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

[Vattila]

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).

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!

 

[Carfax83]

Talking about Raptor lake in a Zen 4 thread is not trolling ? And vcache primarily gaming ? Its started in Milan-x, and THEN went to Zen 3. There are a lot more places that cache helps other than gaming, you guys act the world revolves around gaming.

Raptor Lake is the competition, so it's naturally going to be discussed when talking about Zen 4 3D's potential performance. And while you are correct that V-cache had its inspiration in servers/HPC, the consumer chips are marketed heavily towards gaming. Consumer oriented productivity workloads don't seem to benefit from the V cache.

Like it or not, gaming is the most cache sensitive workload for consumers.

Edit: and the 5800x3d is known as the best gaming chip out there.

Bang for buck I would agree, but as for best performance it's not even in the top 5 since Raptor Lake and Zen 4 are out.

 

[Exist50]

These numbers aren't stock comparisons of 5800X vs. 3D - they are at normalised 4,4Ghz frequency, increasing the relative advantage. The reasoning being that the frequency gap will be much less this time.

The frequency difference between the 5800X and 5800X3D is what, like 5% if that? There's still a large gap unaccounted for. And then consider that closing that frequency gap will probably be the main improvement from Zen4 X3D, so where're the additional gains coming from?

 

[Exist50]

Again, you only speak for some people, not the DC community. The upcoming project (tomorrow) would be killer if both ccds' had 96m cache. See the DC forum for details, but stop speaking for everyone.

The link in question was literally about gaming. But what I said applied identically to any workload. Cross-CCX cache snooping makes no sense with the latencies involved, so there's going to be no gains there, and you're presumably loading each core anyway. Why would you expect a dual die config to provide more advantage from V-Cache than a single die?

 

[Exist50]

The 7600X is already beating the 13900K in gaming with good RAM Kits.

This is just delusional.

 

[Exist50]

He isn't speaking for everyone or anything lmao, idk what's up with this "white knighting" for the DC community.

It's an excuse for him to make absurd performance claims and then defend those as "curing cancer" or some other non-sequitur. He used to say that Zen 3 had twice the performance per core at half the power vs Golden Cove. Obviously, no benchmarks exist to back up those claims, but as we see here, clearly that doesn't stop some people.

This shouldn't even be a discussion. The original link we were responding to was explicitly about gaming performance, so of course that'll be the topic of discussion. And given that AMD explicitly markets its consumer 3D V-Cache chips for gaming, seems perfectly applicable too. Obviously, if a 16c V-Cache chip does come out, it would be useful for other things, so it would be on-topic and pertinent to know performance, but I'm not seeing anyone posting benchmarks for those.

 

[igor_kavinski]

Why would you expect a dual die config to provide more advantage from V-Cache than a single die?

Isn't it possible for game developers to detect the presence of a dual die V-cache CPU and then peg various threads across the two CCXs? That would maximize the amount of cache available to the game.

 

[nicalandia]

This is just delusional.

Really? Why? Because you say so?

The 7600X has been shown to be faster than the 5800X3D(both using 4090 GPU)

The Best Value Gaming CPU: 13600K vs 12600K vs 7600X vs 5800X3D vs 5600X

The Core i5-13600K is possibly the best value CPU all-rounder right now, but when looking purely at gaming the Ryzen 5 7600X is a strong contender as...

www.techspot.com

Here the most recent and comprehensive review of the 5800X3D vs 13900K, The top of the line Intel is 1.5% faster on average than the 13900K where it matters the most.

RTX 4090 & 53 Games: Core i9-13900K vs Ryzen 7 5800X3D Review

Intel's new 13900K offers amazing gaming performance, thanks to improvements to caches, IPC and higher operating frequencies. But is that enough to beat the 3DV Cache-powered AMD Ryzen 7 5800X3D? What about platform cost?

www.techpowerup.com

So who is delusional now? The guy who say that the Stack 3D V cache will not matter because Intel enjoys a hefty 10% Gaming lead or me who is saying that Zen4(7600X) and Raptor Lake(13900K) are basically a match and that the 7800X3D will more than likely beat it soundly?

 

[Exist50]

Isn't it possible for game developers to detect the presence of a dual die V-cache CPU and then peg various threads across the two CCXs? That would maximize the amount of cache available to the game.

My understanding of current game's behavior is that they highly prefer to share one CCX for the sake of inter-thread communication and common data sharing. I doubt a separate V-Cache CCX would substantially change things independent of the core count increase.

 

[nicalandia]

Overall, the 7600x is not in the same league as the 13900K.

