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

Page 414 - 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

Kaluan

Senior member
Jan 4, 2022
500
1,071
96
Also a single CCD with 3D V Cache is better for gaming
View attachment 69205

View attachment 69206

As you can see... 1 CCD is better than 2 CCD for gaming(we all knew that since the 3950X days), so a 7800X3D will just be enough to counter any 13900KS(which has 8P cores and don't use the e cores for gaming)
Sample size: 1 game. :grin:

Dude, you're kidding, right?

This one has a much better sample size (and transparency in testing)... and it says the opposite.

Jeez, that grifter didn't even mention what res that was at... 01-FPS.png
Just peeked at a few reviews that tested 7700X and 7950X with Metro Exodus and the game consistently preffers the single-CCD 7700X. Who would've thought there are some games that do that? Clearly taking a sample size of 1 and making definitive claims is a good idea. :p

Edit: What I'd be more interested in is 16C/16T 7950X tests, gaming or otherwise.
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
Sample size: 1 game. :grin:

Dude, you're kidding, right?

Using a sample size 1 to enforce a "known Fact" it's a valid point. Are you going to be the guy that will contradict the point? That One CCD is better at gaming than Two CCDs? Just to be clear, this are equal binned CCDs(From the same CPU) this is not a stock 7700X vs 7950X(it could if they were at ISO Speed, say 5 Ghz All Core)
 
  • Haha
Reactions: Atari2600

biostud

Lifer
Feb 27, 2003
18,194
4,675
136
Also a single CCD with 3D V Cache is better for gaming
View attachment 69205

View attachment 69206

As you can see... 1 CCD is better than 2 CCD for gaming(we all knew that since the 3950X days), so a 7800X3D will just be enough to counter any 13900KS(which has 8P cores and don't use the e cores for gaming)
Can't that be handled by software to keep it running on only one of the CCDs?

Also does it prefer ADL with or without e-cores?
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
I wonder if it's due to a scheduling mess WIndows is making or due to latency between CCDs?

Its both, here he tested 2 CCD with SMT Disabled. But still a single CCD is faster

FfH53PWXoAAV6n0.jpeg

Can't that be handled by software to keep it running on only one of the CCDs?

Also does it prefer ADL with or without e-cores?
Project Lazo for gaming perhaps. Since hardly any games like more than 8 Cores And the 13900K has 8P core, games more likely than not will only use 8P Cores for gaming(the 16C/16T e cores will likely be doing background tasks which will help on system stability of course.)
 

biostud

Lifer
Feb 27, 2003
18,194
4,675
136
Using a sample size 1 to enforce a "known Fact" it's a valid point. Are you going to be the guy that will contradict the point? That One CCD is better at gaming than Two CCDs? Just to be clear, this are equal binned CCDs(From the same CPU) this is not a stock 7700X vs 7950X(it could if they were at ISO Speed, say 5 Ghz All Core)
You could argue, if you never play Metro, then the 7700x and 7950x are pretty equal.

So who is to blame? If 99.9% of games run equally good on both processors or the one game, that runs worse with two CCDs present. Maybe they should make some changes in their game engine, so it wasn't hit with a performance penalty.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,620
136
Using a sample size 1 to enforce a "known Fact" it's a valid point.
For the specific games tested, yes maybe. For generalization it's little more than anecdotes until you get decent coverage.

That SMT and cross CCX communication can have negative performance impact on some games is nothing new and are talked about since each of those exist. Since developers of the games affected don't appear to care, maybe somebody should write a process lasso like app specific for gamers that automatically applies game specific core/thread mappings to fix those.
 

Kaluan

Senior member
Jan 4, 2022
500
1,071
96
Using a sample size 1 to enforce a "known Fact" it's a valid point.
What's stopping me from using a sample size of 1 to enforce the opposite of "known facts"? lol

Some games literally only run slightly faster on just one CCD from a 7950X just because that CCD can clock higher than the other one/combined, at least as long as that title can only use < 8c/16t.
So it's not just a question of latency and scheudling, at least not always.

