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]

Seems like ya'll are grasping at straws.

Its in the interest of fairness. If you're going to make an honest comparison between products and you intentionally optimize your own to show it in the best light while neglecting to do the same for the competition then it just looks bad.

From what I've read, Zen 4 likely won't scale well with higher memory speeds unless you overclock the IF as well and keep it in gear 1 like Zen 3, but Golden Cove can scale with faster memory in gear 2 mode; presumably because it's a monolithic design. Raptor Lake's memory controller will be even more capable of utilizing higher memory speeds.

Wouldn't it be weird if a reviewer configured Zen 4 in gear 2 rather than gear 1? That's intentionally crippling the CPU and you'd scream bloody murder.

Little details like this matter.

 

[RTX2080]

Zen 4 has some surprising tweaks:

The front end to be significantly improved in Zen 4 is not something I heard before.

So apparently the mop cache is more than 50% larger...

https://twitter.com/x/status/1564413952108335105

Anybody remember the leak I retweeted several days ago?

Although the rumor mill is annoying, it would still be some gold here that glitters sometimes. I just lucky enough to catch a glimpse of them.

I'm also very surprise that most performance uplift are brought by these changes. WOW.

https://twitter.com/x/status/1559897403532394501

ROB 320，OP Cache 6.75K

for comparison:
zen2: 224 rob, 4k op cache
zen3: 256 rob, 4k op cache

edit: wow the twitter link got killed quickly!

 

[beginner99]

Havne't been reading this rumor thread and so seeing the news of up to 5.7Ghz. holly molly! never saw that coming. Beating intel at its own game. That must hurt big time.

 

[exquisitechar]

Havne't been reading this rumor thread and so seeing the news of up to 5.7Ghz. holly molly! never saw that coming. Beating intel at its own game. That must hurt big time.

Not really beating, they're on par at stock. The 13900K boosts to 5.8GHz.

 

[Rigg]

Its in the interest of fairness. If you're going to make an honest comparison between products and you intentionally optimize your own to show it in the best light while neglecting to do the same for the competition then it just looks bad.

From what I've read, Zen 4 likely won't scale well with higher memory speeds unless you overclock the IF as well and keep it in gear 1 like Zen 3, but Golden Cove can scale with faster memory in gear 2 mode; presumably because it's a monolithic design. Raptor Lake's memory controller will be even more capable of utilizing higher memory speeds.

Wouldn't it be weird if a reviewer configured Zen 4 in gear 2 rather than gear 1? That's intentionally crippling the CPU and you'd scream bloody murder.

Little details like this matter.

That's all well and good. I don't really disagree with most of what you've just posted. I wasn't disputing that the 12900k setup is non optimal, I'm questioning if it actually matters. Do you have data that shows a significant performance difference between a 12900k w/ 6400 CL32 vs 6000 CL30 in gaming or any other meaningful benchmark? Is this a small detail that makes a practical, demonstrable difference or is it a nitpick? I'm not trying to pick a fight I'm genuinely interested.

 

[coercitiv]

Its in the interest of fairness. If you're going to make an honest comparison between products and you intentionally optimize your own to show it in the best light while neglecting to do the same for the competition then it just looks bad. [...]

Little details like this matter.

On the topic of memory scaling, what are your thoughts on Zen4 DDR5 scaling in gaming?

 

[exquisitechar]

Raichu on Twitter: "This is why I just said almost all rumor of the Zen 4 dram part is not right. FCLK freq is only can OC to almost 2000mhz and higher freq will have whea19. 48 ns is impossible. In fact, most of the latency of OC DRAM will fall in the 6x ns range and a little can jump into 5x ns.😑" / Twitter

That seems reasonable, some of those leaks were silly. Memory OC looks like the same story as usual.
Gaming wise, Raphael definitely won't look good compared to Raptor Lake once overclocking comes into play. More likely than not, even the V-cache parts will lose to OC Raptor Lake, like the 5800X3D loses to OC Alder Lake. I don't do much gaming or overclocking these days, but that's one more unappealing thing about Raphael for enthusiasts.

 

[Thunder 57]

Raichu on Twitter: "This is why I just said almost all rumor of the Zen 4 dram part is not right. FCLK freq is only can OC to almost 2000mhz and higher freq will have whea19. 48 ns is impossible. In fact, most of the latency of OC DRAM will fall in the 6x ns range and a little can jump into 5x ns.😑" / Twitter

That seems reasonable, some of those leaks were silly. Memory OC looks like the same story as usual.
Gaming wise, Raphael definitely won't look good compared to Raptor Lake once overclocking comes into play. More likely than not, even the V-cache parts will lose to OC Raptor Lake, like the 5800X3D loses to OC Alder Lake. I don't do much gaming or overclocking these days, but that's one more unappealing thing about Raphael for enthusiasts.

I've had more than enough of the claims from Raichu. I do not find that source believeable.

 

[Rigg]

Raichu on Twitter: "This is why I just said almost all rumor of the Zen 4 dram part is not right. FCLK freq is only can OC to almost 2000mhz and higher freq will have whea19. 48 ns is impossible. In fact, most of the latency of OC DRAM will fall in the 6x ns range and a little can jump into 5x ns.😑" / Twitter

That seems reasonable, some of those leaks were silly. Memory OC looks like the same story as usual.
Gaming wise, Raphael definitely won't look good compared to Raptor Lake once overclocking comes into play. More likely than not, even the V-cache parts will lose to OC Raptor Lake, like the 5800X3D loses to OC Alder Lake. I don't do much gaming or overclocking these days, but that's one more unappealing thing about Raphael for enthusiasts.

