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

Page 27 - 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

french toast

Senior member
Feb 22, 2017
988
825
136
Very disappointing about Warhol IMO, was expecting some mild IPC improvements ~8% and +200mhz clocks, plus new IOD with AM5, seems we are getting another ZEN+.

This may have been taken out of AMD's hands somewhat with struggles to get EUV machines and capacity wars at TSMC.
At the end of the day we all know that having the best chip and technology doesn't guarantee you improve market share and margin, rather capacity and costs play a huge role in that, it seems AMD looked at what's guaranteed to be available at that time frame and figured they would save development costs and get more of the less leading edge wafers at a cheaper price, rather than much less of the more expensive wafers at the leading edge.
Shame on a technical point of view but understandable on the business side of things as AMD already has brilliant chips, - they just can't make enough of them to take market share.

I do feel this is an opportunity missed to crush Intel on the technical side, if Alderlake S comes around in volume H2 2021 then AMD might become the budget option again.
Disappointing but understandable is my take.
 

french toast

Senior member
Feb 22, 2017
988
825
136
Well they got quite backfired for paperlaunching both Ryzen 5xxx and RDNA2 so they might not want to stretch it too much.
AMD need to get another supplier going at Samsung, I know they have global foundries for the IOD, but if the want to improve output and market share they need Samsung to produce a portion.

Samsung's process is ideal for low power applications, that is what they have earned their bread and Butter on from smartphones.
By all intents and purposes AMD has a close relationship with their chip division, they should make the laptop chips and future IOD at Samsung to free up capacity IMO.
 
  • Like
Reactions: Tlh97 and Thibsie

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Very disappointing about Warhol IMO, was expecting some mild IPC improvements ~8% and +200mhz clocks, plus new IOD with AM5, seems we are getting another ZEN+.

Can't say I blame you, but honestly what do you expect? AMD is going to milk their advantage. Also, contrary to my previous assumptions, it may well be that it took AMD longer to bring Zen3 to market than Zen2 or Zen+ (relatively speaking), so they may want to milk it for awhile.

I do feel this is an opportunity missed to crush Intel on the technical side, if Alderlake S comes around in volume H2 2021 then AMD might become the budget option again.
Disappointing but understandable is my take.

Agreed, though my hope was for Zen4 on AM5 in 2021, which is definitely not going to happen.
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
if Alderlake S comes around in volume H2 2021 then AMD might become the budget option again.
Do you seriously expect that? With Intel lacking at the execution side for like half a decade now (which over time "helped" it lose all of the advantages it initially had over AMD) I find it hard to imagine that Alder Lake alone will not only make Intel competitive again across the product range but also force down AMD to become a budget option again.
 

exquisitechar

Senior member
Apr 18, 2017
655
862
136
Do you seriously expect that? With Intel lacking at the execution side for like half a decade now (which over time "helped" it lose all of the advantages it initially had over AMD) I find it hard to imagine that Alder Lake alone will not only make Intel competitive again across the product range but also force down AMD to become a budget option again.
Golden Cove is likely something like Sunny Cove in terms of the IPC improvement. The only way it won't make AMD the budget option for a while (bar MT performance when it comes to the 5950x) is if it's pushed back to next year, closer to Raphael than Warhol.
 
May 17, 2020
122
233
86
OK, in that case It's most probably CDNA + HBM + Zen3 on package (considering what they hinted at the announcement and the money that was thrown AMD's way to make software for it)
In frontier Trento CPU will be combined with GPU MI200 (GPU build on MCM/chiplet and (based on CDNA2 ??)), so what is purpose to put CDNA(2) GCD (Graphic Core Die) chiplet on EPYC package ?
 

inf64

Diamond Member
Mar 11, 2011
3,685
3,957
136
Golden Cove is likely something like Sunny Cove in terms of the IPC improvement. The only way it won't make AMD the budget option for a while (bar MT performance when it comes to the 5950x) is if it's pushed back to next year, closer to Raphael than Warhol.
So Zen3 which has around 7% higher IPC than WC does not make WC a budget option, but a *possible* IPC lead of around 10% GC might have over Zen3 will make AMD a budget option? That is ridiculous.

Not to mention AMD is a moving target and Zen4 is coming in early 2022, so whatever IPC lead they may have will evaporate quickly. Then there is clock speed, core count, platform features, longevity etc. Intel is in a world of hurt and will be a slower(budget) option until 2024.
 

teejee

Senior member
Jul 4, 2013
361
199
116
AMD need to get another supplier going at Samsung, I know they have global foundries for the IOD, but if the want to improve output and market share they need Samsung to produce a portion.

Samsung's process is ideal for low power applications, that is what they have earned their bread and Butter on from smartphones.
By all intents and purposes AMD has a close relationship with their chip division, they should make the laptop chips and future IOD at Samsung to free up capacity IMO.


All of their budget CPU's like 3200G, 3400G, 2600, 240GE, 200GE , 3000G and more are from Global Foundry, so they are using GF pretty well.
 

dr1337

Senior member
May 25, 2020
309
503
106
In frontier Trento CPU will be combined with GPU MI200 (GPU build on MCM/chiplet and (based on CDNA2 ??)), so what is purpose to put CDNA(2) GCD (Graphic Core Die) chiplet on EPYC package ?
Heterogeneous compute. Accelerators already work in conjunction with CPUs, its just through (relatively) slow and latent PCIE interconnects. By moving all main components onto the same package you significantly increase communication speed, and you can reap other benefits such as shared memory and shared caches. Basically, the accelerator can load data and work faster by being on the same package. Having memory on package is part of this too. You can significantly increase bandwidth and/or significantly improve latency for both major components at the same time.
 
  • Like
Reactions: moinmoin

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Oh and as an aside, it does seem like Warhol is a straight Vermeer refresh akin to the Zen 2 XT skus.

I was hoping it would be AM5 but unfortunately not, seems they're focusing on clocks. Wondering what is possible though given it is possible - even if rare - to get Zen 3 chips clocking above 5GHz already.

Well, IIRC, there are two other performance N7 process nodes still on the table at TSMC (Like EUV). But, if all we get is a +200MHz bump in clocks, Warhol would be pretty lame.
 

french toast

Senior member
Feb 22, 2017
988
825
136
Can't say I blame you, but honestly what do you expect? AMD is going to milk their advantage. Also, contrary to my previous assumptions, it may well be that it took AMD longer to bring Zen3 to market than Zen2 or Zen+ (relatively speaking), so they may want to milk it for awhile.




Agreed, though my hope was for Zen4 on AM5 in 2021, which is definitely not going to happen.
Yes it looks like we all expected too much, remember all that 5nm Zen 3 fluff that was doing the rounds on here a few months ago?
I thought I was one of the more realistic ones.
I was in the Warhol being zen 3 refresh on AM5.. Also nope.
Do you seriously expect that? With Intel lacking at the execution side for like half a decade now (which over time "helped" it lose all of the advantages it initially had over AMD) I find it hard to imagine that Alder Lake alone will not only make Intel competitive again across the product range but also force down AMD to become a budget option again.
Whether we like it or not, gaming seems to dictate which is the premium products on the desktop, why else was AMD the budget option with zen 1 and Zen + when it had more cores and all the other advantages? Single core performance and gaming.
If Alderlake S comes H2 2021, it will easily have the ST and gaming crown, it may also have efficiency crown in some scenarios also, never mind new brand new platform and all the marketing opportunities that creates.
So Zen3 which has around 7% higher IPC than WC does not make WC a budget option, but a *possible* IPC lead of around 10% GC might have over Zen3 will make AMD a budget option? That is ridiculous.

Not to mention AMD is a moving target and Zen4 is coming in early 2022, so whatever IPC lead they may have will evaporate quickly. Then there is clock speed, core count, platform features, longevity etc. Intel is in a world of hurt and will be a slower(budget) option until 2024.
Do we know zen 4 desktop is coming that early? I would argue we won't see 5nm on desktop till H2 2022, one year later than Alderlake S.
Willow Cove might actually beat zen 3 on St and gaming performance, which makes them a premium product in the eyes of the public.
So next year Alderlake S would be the premium product no doubt, which makes anything less from 16 cores from AMD the budget option.. Comparatively speaking.

AMD needs AM5 in 2021 and a solid increase in Cpu performance to keep them up there...i thought Zen 3 + on 6nm was realistic..
All of their budget CPU's like 3200G, 3400G, 2600, 240GE, 200GE , 3000G and more are from Global Foundry, so they are using GF pretty well.
True enough, I still think they need to outsource more from Tsmc, not enough capacity there at all for all those products.
Samsung 5nm Lpp in 2022 would be great for laptop and budget GPUs.
 
Last edited:

exquisitechar

Senior member
Apr 18, 2017
655
862
136
Do we know zen 4 desktop is coming that early? I would argue we won't see 5nm on desktop till H2 2022, one year later than Alderlake S.
Willow Cove might actually beat zen 3 on St and gaming performance, which makes them a premium product in the eyes of the public.
So next year Alderlake S would be the premium product no doubt, which makes anything less from 16 cores from AMD the budget option.. Comparatively speaking.
Hopefully, we'll see Zen 4 desktop by the end of Q2 2022 at least. Willow Cove and Zen 3 are pretty similar in terms of ST performance.
So Zen3 which has around 7% higher IPC than WC does not make WC a budget option, but a *possible* IPC lead of around 10% GC might have over Zen3 will make AMD a budget option? That is ridiculous.
Rocket Lake appears to have pretty similar ST performance to Vermeer. Add 15-20% and Zen 3 desktop CPUs will be firmly the budget option compared to Alder Lake-S if they are to compete.
 
  • Like
Reactions: french toast

inf64

Diamond Member
Mar 11, 2011
3,685
3,957
136
Hopefully, we'll see Zen 4 desktop by the end of Q2 2022 at least. Willow Cove and Zen 3 are pretty similar in terms of ST performance.