That's insane. The 7600X is very much a match in gaming for the 13900K, if The 5800X3D is a virtual match(1.3% with 4090) so is the 7600X who is actually a bit faster than the 5800X3D

 

[Exist50]

The 7600X has been shown to be faster than the 5800X3D(both using 4090 GPU)

The Best Value Gaming CPU: 13600K vs 12600K vs 7600X vs 5800X3D vs 5600X

The Core i5-13600K is possibly the best value CPU all-rounder right now, but when looking purely at gaming the Ryzen 5 7600X is a strong contender as...

www.techspot.com

First of all, your link doesn't even include the 13900k. Beyond that, it's been noted on many occasions that Techspot/Hardware Unboxed's results are highly anomalous. You've already been linked a meta analysis, so I'll avoid the redundancy of posting it again.

So who is delusional now? The guy who say that the Stack 3D V cache will not matter because Intel enjoys a hefty 10% Gaming lead or me who is saying that Zen4(7600X) and Raptor Lake(13900K) are basically a match and that the 7800X3D will more than likely beat it soundly?

No one is claiming that 3D V-Cache "will not matter", nor that it will fail to catch up and eclipse Raptor Lake's gaming performance. The claim in question is that the Zen 4 implementation of 3D V-Cache will provide a substantially greater relative benefit compared to the Zen 3 one, and there's every reason to question the numbers presented.

+15-20% would easily be enough for Zen 4 w/ V-Cache to take the crown, but the claims are about 30%+, and that requires more justification.

 

[Markfw]

The link in question was literally about gaming. But what I said applied identically to any workload. Cross-CCX cache snooping makes no sense with the latencies involved, so there's going to be no gains there, and you're presumably loading each core anyway. Why would you expect a dual die config to provide more advantage from V-Cache than a single die?

Yes, with lasso or other software like it to keep 8 threads of a task on one ccx. And for EPYC, we use 8 threads and enable a bios option that makes each a numa node, and that works the same. (essentially). If curious, see the DC forum.

And again, speaking for regular software, why would Milan-x do so well on quite a few different pieces of software ? Yes, its not always true that it makes a big difference, but it is certainly more than games that benefit.

 

[nicalandia]

No one is claiming that 3D V-Cache "will not matter", nor that it will fail to catch up and eclipse Raptor Lake's gaming performance.

It doesn't need to as Zen4 is already at virtual game parity with Raptor Lake. Whoever thinks otherwise is just clearly Biased.

The 7800X3D Will be a game changer as much as the 5800X3D has been(within 1.3% from the 13900K) and if you think otherwise you are beyond hope.

 

[MarkPost]

7600X vs. 13600K - 54 Game Benchmark | TechSpot

GPU: 4090
MEM: 6000 (7600X) - 6400 (13600K)





13600K vs 7600X | Jarrod'sTech

GPU: 4090
MEM: 6000




13700K vs 7700X | Jarrod'sTech

GPU: 4090
MEM: 6000




13900K vs 7950X | Jarrod'sTech

GPU: 4090
MEM: 6000

 

[Exist50]

Yes, with lasso or other software like it to keep 8 threads of a task on one ccx. And for EPYC, we use 8 threads and enable a bios option that makes each a numa node, and that works the same. (essentially).

That explains how you'd see the same relative improvement from a 2x version as a 1x version. It does not explain how you'd see greater improvement.

And again, speaking for regular software, why would Milan-x do so well on quite a few different pieces of software ? Yes, its not always true that it makes a big difference, but it is certainly more than games that benefit.

I never said that gaming is the only use case that benefits, but that's irrelevant to the question.

 

[Exist50]

It doesn't need to as Zen4 is already at virtual game parity with Raptor Lake. Whoever thinks otherwise is just clearly Biased.

The 7800X3D Will be a game changer as much as the 5800X3D has been(within 1.3% from the 13900K) and if you think otherwise you are beyond hope.

You were the one claiming a 7600X beats the 13900k in gaming. Now you're back to "Zen4 is virtual parity"? Pick one. I responded to what you wrote verbatim, and the data is very clear on the results.

And in the very comment you responded to, I explicitly highlighted that Zen 4 X3D will be the leading gaming chip, and I even gave a range slightly above that which we see for Zen 3. If you cannot be bothered to read my comments, don't waste my time responding to them.

 

[Mopetar]

What's up with Raptor Lake on Battlefield V? That's such a massive outlier that it practically accounts for half of the result at 4K.

 

[nicalandia]

What's up with Raptor Lake on Battlefield V? That's such a massive outlier that it practically accounts for half of the result at 4K.

Not really, from the review

"In our day-one review data, which is based on a 12 game sample, we had the 7600X leading the 13600K by a 3% margin. With that testing expanded to 54 games, the 7600X is 5% faster. If we remove the potentially bugged Battlefield V data (issue with the E-cores?), the 7600X was just 4% faster. Either way as we noted before, we always deem margins of 5% or less to be insignificant, or in other words a tie.

