Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[DisEnchantment]

So the question is, are Zen 4's reputed IPC gains a range, or an average. If it's the former, then that won't be very impressive you ask me but if it's an average, then it certainly will be. As for what he says about Genoa, a lot of that could be due to DDR5's bandwidth. I'm not trying to downplay Zen 4 as I'm sure it will be a really good CPU, but I think there's too many here inserting their own confirmation bias into rumors about Zen 4's IPC gains, which are subject to interpretation.

There is no reliable data.
But you can just ask yourself few questions and try to connect the dots instead of just listening to YouTube click baiters.
(Personally I cannot stand AdoredTV, his arrogance is unbearable. RGT takes 10 mins of my time instead of reading one line of tweet from Greymon or Execufix, and MLID's attitude is like he is shining light on the world of tech darkness)
I am sure you will have a more or less consistent idea.

What will AMD do in 2022 with Zen 4 vs 2020 with Zen 3 considering the following

• 2x transistor density offered by N5
• 2x efficiency gain offered by N5
• 1.25x performance gain offered by N5
• 1.6x more TDP offered by AM5
• 2 more years of development
• massive boost in R&D in previous 2 years.
• New packaging concepts available like SoIC stacking and EFB
• DDR5

 

[gruffi]

This could be interpreted in so many ways. I'm no engineer or industry professional, but it really does appear to me that IPC is a nebulous concept which is difficult to measure properly from one architecture to another. According to Intel, Golden Cove has 19% greater IPC than Cypress Cove, but in many workloads and apps Golden Cove is much faster. I've also seen several forum members claim that Golden Cove's IPC advantage over Zen 3 is in the single digit range, which is crazy to me because in several workloads, Golden Cove straight up destroys Zen 3 by margins I haven't seen in years when it comes to CPU performance.

What? Are you cherry picking? Golden Cove is far from "destroying Zen 3". Zen 3 compared to 10th gen was more impressive than 12th gen compared to Zen 3 considering IPC. The advertised IPC improvements of AMD and Intel are average. But of course not every test suite will show that numbers. Golden Cove might have higher than advertised IPC improvements in some scenarios. But the same applies to Zen 3. And Zen 4 won't be any different. In other scenarios Golden Cove's IPC is hardly higher than Willow Cove or Zen 3.

 

[Carfax83]

There is no reliable data.
But you can just ask yourself few questions and try to connect the dots instead of just listening to YouTube click baiters.
(Personally I cannot stand AdoredTV, his arrogance is unbearable. RGT takes 10 mins of my time instead of reading one line of tweet from Greymon or Execufix, and MLID's attitude is like he is shining light on the world of tech darkness)
I am sure you will have a more or less consistent idea.

What will AMD do in 2022 with Zen 4 vs 2020 with Zen 3 considering the following

• 2x transistor density offered by N5
• 2x efficiency gain offered by N5
• 1.25x performance gain offered by N5
• 1.6x more TDP offered by AM5
• 2 more years of development
• massive boost in R&D in previous 2 years.
• New packaging concepts available like SoIC stacking and EFB
• DDR5

This is a much more sober assessment if you ask me and you make solid points. AMD has a lot going for it this cycle and can potentially leapfrog Intel if they play their cards right. If that 25% figure is really an average, then that would be very impressive indeed and would be the biggest jump in IPC in x86-64 CPUs since the original Core series CPUs post Pentium IV era.

 

[Carfax83]

What? Are you cherry picking? Golden Cove is far from "destroying Zen 3". Zen 3 compared to 10th gen was more impressive than 12th gen compared to Zen 3 considering IPC.

I said "in several workloads," not across the board. Those workloads are predominantly CPU limited games, single threaded applications, code compiling and encoding. When I was looking at the reviews, those workloads stood out to me as great representations of Golden Cove's performance advantage over Zen 3.

The advertised IPC improvements of AMD and Intel are average. But of course not every test suite will show that numbers.

Which is my point. The rumors surrounding Zen 4's IPC obviously don't specify whether it's an average or a range. Based on what I've seen, it could be interpreted either way it seems.

Case in point, that excerpt from Chips and Cheese on the previous page said:

From IPC gains over 25%

The wording to me seems like he's indicating more of a range than an average.

Golden Cove might have higher than advertised IPC improvements in some scenarios. But the same applies to Zen 3. And Zen 4 won't be any different. In other scenarios Golden Cove's IPC is hardly higher than Willow Cove or Zen 3.

