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

Page 278 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Vattila

Vattila

Senior member
Oct 22, 2004
821
1,457
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
inf64

inf64

Diamond Member
Mar 11, 2011
3,884
4,692
136
Wccftech reported today that AMD will announce the parts on Aug 29 while Press/Sales/Review embargo will lift on Sep 13. It won't be long now.

I've seen the well known slide again (the Ryzen 7000 specification) and it got me thinking:
1659516732847.png

The maximum ST performance could be actually hidden (in plain sight) in this slide - the max boost is 5Ghz(+) and ST uplift is >15%. So these two numbers give us bottom range for ST uplift for the lowest end Ryzen 7000 part (7600X?). If this is true, It's easy to extrapolate how 7950X will perform vs 5950X : 5.7Ghz (assumed max boost) / 5Ghz x 1.15 = 1.31 or 31% faster.

It's interesting to note that if ~15% higher ST is calculated for 7600X part with 5Ghz max boost (like the ES we saw on UB), then the average IPC is a bit higher than what AMD showed us in the reveal : 1.15 / (5/4.9) = 1.127 or ~12-13%. The IPC calculation they used is an average that has Geekbench and R23 numbers along with SPEC. I suspect that GB and R23 are dragging the average by quite a bit (being only ~7-8% better as shown in the Genoa ES leak we had), while SPEC is more representative of the common IPC uplift we will see in majority of desktop workloads (~12-13%). If we used 7% for R23, ~8% for Geekbench and ~13% for SPEC, both the geo-mean and arithmetic mean land at ~9%, smack in the middle of what AMD showed in their slides (8-10%).
 
  • Like
Reactions: Tlh97, Schmide, ftt and 1 other person
E

eek2121

Diamond Member
Aug 2, 2005
3,410
5,049
136
Meanwhile, no Zen 5 until 2024. How will AMD fill the void?

New details on the “Zen 4” core architecture, expected to deliver significant performance and power efficiency improvements over the previous generation.

The “Zen 5” core planned for 2024, which is built from the ground up to extend performance and efficiency leadership across a broad range of workloads.
Click to expand...
 
  • Like
Reactions: lightmanek
D

deasd

Senior member
Dec 31, 2013
603
1,033
136
inf64 said:
View attachment 65286


It's interesting to note that if ~15% higher ST is calculated for 7600X part with 5Ghz max boost (like the ES we saw on UB)
Click to expand...

Sometimes I wanted to say something about UB results/database but since UB's credibility is being questioned I just gave up.

But when I see this old pic(>15%, 5Ghz+)you post, much more doubt is rising in my mind: are we being misleaded by these rumors floating around?

UB's 7600X(ES) has only 4.9Ghz, and there's an 8 core leaked half a year ago that it showed only 5.2Ghz, IIRC. If we looked into UB's database, the benches across different thread/core counts showed the clock are always same or slightly below the ST turbo,

For example the results 12900k's users uploaded, clocks are between 4.7-5.1, which is avg clock

and 5600x between 4.3-4.6

Since UB's admin is crazy and mad against 7600X(ES), I have a feeling that UB's result is legit, but with some errors. Its 2-cores, 4-cores, 8-cores bench result seems to be very closed to the '16 cores faster than 12900k, 31% or 45%' claimed by AMD(I forget that guy's name sorry)

This is my last thought before tomorrow's showcase, I hope it clear some doubts.
 
Last edited:
  • Like
Reactions: Tlh97 and Kaluan
N

NostaSeronx

Diamond Member
Sep 18, 2011
3,811
1,290
136
A Zen4 within Zen5 rumor I found on the internet which requires validation:
The Zen5 IOD will not be re-using the 6nm Zen4 IOD instead it will use a new 4nm IOD. That 4nm IOD has been inserted into Zen5 CPU-S/CPU-H. That IOD is a SoC which has Zen4 cores in it.

With it comes a change in naming of the chiplets, which are changed to Central [System might be used here] Die and [Central might be used here as well] Computational Die(s).

Edit: Forgot to add something here.
AM5 is stuck with symmetrical computational dies. There is no Zen5+Zen4c(or d) configuration. It is just Central Die with energy-efficient(Zmin-orientated) Zen4 cores and up to two Computational Dies with high-performance(Zmax-orientated) Zen5 cores.

Speculation through LNKN/GP/RPDF:
- Zen4 in central die is used majority of the time, most processes in Windows are very low power tasks aka Zmin-orientated workloads.
- Zen5 in computational die is only used as an accelerator rather than previous CCD-style, only used for high power tasks aka Zmax-orientated workloads.
- Given the above, the computational dies use CPU drivers. So, when the dGPU is loaded the computational die is loaded instead of the central die. There is no case where a game will not run on the Zen5 cores. The CCD's are not to be flooded with background tasks w/ ~110ns DRAM<->CCD, bg tasks instead will use the Central die w/ sub~64ns DRAM<->CCX.
 
Last edited:
  • Like
Reactions: RnR_au, Tlh97 and Kaluan
E

eek2121

Diamond Member
Aug 2, 2005
3,410
5,049
136
Timorous said:
Zen 4c refresh with multi layer v-cache?
Click to expand...