Rocket Lake appears to have pretty similar ST performance to Vermeer. Add 15-20% and Zen 3 desktop CPUs will be firmly the budget option compared to Alder Lake-S if they are to compete.
Rocketlake is still vaporware with no official benchmarks. We had a few leaks here and there but nothing substantial. Like I said, Zen3 has a nice IPC lead over the best Cove there is, so a backported IceLake on 14nm can only clock higher and that is about it. It will have 2x less L3 cache per core than 5800X which is a huge hindrance in gaming.
 

yuri69

Senior member
Jul 16, 2013
373
573
136
Rocketlake is still vaporware with no official benchmarks. We had a few leaks here and there but nothing substantial. Like I said, Zen3 has a nice IPC lead over the best Cove there is, so a backported IceLake on 14nm can only clock higher and that is about it. It will have 2x less L3 cache per core than 5800X which is a huge hindrance in gaming.
*Mobile* WC is about 5% slower than *desktop* Zen 3. Mobile Zen 3 SKUs won't feature that large L3 - WC will be closer.

RL should feature a bit lower IPC compared to Zen 3. OTOH it'll clock beyond Zen 3 - at least those ES do. This combination will target stealing Zen 3 gaming crown. It will probably succeed given its lower latencies in the memory subsystem. That's the point Intel has with the CPU.

There is nothing known about GC besides aiming at "IPC" gains (SC => around 1.2x). This thing being at a newer 10nm revision than Tiger Lake promises some very nice frequency targets. Apparently, Intel targets both 2021 servers and desktop with GC.

It seems nice like a competitive offering. I mean having a modern DDR5, PCIe5 platform equipped with beyond Zen 3 IPC in 2021/2022. Right?
 
  • Like
Reactions: french toast

HurleyBird

Platinum Member
Apr 22, 2003
2,670
1,250
136
Whether we like it or not, gaming seems to dictate which is the premium products on the desktop, why else was AMD the budget option with zen 1 and Zen + when it had more cores and all the other advantages?

Probably because of comparison equivocation with Skylake-X and a very low reputation post bulldozer. And where AMD did lose, it was often more than 7%.

Zen 2 still lost in gaming and absolute ST performance but wasn't considered a budget option at all.

If Alderlake S comes H2 2021, it will easily have the ST and gaming crown, it may also have efficiency crown in some scenarios also, never mind new brand new platform and all the marketing opportunities that creates.

Given Intel's execution, I'd be surprised if it came out any time before late Q4 2021, and I'd peg 2022 as more realistic. Shades of Rocket Lake in late 2019 and then 2020. I'd very much like to be wrong, however.

Do we know zen 4 desktop is coming that early? I would argue we won't see 5nm on desktop till H2 2022, one year later than Alderlake S.

I'd posit that to the extent Alderlake-S is impressive and the extent to which it launches earlier than expected, will help drive the Zen 4 release schedule, and while I won't discount entirely the idea of Zen 4 releasing a full year later, I find that scenario highly dubious except if Alderlake-S seriously underperforms expectations or AMD experiences significant issues with bringup.
 
Last edited:

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
Whether we like it or not, gaming seems to dictate which is the premium products on the desktop, why else was AMD the budget option with zen 1 and Zen + when it had more cores and all the other advantages? Single core performance and gaming.
If Alderlake S comes H2 2021, it will easily have the ST and gaming crown, it may also have efficiency crown in some scenarios also, never mind new brand new platform and all the marketing opportunities that creates.
Just a reminder that Intel still hasn't managed to release any node newer than 14nm on desktop (and even on server that has been repeatedly delayed, maybe we'll see Ice Lake-SP soon).
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
I don't uderstand what is the codename Trento for ? An EPYC between Rome and Genoa ?
i wonder what'll be the main difference, DDR5?
Possibly, or CCD cache amounts and sheer IOD capacity (ie more PCIe lanes) I would think.

Maybe also extra IO PHY's beyond what they plan to give Milan like CXL, CCIX, GenZ etc etc - perhaps even baked in state of the art ethernet like 100Gb/s?
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Just a reminder that Intel still hasn't managed to release any node newer than 14nm on desktop (and even on server that has been repeatedly delayed, maybe we'll see Ice Lake-SP soon).

. . . which is probably one reason why AMD isn't launching Raphael until 2022, more-likely-than-not. That and maybe it really is taking that long to pull it together.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,670
1,250
136
*Mobile* WC is about 5% slower than *desktop* Zen 3. Mobile Zen 3 SKUs won't feature that large L3 - WC will be closer.

But it does feature a significant better memory controller, and while Renoir had 1/4 of Matisse's L3, Cezanne ups that to 1/2 of Vermeer's.

Most likely mobile Zen 3 will performance will be roughly comparable to desktop. A tad slower overall, but moderately exceeding its desktop counterpart in gaming workloads.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
It will have 2x less L3 cache per core than 5800X which is a huge hindrance in gaming.
Huge hindrance is a real stretch.

Anyone truly serious about gaming to the point that this would bother them is not doing so on a mobile APU chip/platform.