Yes I agree, which is why I said IPC is inherently nebulous and difficult to measure, especially across architectures because it can vary so much across applications. But a lot of people are acting as though Zen 4 will have an average improvement of 25%, which would be really astonishing if true.

With Zen 3, it seemed though AMD used Spec 2017 as the source of their 19% IPC uplift over Zen 2. I hope they don't just rely on Spec this time, and include other applications like what Intel did to get their 19% IPC figure for Golden Cove.

Zen 3:

Golden Cove:

 

[exquisitechar]

There is no reliable data.
But you can just ask yourself few questions and try to connect the dots instead of just listening to YouTube click baiters.
(Personally I cannot stand AdoredTV, his arrogance is unbearable. RGT takes 10 mins of my time instead of reading one line of tweet from Greymon or Execufix, and MLID's attitude is like he is shining light on the world of tech darkness)
I am sure you will have a more or less consistent idea.

What will AMD do in 2022 with Zen 4 vs 2020 with Zen 3 considering the following

• 2x transistor density offered by N5
• 2x efficiency gain offered by N5
• 1.25x performance gain offered by N5
• 1.6x more TDP offered by AM5
• 2 more years of development
• massive boost in R&D in previous 2 years.
• New packaging concepts available like SoIC stacking and EFB
• DDR5

This is what makes me hopeful as well. Zen 3 was a very good improvement for a relatively small increase in xtor count (with a mostly untouched IOD, too), although it was a “from the ground up” new core according to AMD and Zen 4 is supposed to be more of an evolution like Zen 2. If Zen 4 is just similar to Zen 2 like the less optimistic predictions suggest, it looks like a miss from AMD considering what it has going for it. I would understand if it was a fast follow up to Zen 3 and more like a “tick”, but it isn’t. Where did all those transistors go in that case?

 

[yuri69]

Taking transistor count figures inferred from marketing slides as *the thing* is misleading. There is always the dark silicon, different EDA libs, different optimization targets, etc.

After Skylake, Intel got a great deal of transistors for their 10nm processors. The results were not stellar.

 

[Hitman928]

With Zen 3, it seemed though AMD used Spec 2017 as the source of their 19% IPC uplift over Zen 2. I hope they don't just rely on Spec this time, and include other applications like what Intel did to get their 19% IPC figure for Golden Cove.

The plot you are showing for Zen 3 over Zen 2 is for Spec2006, not Spec2017. Spec 2017 is a much better representation than 2006 and is what Anandtech uses to check IPC gains. I'm not sure AMD ever says specifically the exact collection of tests they use to get their IPC numbers. Additionally, Zen3 IPC gains were confirmed by multiple independent reviewers all running different test suites so no weird benchmark shenanigans going on. Anandtech's Spec2017 tests show GC with about a 10% lead in FP IPC over Zen 3 and a very slight lead in INT IPC.

 

[Carfax83]

The plot you are showing for Zen 3 over Zen 2 is for Spec2006, not Spec2017. Spec 2017 is a much better representation than 2006 and is what Anandtech uses to check IPC gains. I'm not sure AMD ever says specifically the exact collection of tests they use to get their IPC numbers. Additionally, Zen3 IPC gains were confirmed by multiple independent reviewers all running different test suites so no weird benchmark shenanigans going on. Anandtech's Spec2017 tests show GC with about a 10% lead in FP IPC over Zen 3 and a very slight lead in INT IPC.

Yeah I wasn't trying to imply that there was any shenanigans going on with the reported IPC gains from AMD for Zen 3, I just personally don't like Spec. It just doesn't seem to reflect the wide range I see in performance between applications between CPUs. Case in point, according to Spec2017, Golden Cove is just barely faster than Zen 3 in code compiling, but when I see actual benchmarks from compiling large code bases, there is a much bigger gap.

 

[Hitman928]

Yeah I wasn't trying to imply that there was any shenanigans going on with the reported IPC gains from AMD for Zen 3, I just personally don't like Spec. It just doesn't seem to reflect the wide range I see in performance between applications between CPUs. Case in point, according to Spec2017, Golden Cove is just barely faster than Zen 3 in code compiling, but when I see actual benchmarks from compiling large code bases, there is a much bigger gap.

Do you have an example?