AMD has stated that Zen4c is EPYC only IIRC.

igor_kavinski said:
During the "void", Intel also doesn't have anything of note. AMD could always do Zen 4 XT SKUs.
Click to expand...

Intel has 2-3 releases before/into 2024. They are iterating in a 10-12 month cadence. Rocket lake will be close to Zen 4 chips in terms of performance and anything after will beat them. 3D V-Cache will help workloads such as gaming, but Intel is going to be bringing further multicore gains.

Note that Intel is apparently moving up the schedule for RKL mobile chips.
 
I

igor_kavinski

Lifer
Jul 27, 2020
28,005
19,125
146
eek2121 said:
Intel has 2-3 releases before/into 2024. They are iterating in a 10-12 month cadence.
Click to expand...
Nothing of note in the sense that I don't expect Intel to release anything groundbreaking that would make Zen 4 look as old as Zen 3 looks now and despite being old, Zen 3 is still pretty competitive. I would also be surprised if RPL mobile is able to deliver better battery life than AMD 6000 series CPUs.
 
  • Like
Reactions: lobz, lightmanek, Tlh97 and 1 other person
E

eek2121

Diamond Member
Aug 2, 2005
3,410
5,049
136
Karnak said:
There is only MTL in H2 2023 besides RTL. After that ARL in H2 2024. Don't see a problem with that. AMDs "new gen" is V-Cache. Helps a lot with gaming performance and efficiency.

And MTL looks like mobile focused so far.
Click to expand...

Meteor Lake is most certainly not mobile focused. It is coming to the desktop and will have double digit multicore gains thanks to a new *mont core. It will also be on a new node which means we can expect much better perf/watt.
 
  • Like
Reactions: Tlh97 and Saylick
K

Kaluan

Senior member
Jan 4, 2022
515
1,092
106
igor_kavinski said:
IF Intel 4 doesn't have any issues...
Click to expand...
There's that rumor that MTL might see a frequency regression, but I don't know where it came from or how believable it is but IT DOES REMIND me of back when Intel's first gen "10nm" (mobile) parts "launched". A disaster.

Has anyone seen this BTW? (source is WCCFTech tho)
videocardz.com

AMD to announce Ryzen 7000 on August 29th, launching September 15th - VideoCardz.com

AMD Ryzen 7000 series to be unveiled this month According to Wccftech, AMD has set the date for hard launch of the new desktop Zen4 CPUs. This is a follow-up to official confirmation to AMD releasing Ryzen 7000 series this quarter. AMD CEO Lisa Su narrowed down the release date during the Q&A...
videocardz.com videocardz.com
 
  • Like
Reactions: Tlh97 and igor_kavinski
D

Doug S

Diamond Member
Feb 8, 2020
3,574
6,311
136
Hans Gruber said:
I think people should consider the computer market is falling apart in this global recession.
Click to expand...


It is just reverting to pre pandemic levels. Other than schools/parents buying Chromebooks/PCs for remote learning by students who previously had none, the pandemic PC bubble pushed purchases forward - i.e. corporations were buying employees brand new laptops regardless of where in the replacement recycle the desktops abandoned in the office were.

So I wouldn't attribute falling to 2016-2019 levels as "falling apart". Not even if it fell below those levels by 10% or so for a year or two, which I think is a strong possibility due to all the corporate purchases pushed forward that won't begin to hit the replacement cycle until at least 2024.

This is to be expected when the fundamentals of the PC market that caused it to slowly shrink off its peak in 2010 or so have not changed. Anyone who thought the recent strength in the PC market represented a new normal was always going to be disappointed.
 
  • Like
Reactions: MangoX, Tlh97, moinmoin and 2 others
biostud

biostud

Lifer
Feb 27, 2003
19,909
7,014
136
eek2121 said:
AMD has stated that Zen4c is EPYC only IIRC.



Intel has 2-3 releases before/into 2024. They are iterating in a 10-12 month cadence. Rocket lake will be close to Zen 4 chips in terms of performance and anything after will beat them. 3D V-Cache will help workloads such as gaming, but Intel is going to be bringing further multicore gains.

Note that Intel is apparently moving up the schedule for RKL mobile chips.
Click to expand...
But how large a percentage of PC users are upgrading their PC for MT performance?
 
Markfw

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,241
16,107
136
Vope45 said:
According to wccf's Hassan motherboard maker source, Zen 4 is still behind AL and RL in cpu sensitive games: https://wccftech.com/amd-ryzen-7000-desktop-cpus-x670-motherboards-launch-on-15th-september/


Look at the comments section: RL > Zen 4 in both latency and bw running the same DDR5 JEDEC speeds . 3 games were tested SOTR, FH5 and FFXIV plus FSU.
Click to expand...
The comments section ? thats the WORST place to find decent believable leaks. I will wait for reviews.

Edit: and the more comments I read, the more BS I see.
 
  • Like
Reactions: Drazick and Kaluan
V

Vope45

Member
Oct 4, 2020
114
168
86
Markfw said:
The comments section ? thats the WORST place to find decent believable leaks. I will wait for reviews.

Edit: and the more comments I read, the more BS I see.
Click to expand...
Hassan said he'd reserve a separate article for now. What's bs about his infos btw?
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom