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!

 

[JoeRambo]

Epic fail for what? I think some reactions are way exaggerated.

Indeed. Big gains were possible when they were catching up With Zen to state of art for 2013 ( Skylake) and then overtaking it with 2020 state of art in Zen3. Or when Intel jumped from same Skylake to 2020 tech with Golden Cove. From this point big gains require equally big changes to uArch.
IPC gain of 5-10% is perfectly reasonable and combined with new node enabling higher clocks and less power usage => will still result in huge gains everywhere from mobile to servers.

 

[RnR_au]

you guys expect that Zen4 deviates from the above trend of >15% IPC uplift with next gen core while they switch from 7nm to 5nm? That's seriously underestimating AMD's capabilities, even IF they wanted to pursue the "higher clocks"/speed demon design.

We can really only go from what they showed today. And it wasn't >15% IPC uplift.

 

[exquisitechar]

Epic fail for what? I think some reactions are way exaggerated.

To me, it seems like it will be an epic fail in the desktop market. I don’t see how Raphael will be an attractive product compared to Raptor Lake. One year later, with a node advantage, they will just match Alder Lake or fall slightly behind, it seems. And it won’t come cheap, the 6nm IOD seems to be of a similar size to the previous one. I don’t need to even point out that the N5 CCDs won’t be cheap either.

Genoa will still be really good. Perf/watt at those clock speeds should be great and they are increasing the core count.

 

[biostud]

AMD had consistent >15% IPC gains year on year with their Zen roadmap. It was specifically their goal to outpace the industry average of the IPC curve. For them to suddenly massively underperform with Zen4 is totally out of character, especially if they had a new node to play with (which grants them the needed die area for beefed up core logic).

So in summary:
- Zen1/Zen1+ brought >50% over previous gen (Steamroller)
- Zen2 brought ~15% IPC on average Vs Zen1, and 2x the AVX throughput, with 14nm-> 7nm node switch
- Zen3 brought ~19% IPC on average Vs Zen2 , on the same node
- you guys expect that Zen4 deviates from the above trend of >15% IPC uplift with next gen core while they switch from 7nm to 5nm? That's seriously underestimating AMD's capabilities, even IF they wanted to pursue the "higher clocks"/speed demon design. Intel pursued much higher clocks and they achieved ~17-19% IPC targets with Icelake and Alder Lake

In some way I think we have been "spoiled" by AMD expecting they can achieve +20% performance uplift (IPC + increased frequency) with each generation, while intel (until Alder lake) was exactly the opposite. 15% performance increase ST is not stellar, but it is not bad either. I don't think it will be enough to beat Raptor lake, but I think AMD will handle that with vcache CPU at some point.

 

[Hans Gruber]

AMD isn't pushing higher CPU mhz to keep up with intel. That is the benefit of the silicon die shrink from 7nm to 5nm. Let's focus on the positives. AMD will now include RDNA2 integrated graphics. It's a nice to have feature if your GPU goes out or if you are in between GPU's at some point. Regardless of the Zen 4 IPC increase. The rumors are that Zen 4 will be 27-40% faster than Zen 3 in various applications.

The big problem is the fanboys here do not realize that Alder Lake is ahead of Zen 3 and Raptor Lake is going to be released later this year by Intel. That means Zen 4 has to punch way above it's weight considering Alder Lake is what people are comparing Zen 4 against. Raptor Lake is the real problem for Zen 4.

This is why I said a few months ago that AMD should release Zen 4 ASAP. The longer it takes for Zen 4 to get to market. The more time Intel has to tweak Raptor Lake and the less time Zen 4 has as the CPU king. Nobody cares about power efficiency when it comes to who has the fastest and most powerful CPU. I say this because Intel has been much better with managing power use. AMD still wipes the floor with energy efficiency and will continue this with Zen 4 because of the 5nm process.

