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!

 

[Hail The Brain Slug]

Ah, where's that (Intel branded) copium huffing device meme when you need it...


AMD Zen4X3D is "only this" and "only that"... my boy knows 13900K is "only" 6-7% faster than 5800X3D at 1080p, right?

Going by AMD's numbers, which they didn't mislead with back in the 5000X3D drum-up period (and I hope they don't do now), 7800X3D is ~21% faster (8 game average) than 5800X3D and 7950X3D is ~16% faster (6 game average) than 13900K. That fits pretty great into the "5800X3D 6% slower than 13900K, 7000X3D 16% faster than 13900K... so ~21% faster than 5000X3D" triangle.
(If we asume 7800, 7900 and 7950 X3Ds all perform roughly the same in games)

It will be interesting to see if the dual CCD Zen 4 3D scheduling will "solve" the cross-CCD penalty that the 7900X and 7950X currently see in some games against the 7700X. If it does, and it makes 3D cache favored games run entirely on the 3D CCD, it should perform practically the same I'd think. Problem is if some games that don't favor the cache end up cross-CCD or not. They do now, and that seems like something AMD could have or might work on in the future.

 

[Hans Gruber]

I think it would make sense for AMD to roll back the TDP for the 3D-vache to the Zen 3 levels (105w). If the non X processors can push 5.2-5.3ghz with reduced TDP numbers. It would make sense for AMD to roll back the TDP to make the power/performance ratio much better.

AMD had a good thing going with Zen2 and Zen3. Why did they decide to screw up their huge power consumption advantage with Zen4?

 

[Kaluan]

It will be interesting to see if the dual CCD Zen 4 3D scheduling will "solve" the cross-CCD penalty that the 7900X and 7950X currently see in some games against the 7700X. If it does, and it makes 3D cache favored games run entirely on the 3D CCD, it should perform practically the same I'd think. Problem is if some games that don't favor the cache end up cross-CCD or not. They do now, and that seems like something AMD could have or might work on in the future.

Here's something that might interest you (and everyone else interested in dual-CCD gaming):

https://twitter.com/x/status/1610943615622348800

Perhaps these W11 changes can exted to non-X3D dual-CCD Ryzen 7000 in some way?


Edit: And a bonus for me, seeing that troll get owned again.

 

[Joe NYC]

So what you guys are saying. They didn't fix the 3d v-cache temperature limit with the Zen 4 3D version. So clock speeds will be 400-500mhz lower than standard 7700x/7800x CPU's?

5800x3d issue was with voltage, not temperature, but these are somewhat related... Temperature is less of a problem than a hard voltage limit - for overclocking...

 

[Saylick]

CapframeX being a Biased Intel fan yet again..
View attachment 74019
View attachment 74021

Ugh, that guy spends more time posting bait tweets that are either pro-Intel or pro-Nvidia rather than tweets that promote his own software.

Also, I don't understand people's self-proclaimed responsibility for having to "keep AMD hype in check", especially when I don't really see the market as being heavily AMD favored to begin with, and not when AMD is still the underdog in terms of company size. I refuse to believe that a handful of Youtube leakers constitute the bastion of "over hyping" that he's trying to "correct".

 

[RnR_au]

Could be a little bit of both. Is the v-cache applied to all chiplets on the wafer before it's diced, or individually?

Individually I believe. I think AMD is using TSMC's CoW tech and not their WoW (wafer-on-wafer).

 

[Hail The Brain Slug]

Here's something that might interest you (and everyone else interested in dual-CCD gaming):

https://twitter.com/x/status/1610943615622348800

Perhaps these W11 changes can exted to non-X3D dual-CCD Ryzen 7000 in some way?


Edit: And a bonus for me, seeing that troll get owned again.

.... Seems suspect because as far as I am aware from every source I've seen, 5800X3D VMax was 1.35V, not 1.1V. Kind of an odd fact to get wrong....?

I have seen the details about PBO+CO being available on Zen 4 3D, as well as the other details regarding scheduler management. My question was made knowing everything presented there.

