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

Page 210 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Vattila

Senior member
Oct 22, 2004
799
1,351
136
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).

Untitled2.png


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! :)
 
Last edited:
  • Like
Reactions: richardllewis_01

Frenetic Pony

Senior member
May 1, 2012
218
179
116
Since 24c is not happening and it seems there is not really a much place under the IHS to fit additional chiplets (for whatever reason i thought it is, i remember seeing some mockup scheme showing it should be possible, but cant find it anymore), do you think AMD does not really count with more than 16 core on mainstream platform in upcoming years? I mean, throughout the life of AM5? Or do you think future chiplets (say as soon as in Zen5) are going to be smaller or pack more cores?

EDIT: found that mockup: https://pbs.twimg.com/media/FGBnRKQXoAADm3G?format=jpg&name=large

Saw quite reasonable sounding speculation that Zen 4c is just Zen 4 with SRAM moved to stack caches. Thus why 50% more cores are possible in what we might reasonably assume is the same socket. If that's the case for servers that could be the case for desktop as well. Could still see an HEDT 24 core CPU launch second half of next year or so.
 
  • Like
Reactions: Tlh97 and Vattila

trivik12

Senior member
Jan 26, 2006
252
205
116
Is there a chance that AMD increased the pipeline to hit higher clock speeds. That would explain why they focused on ST improvement vs IPC gains when previously for Zen 3 they focused on IPC gains. Its easier to sell higher clock speed than IPC gains. Only hardcore techies care about IPC.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Do you know this to be true or you just guessing?

Leakers said up to 37% ST uplift for Zen 4 and people hyped Zen 4 so much. Yeah, after this Zen 4 reveal I am not trusting anymore leaks.

It took AMD nearly 24 months to deliver Zen 4 and we shall see later this year if Zen 4 "wrecks Intel".
OK, AMD is the uncontested winner in server, HEDT and desktop productivity. AMD is pretty much the gaming king, but there are a few exceptions. AMD is really strong in the mobile/laptop space, but I don't follow that as closely. In efficiency, AMD is currently king. The 6000 series appears to be a winner, but I will leave that one as a tie.

You don't think that already "wrecks Intel" ? And this is before Zen 4, which everybody seems to think will be at least 15% faster. And until I see any real benchmarks, AMD has so surpassed Intel in the server space, there is not even a date when I think they will be competitive.

Edit: Oh, and integrated graphics, I believe that AMD has that locked up as the best....
 

poke01

Senior member
Mar 8, 2022
651
611
106
OK, AMD is the uncontested winner in server, HEDT and desktop productivity. AMD is pretty much the gaming king, but there are a few exceptions. AMD is really strong in the mobile/laptop space, but I don't follow that as closely. In efficiency, AMD is currently king. The 6000 series appears to be a winner, but I will leave that one as a tie.

You don't think that already "wrecks Intel" ? And this is before Zen 4, which everybody seems to think will be at least 15% faster. And until I see any real benchmarks, AMD has so surpassed Intel in the server space, there is not even a date when I think they will be competitive.

Edit: Oh, and integrated graphics, I believe that AMD has that locked up as the best....
In laptops AMD is weak in terms of actually having latest 6000 series AMD laptops in stock in stores. Intel has lots of 12th gen laptops in stock online and in-store.

Zen 4 is competing with 13th gen Raptor Lake and Intel will finally have enough threads to match AMD's 32T chip. Productivity I think will be similar.

Yes AMD is king in server. However productivity and gaming and the consumer market AMD and Intel will trade blows.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
In laptops AMD is weak in terms of actually having latest 6000 series AMD laptops in stock in stores. Intel has lots of 12th gen laptops in stock online and in-store.

Zen 4 is competing with 13th gen Raptor Lake and Intel will finally have enough threads to match AMD's 32T chip. Productivity I think will be similar.

Yes AMD is king in server. However productivity and gaming and the consumer market AMD and Intel will trade blows.
I was talking today. When Zen 4 and Raptor lake comes out, we will see. And having stock on the 6000 series, maybe they are outselling so much, there is not enough stock ? Any numbers there ? Or just you can;t find one.?
 
  • Like
Reactions: Tlh97 and Drazick

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
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.

In the AMD footnotes, the Zen 4 chip was 46% faster than the 12900k.

People are losing their minds over nothing.