 

[biostud]

There could be a trend that modern AAA titles with RT enabled actually also taxed the CPU quite heavily like Spiderman: Miles Morales, in which case Intel would probably gain an advantage with their e-cores, in the lower segment with AMD's 6-8 core CPU's.

But no matter if you decide to buy Intel or AMD, you will probably not notice the difference in games, when you're not benchmarking if you buy intel or AMD at the same pricepoint.

 

[nicalandia]

There could be a trend that modern AAA titles with RT enabled actually also taxed the CPU quite heavily like Spiderman: Miles Morales, in which case Intel would probably gain an advantage with their e-cores, in the lower segment with AMD's 6-8 core CPU's.

Currently those e cores are pretty useless at gaming.

RTX 4090 & 53 Games: Core i9-13900K E-Cores Enabled vs Disabled Review

In this week's TPU50 Megabench we're testing whether you can unlock additional gaming performance by disabling the E-Cores on a Raptor Lake Core i9-13900K processor. For our benchmarks we used the mighty GeForce RTX 4090, with 53 games at three resolutions.

www.techpowerup.com

 

[ondma]

There could be a trend that modern AAA titles with RT enabled actually also taxed the CPU quite heavily like Spiderman: Miles Morales, in which case Intel would probably gain an advantage with their e-cores, in the lower segment with AMD's 6-8 core CPU's.

But no matter if you decide to buy Intel or AMD, you will probably not notice the difference in games, when you're not benchmarking if you buy intel or AMD at the same pricepoint.

True, but problem is whether the games can effectively use the E cores. I dont know why someone hasnt done streaming or other heavy tasks while gaming to see if the E cores can help offload the cpu when other tasks are active during gaming.

 

[biostud]

Currently those e cores are pretty useless at gaming.

View attachment 72402

RTX 4090 & 53 Games: Core i9-13900K E-Cores Enabled vs Disabled Review

In this week's TPU50 Megabench we're testing whether you can unlock additional gaming performance by disabling the E-Cores on a Raptor Lake Core i9-13900K processor. For our benchmarks we used the mighty GeForce RTX 4090, with 53 games at three resolutions.

www.techpowerup.com

I know (and so is most cores above the first six) , that is why I mentioned a single game, and only with RT on. It would be interesting to see how it behaved with e-cores disabled.

There are some outlier games that can tax the CPU quite heavily, and small amount of these can utilize more than 6-8 cores and see a performance increase. But before we see this as normal, Zen 6 has probably been released.

 

[Kaluan]

The frequency difference between the 5800X and 5800X3D is what, like 5% if that? There's still a large gap unaccounted for. And then consider that closing that frequency gap will probably be the main improvement from Zen4 X3D, so where're the additional gains coming from?

If that? Data is out there if one wants to check it
~5% higher clock in mixed/light workloads/typical gaming
~6% higher in heavy multithreaded (eg Blender)
~9% higher in single threaded
Not bad considering it's barely any slower (2%?) than vanilla in productivity, something they never marketed it for (this may change with Zen4 X3D).

As to where other potential gains could be coming from? Who know, they may lower the L3 latency penalty from 3-4ns to 3ns or less somehow. We also have no idea how changes (cache or otherwise) in Zen4 react to increased L3 pools. Seeing how forward thinking Zen3 design was, at least 1 year post launch, with revolutionary stacking tech obviously not being a afterthought, I doubt Zen4 would be just a repeat. Pretty smug to think we know everything and won't be surprised TBH.

Reminder to people that this exists:

Deep Diving Zen 3 V-Cache

This is the deeper dive of AMD’s V-Cache that we teased with our short latency article and we will be covering a little more on the latency front along with the bandwidth behavior of V-Cache …

chipsandcheese.com

 

[Kaluan]

What's up with Raptor Lake on Battlefield V? That's such a massive outlier that it practically accounts for half of the result at 4K.

Oh no, not this again ☠☠☠

 

[coercitiv]

that is why I mentioned a single game

AFAIk Spiderman is to be considered an outlier from a CPU load PoV, it leans on the CPU for a workload that will soon be GPU accelerated. The CPu grunt is welcomed for the transitional period, but does not necessarily define a trend.

 

[ondma]

7600X vs. 13600K - 54 Game Benchmark | TechSpot

GPU: 4090
MEM: 6000 (7600X) - 6400 (13600K)

View attachment 72391
View attachment 72393
View attachment 72394

13600K vs 7600X | Jarrod'sTech

GPU: 4090
MEM: 6000

View attachment 72396


13700K vs 7700X | Jarrod'sTech

GPU: 4090
MEM: 6000

View attachment 72395


13900K vs 7950X | Jarrod'sTech

GPU: 4090
MEM: 6000

View attachment 72397

These results seen a bit low for intel, compared to other tests I have seen, especially the 7700k on average being more than 10% faster than 13700k, and if I read the charts correctly that is with BF V removed.