Current behavior seems to be to favor CCD0, and if enough threads are requested (even if they do little work) it will begin to schedule them on CCD1. Just because they instruct the scheduler to favor the 3D Cache CCD doesn't mean it won't spill over additional threads.

Granted they seem to be putting forth game-specific and application-specific instructions for the scheduler, according to what I've seen, so perhaps they will also tackle this issue for both 3D and non 3D dual CCD chips.

 

[Karnak]

CapframeX being a Biased Intel fan yet again..
View attachment 74019
View attachment 74021

Ah these guys. Especially the last tweet and talking about AMD CPUs getting overhyped.

https://twitter.com/x/status/1322844188443090945

I'm still waiting for RKL and 5.6GHz lol.

 

[nicalandia]

Amd 7950X3D. 3DX CCD on the left

 

[Hans Gruber]

5800x3d issue was with voltage, not temperature, but these are somewhat related... Temperature is less of a problem than a hard voltage limit - for overclocking...

You can still push a Zen4 3D chip to 1.4v with a 105w TDP. That's all I am saying.

 

[Saylick]

Ah these guys. Especially the last tweet and talking about AMD CPUs getting overhyped.

https://twitter.com/x/status/1322844188443090945

I'm still waiting for RKL and 5.6GHz lol.

I guess when you are an Intel fanboy, even the mildest consumer interest in the competing company's new products constitutes as too much hype.

 

[Markfw]

Amd 7950X3D. 3DX CCD on the left

View attachment 74022

Has it been confirmed that only one ccd has the extra cache ? For a 7950x3d, the extra cache would be a benefit for BOTH ccds' when the software uses all 32 threads (primegrid for example)

 

[Thunder 57]

I'm curious about the performance of Zen 4 3D in games that use ray tracing. Raptor Lake has a huge performance lead over Zen 4 in heavy RT games like Spider-Man Miles Morales that use the CPU for BVH building and maintenance. Tests have it at nearly 50%.

The 3D cache should make a nice dent in that gap, but I doubt it will be enough to fully close it.

If I had $1 every time you posted this. If Intel was 50% behind, I bet you'd be blaming the game and calling it broken.

 

[Harry_Wild]

AMD Ryzen 7000 Series Desktop Line-Up
AnandTech	Cores Threads	Base Freq	Turbo Freq	Memory Support	L3 Cache	TDP	MSRP
Ryzen 9 7950X	16C / 32T	4.5 GHz	5.7 GHz	DDR5-5200	64 MB	170 W	$699
Ryzen 9 7900X	12C / 24T	4.7 GHz	5.6 GHz	DDR5-5200	64 MB	170 W	$549
Ryzen 9 7900	12C / 24T	3.6 GHz	5.4 GHz	DDR5-5200	64 MB	65 W	$429
Ryzen 7 7700X	8C / 16T	4.5 GHz	5.4 GHz	DDR5-5200	32 MB	105 W	$399
Ryzen 7 7700	8C / 16T	3.6 GHz	5.3 GHz	DDR5-5200	32 MB	65 W	$329
Ryzen 5 7600X	6C / 12T	4.7 GHz	5.3 GHz	DDR5-5200	32 MB	105 W	$299
Ryzen 5 7600	6C / 12T	3.8 GHz	5.1 GHz	DDR5-5200	32 MB	65 W	$229

 

[nicalandia]

Has it been confirmed that only one ccd has the extra cache ? For a 7950x3d, the extra cache would be a benefit for BOTH ccds' when the software uses all 32 threads (primegrid for example)

100% connfirmed by AMD.

 

[Saylick]

Thanks for posting the table. I added in the current AMD Store prices, if you don't mind.

