]

This is the original source...

Thanks! Some of the fineprint:

• This article has SPECrate®2017 estimates, once with DDR5-4800 and once with DDR5-6000, which are from 1-copy 1-threaded runs AFAIU.
• The measurements were taken mid September 2022 (or earlier) from engineering samples.
• Edit 1: The article doesn't mention if the figures represent "base" or "peak", and other details. Edit 2: "Base" is less work for the user, so maybe it's "base".

Clock-normalized n-copy or n-threaded estimates (i.e. "rate-n" runs with fixed core clock speed) are certainly out there somewhere too...

Defining real-world benches might be helpful for this discussion?

@ikjadoon, why are you first asking @Markfw to define real-world benchmarks, then leave the real world and provide Cinebench 1T figures? The Cinebench benchmark measures performance of a rendering engine which, in the real world, is used for widely parallel problems and, importantly, in its much more effective GPGPU implementation, not in the plain CPU implementation anymore these days.

I understand that @Markfw was primarily referring to fully parallel scientific computing. E.g. molecular dynamics, telescope data processing, number-theoretical transforms... That's often n×1-copies×threads, sometimes n×m-copies×threads. I am running such stuff myself and occasionally implemented actual reproducible benchmarks based on selected workunits of such science tasks. (But the purpose of such benchmarks was to enable myself and others to optimize BIOS settings, application settings, and task scheduling, not as input to purchasing decisions of end-users or to assess market competitiveness of hardware vendors. Due to this and for lack of spare time, I covered only some number-theoretical transforms so far, not any of the other fields.) Alas I don't have current-generation Intel hardware, nor do I recall anybody who has got my little benchmark frameworks plus Intel gear, in order to be able to post "IPC" a.k.a. clock-normalized benchmark results. Worse, I don't even have any hardware which I could run at fixed core clock frequency, for correct clock-normalized evaluation, apart from some very old kit.

Anyway; this side discussion was started with a statement that Intel had "better IPC" (without further qualification) and the implication that this is one of the aspects why the poster thought that AMD should offer Zen 5 based products as soon as they can. Meanwhile we have seen that there are indeed situations in which higher clock-normalized performance is observed on Intel CPUs, while on the other hand everybody can have a different opinion on how much this fact can or should influence AMD's time-to-market efforts. :-)

 

[uzzi38]

Phoenix : 12 CU RDNA3 : 100%

Strix Point : 16 CU RDNA3.5 : 150%

Strix Halo : 40 CU RDNA3.5: 375%

Theoretical performance estimation

I'll preface this by saying I don't know Strix Point perf numbers, but I still wouldn't expect a 50% bump over PHX personally.

I'm expecting more along the lines of ~25-30% perf at best. Memory bandwidth is the concern really, PHX seems massively mem bw bound by 28w. Gains from increasing power limits after 28w are minimal, despite the iGPU clocks being significantly higher (in games PHX's iGPU sits at 1.3GHz or lower at 15w from my own testing, can't measure at 28w because the GPD Win Mini chassis can't take it, but even 20W sees the iGPU shift up to ~1.7GHz peak - by 28W 2GHz+ is pretty much guaranteed).