After Raptor Lake, Intel shrinks down to 7nm. Raptor Lake is the final Intel CPU on 10nm. I would think that AMD should be worried about Intel considering Zen 4 has not made it to market yet. They say Zen 5 is going to be released in 2024. By that time Intel may have wiped AMD off their windshield and put in the rear view mirror.

AMD would have been better served releasing Zen 4 at the beginning of 2022. I am kind of in disbelief that Zen 4 will not have either a 24core 48 thread or a 32core 64thread CPU. I guess AMD is holding back until they see what Raptor Lake has to offer. Regardless, AMD has had a great 5 year run.

 

[Abwx]

Epic fail for what? I think some reactions are way exaggerated.

At 65W it should perform like a 12900K@180W and a 5950X@125W but there s still people who would find this underwhelming.

 

[inf64]

We can really only go from what they showed today. And it wasn't >15% IPC uplift.

And the only thing we have is >15% ST uplift in one benchmark that is not cache sensitive, all done on a pre-prod. sample running at unknown clock speeds. Too much is unknown to call it a failure at this point. If all they managed to get is a clock boost with minor IPC while using the 5nm node then something is way off versus all the previous Zen cores.

 

[igor_kavinski]

At 65W it should perform like a 12900K@180W and a 5950X@125W but there s still people who would find this underwhelming.

That would be excellent if that happens to be true.

By the way, I'm guessing AMD is sandbagging Intel, not us.

 

[yuri69]

And the only thing we have is >15% ST uplift in one benchmark that is not cache sensitive, all done on a pre-prod. sample running at unknown clock speeds. Too much is unknown to call it a failure at this point. If all they managed to get is a clock boost with minor IPC while using the 5nm node then something is way off versus all the previous Zen cores.

While I agree about the "single benchmark sux" part, let's not get ourselves into the 'JF-AMD' style of damage control...

The Zen IPC was improved gradually, but the achieved frequency was not as high as Intel designs. Both seem to change with Zen 4. As weird it might look it is still a valid design choice.

 

[RnR_au]

And the only thing we have is >15% ST uplift in one benchmark that is not cache sensitive, all done on a pre-prod. sample running at unknown clock speeds. Too much is unknown to call it a failure at this point. If all they managed to get is a clock boost with minor IPC while using the 5nm node then something is way off versus all the previous Zen cores.

So you and @Saylick seems to be arguing that their >15% ST uplift benchmark will be on the far left handside on a graph like this;



That AMD is 'sandbagging' for whatever reason.

If true, its weird marketing. They had no issues in showing 5.52Ghz in a gaming scenario, so I can't understand why they couldn't have thrown in a cherry picked benchmark. Its not like they were running out of time during this presser... ie pressure sensitive wasd keys <scratches head>.

 

[inf64]

While I agree about the "single benchmark sux" part, let's not get ourselves into the 'JF-AMD' style of damage control...

The Zen IPC was improved gradually, but the achieved frequency was not as high as Intel designs. Both seem to change with Zen 4. As weird it might look it is still a valid design choice.

I've just checked Zen2->Zen3 uplift in R23, it's ~14% (iso clocks; source). The best case scenario for what AMD showed today is the Zen4 sample running ST ~5.2Ghz boost (my guess) that achieves ~15% higher ST score in R23 vs 5950X. That would put IPC jump in R23 to around ~10%. R23 had much lower gain on Zen2->Zen3 (~14%), which is a diminishing return VS Zen1+->Zen2 (~18% at iso clock; source). It looks like we have diminishing IPC gains in R23, at least for AMD (intel had no problems getting >15+ IPC with their last 2 new cores).

So you and @Saylick seems to be arguing that their >15% ST uplift benchmark will be on the far left handside on a graph like this;

That AMD is 'sandbagging' for whatever reason.

If true, its weird marketing. They had no issues in showing 5.52Ghz in a gaming scenario, so I can't understand why they couldn't have thrown in a cherry picked benchmark. Its not like they were running out of time during this presser... ie pressure sensitive wasd keys <scratches head>.

That's my guess at least. See my post above.