EDIT: We will see what the results are in other workloads. However, Raptor Lake likely won't be 40+% faster unless Intel raises power limits. However, 46% faster while consuming at least 25-35% less power in blender? that seems like a huge win to me. Oh, and they are officially beating Intel at the frequency game as well. 5.52ghz while gaming, that is higher than the 12900ks gets unless you overclock the daylights out of it. Not trying to shill for AMD, however, I see a lot of folks who were disappointed at the limited data shown. I'm just baffled.
 
Last edited:

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
Do you know this to be true or you just guessing?

Leakers said up to 37% ST uplift for Zen 4 and people hyped Zen 4 so much. Yeah, after this Zen 4 reveal I am not trusting anymore leaks.

It took AMD nearly 24 months to deliver Zen 4 and we shall see later this year if Zen 4 "wrecks Intel".

No and no, as far i now there is only a couple of "reliable leakers"= none of them claimed this red blah. :mask:

If you count leakers from Klingon, hm this is kind of complicated situation. :grinning:
 
Last edited:
  • Haha
Reactions: Tlh97 and MangoX

poke01

Senior member
Mar 8, 2022
651
611
106
In the AMD footnotes, the Zen 4 chip was 46% faster than the 12900k.
At blender which is MT based. Not ST. Zen 4 MT will be killer due to higher clocks. But the >15% ST uplift on Cinebench R23 compared to 5950X just makes it equal to 12900K.

Rememeber Intel will be doubling e-cores on desktop. AMD has nothing of value in the mid to low end
 

Schmide

Diamond Member
Mar 7, 2002
5,581
712
126
The 6nm IO die with graphics.

AMD with its instinct processors has put its memory/network interfaces on the GPU so CPU cores feed off the GPUs. Although the CPUs do have their own local memory, they are basically cut off from the outside world.

Apple has basically done the same with its M line.

nVidia has specific network processors that are basically GPUs. Although I think grace seems to keep the IO on the CPUs. They also have that raid controller GPU thing.

Add to this the new microsoft direct storage API.

Wonder if there will be some relationship in the Zen4?
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
At blender which is MT based. Not ST. Zen 4 MT will be killer due to higher clocks. But the >15% ST uplift on Cinebench R23 compared to 5950X just makes it equal to 12900K.

Rememeber Intel will be doubling e-cores on desktop. AMD has nothing of value in the mid to low end

You do realize > means greater than, right? As in more than 15%. It is not a ceiling, it is a floor.
 
  • Like
Reactions: lightmanek

poke01

Senior member
Mar 8, 2022
651
611
106
You do realize > means greater than, right? As in more than 15%. It is not a ceiling, it is a floor.
That figure is so ambiguous because they tested against a 5950X in r23. There is no reason to state a >15% when you can get a definite % increase in Cinebench r23.

Still no mention of IPC like previous Zen arch reveals.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Do you know this to be true or you just guessing?

Leakers said up to 37% ST uplift for Zen 4 and people hyped Zen 4 so much. Yeah, after this Zen 4 reveal I am not trusting anymore leaks.

It took AMD nearly 24 months to deliver Zen 4 and we shall see later this year if Zen 4 "wrecks Intel".
If I knew for sure then I would be a “real” leaker. Note the big difference between “>15%” and “up to 37%”. These are not mutually exclusive. It wouldn’t surprise me if some applications see a massive boost; some perform exceptionally well with larger L2. You throw AVX512 into the mix, and I wouldn’t be surprised at “up to” a large improvement.

I don’t pay much attention to Intel these days since we have standardized on Epyc at work, so I really don’t have an idea of how much AMD will be “wrecking intel”. Intel has been talking a lot though while AMD has been in super stealth mode still, so I think AMD has the significantly stronger hand.
 
  • Like
Reactions: Tlh97 and majord

RnR_au

Golden Member
Jun 6, 2021
1,675
4,078
106
That figure is so ambiguous because they tested against a 5950X in r23. There is no reason to state a >15% when you can get a definite % increase in Cinebench r23.
A point was made earlier that the slide mentioning >15% uplift was for Zen 4 in general, so has to cover 8C parts which will run slower than 16C parts.
 

tomatosummit

Member
Mar 21, 2019
184
177
116
The 6nm IO die with graphics.

AMD with its instinct processors has put its memory/network interfaces on the GPU so CPU cores feed off the GPUs. Although the CPUs do have their own local memory, they are basically cut off from the outside world.

Apple has basically done the same with its M line.

nVidia has specific network processors that are basically GPUs. Although I think grace seems to keep the IO on the CPUs. They also have that raid controller GPU thing.

Add to this the new microsoft direct storage API.

Wonder if there will be some relationship in the Zen4?
Consumer cpus have been very reluctant to use on cpu network devices. They've been present on all zen1 through 3 cpus and it's hard to find one in use. I'd be less suprised if they did away with the network interfaces going forwards on all but dies with embedded applications.

Although I did notice there are now 4 display outs on the am5 platform that I thought was a nice improvement.
Might be nice to get two of the same display port instead of having to use an adaptor for a second screen,
 
  • Like
Reactions: Tlh97 and majord

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
At 65W it should perform like a 12900K@180W and a 5950X@125W but there s still people who would find this underwhelming.
Lots of folks who do intensive work with their PCs really need the constant uplift in performance in order that their work PC of today is no slower than the work PC of last year when doing the same work at the time/era of its launch, because most of the intensive work grows in complexity every year.

Many dev work for example, like AOSP compilation/dev grew tremendously in code base and complexity in the last two years.
My 5950X starts feeling slow in this case and I find myself waiting far too long more often than not.
High thread count CPUs help but not always. For example, for AOSP compilation/dev, before the soong build system could use all available threads it has to run an intensive ST computation, then comes the compile with all the threads, and then linking, image generation etc. are ST bound and finally emulation is ST bound.
Having a high thread count HEDT helps, but there are these ST compute phases which really gets hindered by the low ST perf of high core count CPUs, because once the parallel work is done you need high ST perf for the next part of the workload.
Therefore something like 5950X is preferable for me. Lots of ST performance but decent MT performance as well.
Ideal CPU would be a 32+ Core CPU with ST perf as much as an 8/16 Core.

Most folks on desktop do not care much about TDP (if it means more performance) if their CPU operates within reasonable range as long as thermals and acoustics are not disturbing the ambience in the work area. Growth in complexity and size of the same workload needs a sizable increase from the HW otherwise with every passing year you would need to wait longer and longer for your work to get done.

For me personally, 12th gen Intel is not something I would be comfortable with, not sure what happens if my docker containers or qemu instances land on a small E core.
Otherwise 16C P core only 12th Gen Intel would be insta buy if the TDP is reasonable enough.
Therefore trying to stick with AMD but seeing the bar for ST perf increase being set so low, I am getting bothered with having to wait more years to get any considerable speedup in my work/hobby.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
In the AMD footnotes, the Zen 4 chip was 46% faster than the 12900k.

People are losing their minds over nothing.

EDIT: We will see what the results are in other workloads. However, Raptor Lake likely won't be 40+% faster unless Intel raises power limits. However, 46% faster while consuming at least 25-35% less power in blender? that seems like a huge win to me. Oh, and they are officially beating Intel at the frequency game as well. 5.52ghz while gaming, that is higher than the 12900ks gets unless you overclock the daylights out of it. Not trying to shill for AMD, however, I see a lot of folks who were disappointed at the limited data shown. I'm just baffled.
I suspect a bunch of marketing people and intel fanboys are “disappointed”. The only thing that disappoints me is how little information was actually released and no teasers on things coming later.

Anyway, if there are real enthusiast that are disappointed, then they must be very young. The kind of generation on generation increases we are getting with AMD’s come back are exceptional compared to the near complete stagnation we had under the intel monopoly. Intel could have made 8 cores mainstream several generations before they were forced to, but that would have cut into Xeon and extreme edition profits, so we had 4 cores forever.
 
  • Like
Reactions: Tlh97 and MangoX

Saylick

Diamond Member
Sep 10, 2012
3,084
6,184
136
That figure is so ambiguous because they tested against a 5950X in r23. There is no reason to state a >15% when you can get a definite % increase in Cinebench r23.

Still no mention of IPC like previous Zen arch reveals.
Just because they can doesn't mean they have to. There's a ton of stuff AMD can tell us, but they choose not to. Better to see ">15% ST" then to see "Up to 15% ST".
 
  • Like
Reactions: Tlh97

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Saw quite reasonable sounding speculation that Zen 4c is just Zen 4 with SRAM moved to stack caches. Thus why 50% more cores are possible in what we might reasonably assume is the same socket. If that's the case for servers that could be the case for desktop as well. Could still see an HEDT 24 core CPU launch second half of next year or so.
I was hoping for some use of silicon bridges. If it is 16 cores on one die, then that is a little problematic since your cache might have to be stacked over some of the cores. It might be that the process tech used for the cpu die is low enough power that it isn’t an issue though.
 

majord

Senior member
Jul 26, 2015
433
523
136
That figure is so ambiguous because they tested against a 5950X in r23. There is no reason to state a >15% when you can get a definite % increase in Cinebench r23.