AMD Ryzen 7000 Series Desktop Line-Up
AnandTech	Cores Threads	Base Freq	Turbo Freq	Memory Support	L3 Cache	TDP	MSRP	AMD Store Price
Ryzen 9 7950X	16C / 32T	4.5 GHz	5.7 GHz	DDR5-5200	64 MB	170 W	$699	$574
Ryzen 9 7900X	12C / 24T	4.7 GHz	5.6 GHz	DDR5-5200	64 MB	170 W	$549	$474
Ryzen 9 7900	12C / 24T	3.6 GHz	5.4 GHz	DDR5-5200	64 MB	65 W	$429	$429
Ryzen 7 7700X	8C / 16T	4.5 GHz	5.4 GHz	DDR5-5200	32 MB	105 W	$399	$349
Ryzen 7 7700	8C / 16T	3.6 GHz	5.3 GHz	DDR5-5200	32 MB	65 W	$329	$329
Ryzen 5 7600X	6C / 12T	4.7 GHz	5.3 GHz	DDR5-5200	32 MB	105 W	$299	$249
Ryzen 5 7600	6C / 12T	3.8 GHz	5.1 GHz	DDR5-5200	32 MB	65 W	$229	$229

 

[Timmah!]

Tough crowd here. The best gaming CPU is "disappointing".

well, dont want to say expectation, but the “wish” was definitely for the CPU clocking as High as 7950x with 64 MB of vcache on each die. That would have been ideal.

instead, we are given sort of hybrid, with vcache on just single die, that has exactly the same downside as 5000 series, so the one thing we wished AMD to fix, the clock penalty, they clearly did not. So as a result we will have to rely on Windows to correctly schedule threads to proper ccds/cores. Thats hardly optimal, its added complexity that may cause issues.

Not to mention because of the lower clocks on the vcache ccd we may see lower scores in Cinebench and the likes, compared to vanilla 7950x, so 3d might not be always the best cpu.

So yes, its disappointing, because its not the ultimate CPU we wanted, but another trade-off, just bit more complex than 5800x3d was.

 

[nicalandia]

Another shot of the 7950X3D

 

[Carfax83]

If I had $1 every time you posted this. If Intel was 50% behind, I bet you'd be blaming the game and calling it broken.

Except it's not just with that particular game or that particular engine. Nice try though.

 

[Carfax83]

If I had $1 every time you posted this. If Intel was 50% behind, I bet you'd be blaming the game and calling it broken.

Except it's not just with that particular game or that particular engine. Nice try though.

 

[Joe NYC]

well, dont want to say expectation, but the “wish” was definitely for the CPU clocking as High as 7950x with 64 MB of vcache on each die. That would have been ideal.

instead, we are given sort of hybrid, with vcache on just single die, that has exactly the same downside as 5000 series, so the one thing we wished AMD to fix, the clock penalty, they clearly did not. So as a result we will have to rely on Windows to correctly schedule threads to proper ccds/cores. Thats hardly optimal, its added complexity that may cause issues.

Not to mention because of the lower clocks on the vcache ccd we may see lower scores in Cinebench and the likes, compared to vanilla 7950x, so 3d might not be always the best cpu.

So yes, its disappointing, because its not the ultimate CPU we wanted, but another trade-off, just bit more complex than 5800x3d was.

In general, I wish that AMD truly recognized the gaming market by selling a single CCD 7850x and 7850x3d with the highest binned parts.

Instead, these CPUs, 5800x, 7700x are a dumpster for worst binned parts. Which would not be a problem if there was also a highest binned part to go along with them for gamers.

We end up with counter-intuitive CPUs created by marketing, where 2 CCD CPUs, which have to split the thermal limit in half, have higher clock speeds than single CCD CPUs that have the full power and thermal envelope at their disposable. Which is all due to weird marketing / segmentation...

As far as the V-Cache versions of Zen 4, we are not really sure how much limitation is due to binning and how much is attributed to other areas / variables.

 

[SK10H]

well, dont want to say expectation, but the “wish” was definitely for the CPU clocking as High as 7950x with 64 MB of vcache on each die. That would have been ideal.

instead, we are given sort of hybrid, with vcache on just single die, that has exactly the same downside as 5000 series, so the one thing we wished AMD to fix, the clock penalty, they clearly did not. So as a result we will have to rely on Windows to correctly schedule threads to proper ccds/cores. Thats hardly optimal, its added complexity that may cause issues.