 

[mikk]

I think in a best case there is hope for 10% improved IPC and in a worst case closer to 5%. I don't think Cinebench is far off from the average in a number of applications, it was quite accurate in the past if I remember.

 

[Asterox]

AMD isn't pushing higher CPU mhz to keep up with intel. That is the benefit of the silicon die shrink from 7nm to 5nm. Let's focus on the positives. AMD will now include RDNA2 integrated graphics. It's a nice to have feature if your GPU goes out or if you are in between GPU's at some point. Regardless of the Zen 4 IPC increase. The rumors are that Zen 4 will be 27-40% faster than Zen 3 in various applications.

The big problem is the fanboys here do not realize that Alder Lake is ahead of Zen 3 and Raptor Lake is going to be released later this year by Intel. That means Zen 4 has to punch way above it's weight considering Alder Lake is what people are comparing Zen 4 against. Raptor Lake is the real problem for Zen 4.

This is why I said a few months ago that AMD should release Zen 4 ASAP. The longer it takes for Zen 4 to get to market. The more time Intel has to tweak Raptor Lake and the less time Zen 4 has as the CPU king. Nobody cares about power efficiency when it comes to who has the fastest and most powerful CPU. I say this because Intel has been much better with managing power use. AMD still wipes the floor with energy efficiency and will continue this with Zen 4 because of the 5nm process.

After Raptor Lake, Intel shrinks down to 7nm. Raptor Lake is the final Intel CPU on 10nm. I would think that AMD should be worried about Intel considering Zen 4 has not made it to market yet. They say Zen 5 is going to be released in 2024. By that time Intel may have wiped AMD off their windshield and put in the rear view mirror.

AMD would have been better served releasing Zen 4 at the beginning of 2022. I am kind of in disbelief that Zen 4 will not have either a 24core 48 thread or a 32core 64thread CPU. I guess AMD is holding back until they see what Raptor Lake has to offer. Regardless, AMD has had a great 5 year run.

Yes, or again very important is Encoding/Decoding VCN hardware found on every Ryzen APU.

Page 169 - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000)

Page 169 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Abwx]

That would be excellent if that happens to be true.

By the way, I'm guessing AMD is sandbagging Intel, not us.

Those benches numbers from AMD are accurate imo.

They cant state something else than a lower limit for CB ST because this has to hold true for all SKUs, including the 8C that will clock less than the top part.

About power they stated 2x the efficency at same design, with an higher IPC they can get even further than 2x the efficency at isoperf.

 

[deasd]

The big problem is the fanboys here do not realize that Alder Lake is ahead of Zen 3 and Raptor Lake is going to be released later this year by Intel.

hate to say this in AMD thread but last rumor I heard RaptorLake's release is much later than we thought. Just don't set anything in stone and being cautious. I didn't mention this in Intel thread is because I'm not sure the source is trustworthy....

'Alder Lake is ahead of Zen 3' Yes. and part of it is generated by ADL's much aggressive turbo/frequency.... but now what we are slamming is the Zen4 IPC uplift is smallish and nobody complaining about 5.5Ghz+ gaming rehearsal.

Since the release is set in Fall there's still long time for arguing and things might be changed. Not to mentioned the detail provided in AMD's slide right now is ambiguous

 

[DrMrLordX]

Everybody talks about AVX-512 like it is the holy grail, and I've never actually understood if it was important to me as a gamer/photo editor, or who really benefits from having these.

They do? Most folks around here seem to agree that AVX512 is niche. It has a lot of problems associated with it.

Raptor Lake is the real problem for Zen 4.

Red herring. Zen4's main competition was meant to be Granite Rapids (it is instead Sapphire Rapids). Raphael is just another generation of x86 CPUs designed for workstation/server and adapted for use in consumer desktops. To AMD's credit, this strategy has worked up 'til now, and may continue to be quite functional for them. But, in terms of desired market share and eventual revenue, Genoa is far more important to them.