 

[deasd]

Charlie teases about Bergamo(Zen4C):

https://twitter.com/x/status/1493977701945860101

 

[xilli_fiberbit]

Nice diagram to have a clear view of AVX512 supports in ZEN4/ZEN4c, no sure for now for Raphael :

https://twitter.com/x/status/1493309403533135878

 

[coercitiv]

With Zen 3, it seemed though AMD used Spec 2017 as the source of their 19% IPC uplift over Zen 2

Or they used the geomean of multiple benchmarks, workloads and games.

 

[yuri69]

Do you have an example?

Mere 4.4GHz 12400 6c often notably beats the 5+GHz 5950 16c in serial tasks:

AMD Ryzen 9 5950X vs. Intel Core i5-12400 Benchmarks - OpenBenchmarking.org

openbenchmarking.org

For compilation tasks, the 4.4GHz 12400 6c is a bit faster than 4.6GHz 8c 5700G:

https://openbenchmarking.org/embed.php?i=2201128-PTS-ALDERLAK15&sha=c7332f9b9869&p=2

 

[Hitman928]

For compilation tasks, the 4.4GHz 12400 6c is a bit faster than 4.6GHz 8c 5700G:

https://openbenchmarking.org/embed.php?i=2201128-PTS-ALDERLAK15&sha=c7332f9b9869&p=2

Compiling is typically very sensitive to cache size and so the Zen 3 APUs with 1/2 the L3 cache will under perform the competition. Desktop/server CPUs with full cache per CCD are a different story. I also expect that ADL benefits from larger memory bandwidth when using DDR5 as large compilation projects typically like large memory bandwidth as well, but even then, the 6c ADL is not going to beat an 8c Zen3 with fully enabled cache. Core for core, comparing GC and Zen3 in compilation, they are probably pretty close but Zen4 with DDR5 will probably crush both of them.

 

[JoeRambo]

I also expect that ADL benefits from larger memory bandwidth when using DDR5 as large compilation projects typically like large memory bandwidth as well, but even then, the 6c ADL is not going to beat an 8c Zen3 with fully enabled cache.

It actually does beat 8C Zen3 in said compilation tests in Phoronix testing:




Lowly 12400 with 4.4Ghz turbo clocks and 18MB of L3 has no trouble keeping up with premier AMD 8C cpu with 32MB of L3.

People keep conveniently forgetting that ADL big core is 5 ALU (each LEA capable), 512 ROB, mostly 6-wide monster. It is let down currently by mobile phone worthy uncore and memory controller, but expect it to scale real well in the future with faster mem.

The only "complex" benchmark i need is Web Speedometer 2.0 (thread is in on this very forum). Since i own 10900K, 5950X and 12900K, the numbers with 3800-3900 HIGHLY tuned DDR4 are the following ~210, ~255 and 330. That's how faster and more smooth ADL is. I believe 12900K does like 300 on stock as well. ~25% faster in my eyes.

 

[Hitman928]

It actually does beat 8C Zen3 in said compilation tests in Phoronix testing:

View attachment 57517


Lowly 12400 with 4.4Ghz turbo clocks and 18MB of L3 has no trouble keeping up with premier AMD 8C cpu with 32MB of L3.

People keep conveniently forgetting that ADL big core is 5 ALU (each LEA capable), 512 ROB, mostly 6-wide monster. It is let down currently by mobile phone worthy uncore and memory controller, but expect it to scale real well in the future with faster mem.

The only "complex" benchmark i need is Web Speedometer 2.0 (thread is in on this very forum). Since i own 10900K, 5950X and 12900K, the numbers with 3800-3900 HIGHLY tuned DDR4 are the following ~210, ~255 and 330. That's how faster and more smooth ADL is. I believe 12900K does like 300 on stock as well. ~25% faster in my eyes.

You said it beats the 8c Zen3 and then post plots showing that it doesn't?

 

[JoeRambo]

You said it beats the 8c Zen3 and then post plots showing that it doesn't?

News at eleven huh? I even wrote that slowest 6C with least cache atm has no trouble matching CPU that turboes 500mhz more and has almost twice L3 and two more big cores?

 

[Thibsie]

News at eleven huh? I even wrote that slowest 6C with least cache atm has no trouble matching CPU that turboes 500mhz more and has almost twice L3 and two more big cores?

Well, you wrote 'beats' didn't you ?

 

[Hitman928]

News at eleven huh? I even wrote that slowest 6C with least cache atm has no trouble matching CPU that turboes 500mhz more and has almost twice L3 and two more big cores?

Then why write this?

It actually does beat 8C Zen3 in said compilation tests

You 'corrected' my statement with evidence that proved my statement?