Maybe we can all chip in and get you one of these.

 

[exquisitechar]

I've had more than enough of the claims from Raichu. I do not find that source believeable.

He's most likely your usual Chinese forum dweller posting on Twitter, but he seems pretty decent at judging which Intel leaks he receives from there are real and which aren't. In light of what AMD has shown, it seems reasonable.

 

[tamz_msc]

I've had more than enough of the claims from Raichu. I do not find that source believeable.

Well, 1 month from announcement till launch could probably mean memory issues with the BIOS - he mentions that current AGESA limits frequencies in the follow-up tweet.

Let's face it, Intel has had almost a year to fine tune its memory controller and BIOS to push higher frequencies. It is probably not unreasonable that Raptor Lake will be better at memory OC than Raphael, at least initially.

 

[yuri69]

That 13% IPC figure truly screams for a SPEC validation. Seeing benches with <5% improvement leads one to doubt. During the Zen 3 showcase the min gain was 9%. For Zen 4 it's 1%.

 

[Vope45]

A lot of bugs guys, a lot of bugs. Not as bad as zen 1, but it's bad.

 

[coercitiv]

More likely than not, even the V-cache parts will lose to OC Raptor Lake, like the 5800X3D loses to OC Alder Lake. I don't do much gaming or overclocking these days, but that's one more unappealing thing about Raphael for enthusiasts.

5800X3D had to overcome a considerable gaming performance deficit when compared to Alder Lake, as vanilla Zen 3 was behind in terms of memory technology, max clocks, and also raw core performance.

Vanilla Zen 4 will start from a much more favorable position, AMD claims 11% better performance over 12900K.



V-cache version will add performance on top of than, a basic estimate discussed on this thread puts the gains anywhere from 15% to 25%, let's assume a conservative 20% with some minor overclocking included.

If we combine the gains, we get 30%+ more gaming performance over 12900K /w DDR5 6000. Do you think OC Raptor Lake will offer more?

 

[tamz_msc]

5800X3D had to overcome a considerable gaming performance deficit when compared to Alder Lake, as vanilla Zen 3 was behind in terms of memory technology, max clocks, and also raw core performance.

Vanilla Zen 4 will start from a much more favorable position, AMD claims 11% better performance over 12900K.

View attachment 66793

V-cache version will add performance on top of than, a basic estimate discussed on this thread puts the gains anywhere from 15% to 25%, let's assume a conservative 20% with some minor overclocking included.

If we combine the gains, we get 30%+ more gaming performance over 12900K /w DDR5 6000. Do you think OC Raptor Lake will offer more?

I don't think Zen 4 V-cache will be giving you +20% in gaming across a large swathe of games. Yes, there will be some instances where the gains are very large, but overall DDR5 gains should somewhat negate the effect of additional V-cache.

 

[tamz_msc]

A lot of bugs guys, a lot of bugs. Not as bad as zen 1, but it's bad.

Mostly related to memory I guess?

 

[Thunder 57]

That 13% IPC figure truly screams for a SPEC validation. Seeing benches with <5% improvement leads one to doubt. During the Zen 3 showcase the min gain was 9%. For Zen 4 it's 1%.

Maybe the CPU-z 1T test is garbage?

 

[tamz_msc]

Maybe the CPU-z 1T test is garbage?

It is the most synthetic of synthetic tests. At least with wPrime or SuperPi, you are calculating a known quantity. CPU-Z just spits out scores.

 

[Vope45]

Mostly related to memory I guess?

Yes, also a lot of blue screen and random restarts but that's a given. Everything should be ironed out before launch.

 

[Saylick]

If we combine the gains, we get 30%+ more gaming performance over 12900K /w DDR5 6000. Do you think OC Raptor Lake will offer more?

Better get that water chiller lined up if one is to have any success at Raptor Lake OC matching Zen 4 V-cache lol

 

[tamz_msc]

Yes, also a lot of blue screen and random restarts but that's a given. Everything should be ironed out before launch.

With one month to go hopefully it's not as bad as Zen 1 was.

 

[Thunder 57]

It is the most synthetic of synthetic tests. At least with wPrime or SuperPi, you are calculating a known quantity. CPU-Z just spits out scores.

And they could have just included it since it makes Zen 4 look bad. But, they did. Ever since Lisa Su and Zen AMD's numbers have been pretty fair compared to 3rd party results. I still await those as they are the best.

 

[Thunder 57]

Yes, also a lot of blue screen and random restarts but that's a given. Everything should be ironed out before launch.

According to who?

 

[gdansk]

That 13% IPC figure truly screams for a SPEC validation. Seeing benches with <5% improvement leads one to doubt. During the Zen 3 showcase the min gain was 9%. For Zen 4 it's 1%.

Yes, validation is in order. But it is not unusual for IPC to regress in certain benchmarks while being generally better. A recent example:

 

[Vope45]

According to who?

Source: just trust me bro