I fully expect that most of the design efforts behind Zen4 were aimed at Genoa and its associated markets.

 

[yuri69]

The big problem is the fanboys here do not realize that Alder Lake is ahead of Zen 3 and Raptor Lake is going to be released later this year by Intel. That means Zen 4 has to punch way above it's weight considering Alder Lake is what people are comparing Zen 4 against. Raptor Lake is the real problem for Zen 4.

This is why I said a few months ago that AMD should release Zen 4 ASAP. The longer it takes for Zen 4 to get to market. The more time Intel has to tweak Raptor Lake and the less time Zen 4 has as the CPU king. Nobody cares about power efficiency when it comes to who has the fastest and most powerful CPU. I say this because Intel has been much better with managing power use. AMD still wipes the floor with energy efficiency and will continue this with Zen 4 because of the 5nm process.

After Raptor Lake, Intel shrinks down to 7nm. Raptor Lake is the final Intel CPU on 10nm. I would think that AMD should be worried about Intel considering Zen 4 has not made it to market yet. They say Zen 5 is going to be released in 2024. By that time Intel may have wiped AMD off their windshield and put in the rear view mirror.

AMD would have been better served releasing Zen 4 at the beginning of 2022. I am kind of in disbelief that Zen 4 will not have either a 24core 48 thread or a 32core 64thread CPU. I guess AMD is holding back until they see what Raptor Lake has to offer. Regardless, AMD has had a great 5 year run.

True dat.

ADL is often >15% ahead of Zen 3. This is understandable since Zen 3 is a 2020 tech.
H2 2022 is for RPL adding a few % above ADL. Zen 4 launches around that time.
H2 2023 sees Zen "4D" against MTL launched around that time.
Then comes Zen 5 - given 12-18 (23 for Zen 4) months launch window it can be a H2 2023 at earliest. 2024 is more realistic tho.

 

[Karnak]

The big problem is the fanboys here do not realize that Alder Lake is ahead of Zen 3 and Raptor Lake is going to be released later this year by Intel. That means Zen 4 has to punch way above it's weight considering Alder Lake is what people are comparing Zen 4 against. Raptor Lake is the real problem for Zen 4.

This is why I said a few months ago that AMD should release Zen 4 ASAP. The longer it takes for Zen 4 to get to market. The more time Intel has to tweak Raptor Lake and the less time Zen 4 has as the CPU king. Nobody cares about power efficiency when it comes to who has the fastest and most powerful CPU. I say this because Intel has been much better with managing power use. AMD still wipes the floor with energy efficiency and will continue this with Zen 4 because of the 5nm process.

After Raptor Lake, Intel shrinks down to 7nm. Raptor Lake is the final Intel CPU on 10nm. I would think that AMD should be worried about Intel considering Zen 4 has not made it to market yet. They say Zen 5 is going to be released in 2024. By that time Intel may have wiped AMD off their windshield and put in the rear view mirror.

"Zen4" is not an architecture just for desktop only. Besides that RTL will get slaughtered in terms of efficiency - which is one of the most important factors out there, like it or not. Genoa will do the same with everything Intel has to offer for servers. Not to mention the huge MT advantage it will have. MT and efficiency Intel is not even close to what AMD has to offer. Even Milan-X right now is a problem for Intel.

In terms of gaming and pure ST (which is only relevant there and maybe for a few prosumer) AMD can just add cache to get massive performance boosts (which also works in the server space) while also increasing the efficiency a lot. 5800X3D says hello.

Because of that Intel needs a V-Cache counter asap. They are behind in literally everything besides ST. V-Cache changes that for games and AMD but RTL won't change that for Intel in general.

 

[Timorous]

So we have a 16 core engineering sample being 45% faster (31% less time to complete is 45% faster) than a 12900K in a sub 300s blender test. Tests that length tend to favour the Intel boost alg where as longer tests work out better for the 5950X.