In terms of comparison of cores, yes, the Zen3 has 78% more L3, but the 12400 has 87.5% more L2 cache and the advantage of DDR5. The Linux Kernel is also just 1 compilation example and in the grand scheme of things, isn't that large of a project. Other projects of course can show different results. Like the results below from Gamers Nexus where the 12400 is 10% faster than the 5600x but will also use about 10% more power in all core loads. Frequency wise, the 5600x probably has a little bit of an advantage, but overall, a Zen 3 core with fully enabled L3 isn't going to be very far behind a GC core which was what was being contested to begin with and I have yet to see any evidence to the contrary outside of specific cases that can go either way.


source

 

[JoeRambo]

You 'corrected' my statement with evidence that proved my statement?

Well, you wrote 'beats' didn't you ?

Are you guys having trouble imagining a proper 6C ADL without small cores? One that is not 4.4ghz? One that is not with 18MB cache? The @Hitman928 was making bold claims about Zen3, thanks god that 12400 did not actually beat that CPU.

Again. See 12400? See 12600K? See where 5800X is? Easy to imagine hypothetical 5Ghz boosting 6C doing real good without small cores. Just look how good 12700K with similar clocks and same 4E cores does due to having 33% more big cores.

 

[Thibsie]

Hypothetically a lot of things but that ain't gonna make you more right or more wrong (or any of us).

If anything, you made a claim and you can't back it with anything but hypothetical evidence which means pretty much nothing.

Any claim is welcome (hypothetically). Just prove it.

 

[Hitman928]

Are you guys having trouble imagining a proper 6C ADL without small cores? One that is not 4.4ghz? One that is not with 18MB cache? The @Hitman928 was making bold claims about Zen3, thanks god that 12400 did not actually beat that CPU.






Again. See 12400? See 12600K? See where 5800X is? Easy to imagine hypothetical 5Ghz boosting 6C doing real good without small cores. Just look how good 12700K with similar clocks and same 4E cores does due to having 33% more big cores.

So a hypothetical 6C ADL clocked at 5 GHz with 30 MB of L3 cache could perform better? Well, so could a hypothetical 6 core Zen 3 at 5 GHz with 64 MB of V-cache added. I don't want to play the hypothetical game so if that's what you want to debate, I'm out of that conversation.

In terms of 12700k versus 12600k, it's not just that it has 33% more P-cores, it also has 25% more L3 cache, 27.5% more L2 cache, and it is also allowed to use around 33% more power so it can add those cores and cache without dropping frequency at all under all core loads and should boost higher when not completely loaded.

Just for clarity, I don't think anyone was claiming that GC doesn't have higher IPC and performance than Zen 3, just that GC doesn't "destroy" Zen 3 which is what the claim was. My only claim was that a 6c ADL is not going to outperform a fully enabled 8c Zen 3 chip in compiling (talking real products here), that Zen 3 isn't that far behind GC in IPC, and that Zen 4 will probably trounce them both in compiling.

 

[Markfw]

Since this is the Zen 4 thread, maybe we could talk about that and stay on-topic.

NO ADL, and NO Zen 3, except comparing Zen 3 to Zen 4.

 

[JoeRambo]

Just for clarity, I don't think anyone was claiming that GC doesn't have higher IPC and performance than Zen 3, just that GC doesn't "destroy" Zen 3 which is what the claim was. My only claim was that a 6c ADL is not going to outperform a fully enabled 8c Zen 3 chip in compiling (talking real products here), that Zen 3 isn't that far behind GC in IPC, and that Zen 4 will probably trounce them both in compiling.

The train of thought "6C ADL is not going to outperform fully enabled 8C ZEN3" => is shown lowest i5 sku 12400 coming within several % => "ZEN3 IPC is not far behind GC"

Any claim is welcome (hypothetically). Just prove it.

Are you guys are somehow missing the 6C vs 8C. Yeah, enough of this ADL talk for sure. Too bad Intel did not release 6C with 3Ghz clock and without L3 cache, now that SKU would have some merits discussing the real products and give me real trouble comparing 6C vs 8C /s

Remains to be seen how good Zen4 can do. It sure will have advantages, but if it is same Zen3 core + AVX512 + 5ghz all core, I expect Intel to remain competitive in performance.

 

[esquared]

Enough of Alder lake. This thread topic is:
Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000)

If you want to discuss alder lake, go to the alder lake thread. Not here.


esquared
Anandtech Forum Director