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

Page 38 - 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

Hitman928

Diamond Member
Apr 15, 2012
5,180
7,630
136
Who are the customers? And what is the volume?

They have many customers. Most that I am aware of are using the process for analog/mixed signal/RF designs. They also have customers in the automotive space or those who can benefit from using IP that are qualified for automotive environments. For the volume, you'd have to check their financial reports to see if they break it down at all between processes.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
Who are the customers? And what is the volume?
Well since the volume capacity for 22FDX is in the 100K wspm range. While the volume capacity for 14LPP/12LP/12LP+ combined is <55K wspm. I would say there are more customers for 22FDX and the volume is at least double that of GloFo's FinFETs.
You need to mention Intel 7nm too.
Intel's 7nm has been here. Most of it's been taped out. The GPU, the Atom Forests, and the that's not a moon Cove core, etc. It's really has just been politics at Intel.

- 14nm backports
- 10nm ongoing but also cancelled (Some of these products are actually 7nm backports; if it aint broke on 7nm, it will be broke on 10nm)
- TSMC 7nm parallel ports (X-Gold 806&816 was TSMC 7nm, but also was Intel 14nm for the XMM 8060 and Intel 7nm for the XMM 8160)
- Internal 7nm NPI (Of which, there is a ton of tape-ins(ARM(CPU/GPU/PHY/etc)) from external sources and tape-outs spawned from that)

There is no depeche mode in 7nm.
---
12FDX so far has been built up for high performance based on: https://www.leti-cea.com/cea-tech/leti/english/Pages/Leti/Projects supported/OCEAN12.aspx

With Rockchip's 22nm: RK3568/RK3566/RK1808
It is very likely that the next process nodes for the succeeding SKUs will potentially be 12FDX. Which unlike 22FDX will have all proposed enhancements at introduction(1.0 PDK) rather than at a + node or a 1.0+ PDK.

---
For "Zen4" it is best AMD abandons GlobalFoundries. Use TSMC's 6nm for the IOD and the chipset instead. 125 mm2 is too big and they need a foundry with proven Analog/Mixed-Signal shrinks on FinFETs like TSMC. N12 at GloFo -> N6 TSMC -> N4 TSMC is the ideal cost-effective path for the IOD given the need of PCIe 5.0 -> PCIe 6.0 & DDR5 -> DDR6 fast-bring-ups.
 
Last edited:

Cardyak

Member
Sep 12, 2018
72
159
106
...and the that's not a moon Cove core, etc. It's really has just been politics at Intel.

Obvious reference to the Star Wars Death Star here with the "that's not a moon" quote. Which is interesting because many months ago you alluded to a Cove architecture Intel was working on code-named "Deadsun" (Literally a euphemism for Death Star)

Where I found it didn't specify what Cove core it was talking about on the roadmap. Other than it was later rather than earlier, since we only know about Goldencove, I originally had Goldencove in the spot.
Palmcove (10nm) -> Sunnycove (10nm+) -> Willowcove(10nm++) -> Goldencove (10nm+++) -> --- Cove (7nm) -> --- Cove (7nm+) --> "Deadsun 1" (7nm++) -> "Deadsun 2" (5nm)
 

amrnuke

Golden Member
Apr 24, 2019
1,181
1,772
136
I don't think zen4 is expected by end of the year, somewhere mid-2022 is more likely. So, no it will not be beaten by zen4 by the end of the year.
Given November Zen 3 launch, and a 15-18 month cadence, a first-half launch seems likely (Feb-May 2022). Is there any reason to think the cadence has dropped to more than 18 months even as Intel are moving their cadence to a faster pace?
 

LP-ZX100C

Junior Member
Mar 16, 2021
10
17
41
Given November Zen 3 launch, and a 15-18 month cadence, a first-half launch seems likely (Feb-May 2022). Is there any reason to think the cadence has dropped to more than 18 months even as Intel are moving their cadence to a faster pace?

Well ,AMD will start ramping up Zen4 (5nm) starting with 2H2021, also ,i think , they'll have products ready for hyperscalers and cloud at the end of the year , and the said launch will happen later.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Just to make sure there's no confusion, 22FDX has been mature and shipping high volume for a while now. 12FDX is still in the oven. I'm guessing it gets launched next year or possibly the year after.
It's almost as ridiculous as a vaporware as cannon lake.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Well ,AMD will start ramping up Zen4 (5nm) starting with 2H2021, also ,i think , they'll have products ready for hyperscalers and cloud at the end of the year , and the said launch will happen later.

Considering how long they sat on Milan, there's no telling when they'll actually launch Genoa. Rome is still faster than Intel's fastest Ice Lake-SP so they're not under a lot of pressure right now.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
But only in Frontier ??
We'll have to wait and see
Although Trento is being designed for Cray, it will also be available to other OEMs/ODMs. The main advantage of this platform is scalability, bandwidth, latency, and of course, the relatively simpler programmability thanks to the use of the Infinity Fabric 3.0 interconnect and a unified memory pool across the CPU and GPU
.
 

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,619
136
Note that the follow on supercomputer El Capitan, which is scheduled to go live early 2023, is announced to use Epyc Genoa chips. So whichever ways Milan and Trento differ Genoa won't need to be adapted to supercomputer use anymore. (And Zen 4 based products should be available by late 2022 at the latest.)

Considering how long they sat on Milan
Considering all the big guys got Milan already AMD effectively just sat on the Milan launch press release.