Edit: 1MB of L2/32MB of L3 only having to be shared between 1/8 threads (SMT off) rather than 2/16 (SMT on) is also a factor I bet.

Edit2: Anyone know of any test done with 5800X3D done with SMT off?
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
What's stopping me from using a sample size of 1 to enforce the opposite of "known facts"? lol

Some games literally only run slightly faster on just one CCD from a 7950X just because that CCD can clock higher than the other one/combined, at least as long as that title can only use < 8c/16t.
So it's not a question of latency, at least not always.
AMD 5900X3D Prototype with 192 Total L3 showed a 15% average gaming performance across the same games that the 5800X3D(with half of the total cache).
1665861022222.png

Are we really arguing this?
 
Last edited:

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
That says 5900X, not 5800X3D...

5900X3D 192L3$ Prototype apples to apples vs 5900X Shows a 15% average gaming uplift.

AMD when showing up the 5800X3D they also did a 5900X apples to apples comparison with a 15% average gaming uplift.

5800X3D vs 5900X
1665869126250.png

5900X3D vs 5900X
1665869398621.png


Therefore: 5900X3D With 192 L3$ = 5800X3D With 96 L3$.

We can use this to extrapolate that 7800X3D with 96 L3$ would be equal in gaming performance to a 7900X3D With 192 L3$
 
Last edited:

inf64

Diamond Member
Mar 11, 2011
3,686
3,958
136
I don't think it's that straightforward. We don't know at what clocks the 12C Vcache prototype ran. 5900X is a superior gaming chip than 5800X, so I expect that Vcache version of 5900X should be seeing similar uplifts as the 5800X3D saw Vs 5800X.
 

Hans Gruber

Platinum Member
Dec 23, 2006
2,092
1,065
136
Regardless of what (October 20th release) Raptor Lake can do. I do not believe it will be able to overcome Zen 4 3d v-cache chips. Gaming performance is the most important metric for PC buyers. If Intel is planning an October surprise and I think they are. The sooner AMD has the 3D v-cache chips, the better. It sounds like they will arrive Q1 2024. That's not soon enough. If anything I think we will have a confused CPU market for the next 4 or 5 months.

I think Intel will have some Raptor Lake CPU's sipping power and performing well on benchmarks. Then I think AMD will update the bios to do the same type of circus tricks with Zen4. AMD will win the efficiency award and Intel will probably take the performance crown until 3D v-cache arrives.
 

Hans Gruber

Platinum Member
Dec 23, 2006
2,092
1,065
136
I don't think it's that straightforward. We don't know at what clocks the 12C Vcache prototype ran. 5900X is a superior gaming chip than 5800X, so I expect that Vcache version of 5900X should be seeing similar uplifts as the 5800X3D saw Vs 5800X.
This is the kind of stuff that will be possible with Zen 4 3D but Zen 3 is not dead yet.
1665871060423.png
 

inf64

Diamond Member
Mar 11, 2011
3,686
3,958
136
That is what I keep saying..! about 15% give 2% more if you are generous, but it's not going to double the performance because they double the CCDs...!
I think 15-20% above 7700X is what we'll get with 7700X3D. It should comfortably be the best gaming chip until Zen 5 or Arrow Lake arrive.
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
From what I have read, Zen 4 should already beat Raptor Lake. By a small margin.
It will be depend on RAM tuning, a Raptor Lake with DDR4 3600-4000 can't beat a tuned up 7700X, but someone will get their hands on 7000+ DDR5 and claim victory.. The 7800X3D just wont care about RAM Speeds.

Also... about Single vs Dual CCD Gaming..
Screenshot_20221015-170501_Chrome.jpg
 
Last edited:

naad

Member
May 31, 2022
63
176
66
CapframeX has an axe to grind against Ryzen for the last half year, and quite publicly at that, I trust his word as much as I trust userbenchmark.
Even if it is true, one two or even five cases where SMT and 1CCD off is beneficial does not negate the hundred others where it matters absolutely none or improves the situation.
Besides that you got the option of pinning threads and cores to a process with various tools so going the chainsaw method of _disabling_ around half your MT performance for some 6/10 shovelware game is just insane.