Still no mention of IPC like previous Zen arch reveals.

Because they don't want to give out exact numbers , and even if they did ,they likely don't have top sku clocks 100% locked in yet.

I dont recall getting even this much performance info for Zen 3 , this early on. Was very little mentioned until the full announcment in october afaik.
 
  • Like
Reactions: Tlh97 and Thibsie

poke01

Senior member
Mar 8, 2022
651
611
106
. Intel could have made 8 cores mainstream several generations before they were forced to, but that would have cut into Xeon and extreme edition profits, so we had 4 cores forever.
That was due to Intel being a market leader and having awful management. The Intel of today is different from 2015-2020 Intel.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Lots of folks who do intensive work with their PCs really need the constant uplift in performance in order that their work PC of today is no slower than the work PC of last year when doing the same work at the time/era of its launch, because most of the intensive work grows in complexity every year.

Many dev work for example, like AOSP compilation/dev grew tremendously in code base and complexity in the last two years.
My 5950X starts feeling slow in this case and I find myself waiting far too long more often than not.
High thread count CPUs help but not always. For example, for AOSP compilation/dev, before the soong build system could use all available threads it has to run an intensive ST computation, then comes the compile with all the threads, and then linking, image generation etc. are ST bound and finally emulation is ST bound.
Having a high thread count HEDT helps, but there are these ST compute phases which really gets hindered by the low ST perf of high core count CPUs, because once the parallel work is done you need high ST perf for the next part of the workload.
Therefore something like 5950X is preferable for me. Lots of ST performance but decent MT performance as well.
Ideal CPU would be a 32+ Core CPU with ST perf as much as an 8/16 Core.

Most folks on desktop do not care much about TDP (if it means more performance) if their CPU operates within reasonable range as long as thermals and acoustics are not disturbing the ambience in the work area. Growth in complexity and size of the same workload needs a sizable increase from the HW otherwise with every passing year you would need to wait longer and longer for your work to get done.

For me personally, 12th gen Intel is not something I would be comfortable with, not sure what happens if my docker containers or qemu instances land on a small E core.
Otherwise 16C P core only 12th Gen Intel would be insta buy if the TDP is reasonable enough.
Therefore trying to stick with AMD but seeing the bar for ST perf increase being set so low, I am getting bothered with having to wait more years to get any considerable speedup in my work/hobby.
I hope AMD has an in between socket for Threadripper and workstation parts with Zen 4. Using a giant Epyc IO die and a giant SP5 sized socket seems very wasteful.
 
  • Like
Reactions: lightmanek

Saylick

Diamond Member
Sep 10, 2012
3,084
6,184
136
I have to wonder, if Zen 4 is really just Zen 3 with 2x L2, 1.5x L2 BTB, and AVX512 on TSMC N5, then it really does seem like the vast majority of engineering resources was allocated towards developing the new platforms. Remember, it's not just AM5 that needed to be designed, they have to design their new server socket as well, SP5.

All of this reminds me of AT's Zen 2 architecture article:
An interesting element to Zen 2 is around the intention. Initially Zen 2 was merely going to be a die shrink of Zen+, going from 12nm down to 7nm, similar to what we used to see with Intel in its tick-tock model for the initial part of the century. However, based on internal analysis and the time frame for 7nm, it was decided that Zen 2 would be used as a platform for better performance, taking advantage of 7nm in multiple ways rather than just redesigning the same layout on a new process node. As a result of the adjustments, AMD is promoting a +15% IPC improvement for Zen 2 over Zen+.
Zen 2 was originally just going to be a die shrink of Zen+ so it likely just meant performance gains came purely from higher clocks. However, as we now know, the IPC increase came from pulling design features that originally were going to be introduced in Zen 3. If the original plan happened, Zen 3 would have likely seen over 30% IPC increase in one generation. With Zen 2, AM4 was already out and any engineering effort spent on the platform was spent figuring out how to route the traces to get chiplets to work. That required a lot of work, but far less than designing a new platform from scratch.

Zen 4 feels like a similar predicament if it just taps N5 for frequency gains and minor adjustments here and there, with the bulk of the work on the platform. I know that AMD is a larger company today than it was 5 years ago, but they have a lot more products to focus on as well. If history does rhyme, then Zen 5 would be analogous to the original trajectory of Zen 3 where it is the tock that brings big IPC uplifts now that the platform work is out of the way.

Anyways, just some musing as I try to make sense of all this.