We also have a ST test vs a 5950X with a 15% performance uplift in CB R23 but at unknown clocks with unknown scores. Still 15% faster than the 5950X is around 1890 which is a long way behind the faster 12th gen parts.

OTOH when looking at the IPC gains from Zen 2 to Zen 3 CB R23 was on the low end so that might be the case again here.

I do get the impression AMD did not want to give Intel too much to aim at with Raptor Lake but if AMD release Zen 4 before Intel release Raptor Lake then Intel can play with the exact SKUs anyway.

 

[coercitiv]

Raptor Lake is the real problem for Zen 4.

If Raptor Lake were the real problem for Zen 4 then we wouldn't have a problem, since Zen 4 will be closer to Raptor Cove than Zen3 is to Golden Cove. Friendly reminder, for the second time today, that AMD is not the only one who presented the public with underwhelming news regarding their upcoming CPU architecture:



So according to marketing departments from both companies:

• Raptor Cove brings up to 10-15% more performance over Golden
• Zen 4 bring more than 15% more performance over Zen 3

Choose your underwhelming upgrade now!

 

[Abwx]

Hans de Vries has some numbers for CB MT :

https://twitter.com/x/status/1528719223786192896

 

[biostud]

If Raptor Lake were the real problem for Zen 4 then we wouldn't have a problem, since Zen 4 will be closer to Raptor Cove than Zen3 is to Golden Cove. Friendly reminder, for the second time today, that AMD is not the only one who presented the public with underwhelming news regarding their upcoming CPU architecture:

View attachment 61966

So according to marketing departments from both companies:

• Raptor Cove brings up to 10-15% more performance over Golden
• Zen 4 bring more than 15% more performance over Zen 3

Choose your underwhelming upgrade now!

Looking forward to performance/watt comparisons.

 

[Timmah!]

16C are PLENTY for tasks CPUs excel at and given AMD's focus on clocks, 5+Ghz MT performance will be awesome.
I fully expect it to beat 13900K in Blender and probably run around equally in Cinebench ( cause Blender is heavier on FP and loves stronger cores, while Intel's small cores are good in CB23 and less strong in previuos CBs that were FP heavier).

The real question is, why would anyone run rendering workload on CPU in year 2022?

View attachment 61945


Yes, 3090 undervolted is 27 times faster than eight 5Ghz Alder Lake P cores without HT. Like 230W vs 130W too... I am sure 4090 will increase the gap to 40-45 times this summer.

this is true. Rendering on CPU nowadays is soo 2000 personally i moved to octane render back in 2009 and never looked back.

then again, for example aforementioned octane has a certain part of the workflow, that is heavily CPU dependent - its the part, where the mesh is loaded into the app and its compiled - i presume this process is well multithreaded and more cores = faster time. So even more than 16 cores would come handy.

finally, it kinda sucks that nowadays, with no intel hedt platform and new threadripper not available to consumer clients anymore , 5 years ago you could actually buy cpu with more cores (18 core skylake) than you can buy now, when it tops at this 16 core ryzen or 12900k. Not to mention 24, 32 and 64 core 3000 series TRs In later years. Its rather disappointing period to buy cpu, really. i almost regret not getting 3960x, while i could, but i was still ok with what i had and did not expect this regress to happen.

 

[Panino Manino]

Oh, already revealed?

Like I feared, Zen 4 is just a small upgrade over Zen 3 focusing on it's weak points, AVX512, "AI", and integrated video encoder to help some tasks.
15% percent with clock increase? Zen Reign is over.
The saving grace is that all cores are equal, but makes me wonder if the 16 core limit means that there's a possibility of an extra 24 and 32 cores chip using Zen4c cores.

 

[Abwx]

15% percent with clock increase? Zen Reign is over.

According to the numbers from Hans de Vries in CB R23 MT and comparatively to Zen 3 :

38% better perf for the 6C
33% for the 8C
32% for the 12C
45% for the 16C.

The huge gap for the 16C is due to TDP increased, at same power the improvement should be closer to the other SKUs.