Not to mention because of the lower clocks on the vcache ccd we may see lower scores in Cinebench and the likes, compared to vanilla 7950x, so 3d might not be always the best cpu.

So yes, its disappointing, because its not the ultimate CPU we wanted, but another trade-off, just bit more complex than 5800x3d was.

From a 5800x owner standpoint having forced to upgrade to a x570 mb for zen3 in early days with 6 yr old Ram still, and having looked at the rumored 7900x3d spec before, it goes from a immediate purchase to a wait and see approach with these "lower than expected" reality.

I can easily justify the platform upgrade cost with 17% clock speed, 10% avg IPC, double vcache on the 6core chiplet for gaming workload at 20% bonus , at the speculated $549 price. It would also be OK if it was double vache @5.3 as well at a lower rumored expectations.

The unbalanced vcache is a little crappy with the 6 cores ccd, since games have 8 cores programmed in mind these days at least and the decision would be should it use a virtual core on vcache ccd or actual physical core on the non vcache. Vcache gaming was all about significantly improving the 1% fps latency so it doesn't chop, not so easy with the unbalanced approach when it ran out of threads.
More time to wait for getting cheaper ddr5 I guess. 😔

 

[inf64]

well, dont want to say expectation, but the “wish” was definitely for the CPU clocking as High as 7950x with 64 MB of vcache on each die. That would have been ideal.

instead, we are given sort of hybrid, with vcache on just single die, that has exactly the same downside as 5000 series, so the one thing we wished AMD to fix, the clock penalty, they clearly did not. So as a result we will have to rely on Windows to correctly schedule threads to proper ccds/cores. Thats hardly optimal, its added complexity that may cause issues.

Not to mention because of the lower clocks on the vcache ccd we may see lower scores in Cinebench and the likes, compared to vanilla 7950x, so 3d might not be always the best cpu.

So yes, its disappointing, because its not the ultimate CPU we wanted, but another trade-off, just bit more complex than 5800x3d was.

AMD addressed this in the PCWorld video. They did the simulations and saw the chip with two stacked Vcaches had marginal benefits Vs the one they are launching. The hypothetical SKU would have lower max boost/all core boost clocks and would generally underperform in non-gaming tasks with worse power/thermals while offering marginal benefits for gaming workloads. It makes perfect sense. Once we get more games to use >16 threads and Vcache not requiring noticeably lower clocks, it will be pointless to have SKUs with such configuration.

This leaves the door open for future hybrid SKUs as someone else mentioned in the Zen 5 thread : one CCD stacked with Vcache while the other using denser variants with 2x the core count and no Vcache.

 

[biostud]

How much difference would there be in all core frequency with same thermal limit for the 3D versions versus the regular?

 

[Timmah!]

AMD addressed this in the PCWorld video. They did the simulations and saw the chip with two stacked Vcaches had marginal benefits Vs the one they are launching. The hypothetical SKU would have lower max boost/all core boost clocks and would generally underperform in non-gaming tasks with worse power/thermals while offering marginal benefits for gaming workloads. It makes perfect sense. Once we get more games to use >16 threads and Vcache not requiring noticeably lower clocks, it will be pointless to have SKUs with such configuration.

This leaves the door open for future hybrid SKUs as someone else mentioned in the Zen 5 thread : one CCD stacked with Vcache while the other using denser variants with 2x the core count and no Vcache.

yes, it would have had lower boost/clocks, cause they failed to fix the clock penalty issue. If they did manage to do it, you would be better off with both dies stacked and all cores being uniform, so there would be no need for any special scheduling.

As it is, this CPU looks to me more like 2x 8 core cpus than single 16 core. It still will be pretty good no doubt, and i will wait for reviews and all those questions answered (like all core clocks and performance), so i might change my mind as a result, but generally i am no fan of these hybrid designs, i got 7950x (still could return it and replace for x3d) over intel after 4 intel rigs since core2duo days, as i did not fancy little cores shenanigans, and while this is not the same, there are similarities, so i would be hypocritical to give this a pass.