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]

You can't just double all the numbers, it does not work like that. Your posts are approaching trolling.

That has been the approach to calculating absolute DRAM latency for a while. However, I will note that platform, DRAM technology, memory controller, etc. all have an effect on this. My Zen 3 at 3800C14 has higher measured latency than my Zen 4 at 6200C28, despite the latency calculation giving my zen 4 a disadvantage.

 

[Hans Gruber]

You can't just double all the numbers, it does not work like that. Your posts are approaching trolling.

You are one of those people

You can't just double all the numbers, it does not work like that. Your posts are approaching trolling.

With all the insults you throw around on this forum. I cannot understand why they allow you to be a moderator.





You still cannot call out moderators by their titles.


esquared
Anandtech Forum Director

 

[itsmydamnation]

Clock for clock. Cas 14 latency DDR4 @ 3800mhz would be equal to Cas 28 DDR5 7600mhz if you double the speed. Obviously DDR5 has some raw performance gains in architecture. Even if you have E die 16Cas @ 3800DDR4 would be Cas 32 DDR5 7600mhz.

I am equaling out the performance to give you an idea of the latencies of top tier DDR4 vs. DDR5.

My point is that AMD needs much higher frequency support for DDR5.

no they don't , which is why it makes no sense

what matter is timing in NS , not Cas in isolation , not clock in isolation , CAS * Clock = timing

If you can make DDR5 8000 at CAS 40 for example , you can likely get DDR4 4000 CAS 20 from the exact same kit.

so if 6400 is the best data rate for AM5 and you can get 8000 for intel , vendors just need to sell AMD targeted Kits @ 6400 with CAS that reaches the first word latency in NS as the 8000 intel kit.

people love to over value memory bandwidth , vast majority of consumer workloads will favour latency

 

[Thunder 57]

You can't just double all the numbers, it does not work like that. Your posts are approaching trolling.

Downvoted for giving correct information. Have an upvote on me!

 

[Hans Gruber]

That has been the approach to calculating absolute DRAM latency for a while. However, I will note that platform, DRAM technology, memory controller, etc. all have an effect on this. My Zen 3 at 3800C14 has higher measured latency than my Zen 4 at 6200C28, despite the latency calculation giving my zen 4 a disadvantage.

That would be the architectural enhancements in DDR5 th

no they don't , which is why it makes no sense

what matter is timing in NS , not Cas in isolation , not clock in isolation , CAS * Clock = timing

If you can make DDR5 8000 at CAS 40 for example , you can likely get DDR4 4000 CAS 20 from the exact same kit.

so if 6400 is the best data rate for AM5 and you can get 8000 for intel , vendors just need to sell AMD targeted Kits @ 6400 with CAS that reaches the first word latency in NS as the 8000 intel kit.

people love to over value memory bandwidth , vast majority of consumer workloads will favour latency

And the numbers speak volumes. DDR4 3600mhz even at Cas 18 smokes DDR4 3200mhz 14-14-14 in performance.

I mentioned in my earlier post that DDR5 has enhancements that improve performance over DDR4. The doubling of Cas latency and frequency is merely a ball park figure.

Consider that Zen 1 maxed out at 3000-3200mhz. I could get 3400mhz but I had a B-die kit and a good bios. 3200mhz was all that was stable. It was not until the big AGESA bios update that allowed Zen2 to achieve 3800mhz memory frequency.

You can search forum history. People said it wasn't a bios issue it was a motherboard hardware issue because some motherboards could do 3800mhz and other could not. It was a bios issue.

You can run Zen 2 and Zen 3 way beyond 4000mhz but you lose your 1:1 memory divider and introduce a ton of latency.

Zen 4 cannot even achieve the DDR5 6400mhz that AMD claimed was the sweet spot for Zen 4 CPU's. I remember when 3733mhz was the coffins corner for Zen 2. It would boot and run but would have memory issues and phantom crashes before the big AGESA bios. AMD said that AM4 would be supported all the way through Zen 3. Then it wasn't for B350/X370 motherboards. AMD released a bios update and now you can run Zen 3 on the oldest AM4 motherboards.

From what I understand with Zen 4. Regardless of memory dividers and timings. You cannot run any DDR5 rated or not stick of memory above 6400mhz without the system crashing or not booting.

 

[scannall]

You can't just double all the numbers, it does not work like that. Your posts are approaching trolling.

Actually you pretty much can. CAS Latency explained. CAS latency is RAM clock cycles, not a measurement of time. So if your DDR5 is taking 28 cycles but running twice as fast a clock speed then they would be *roughly* equal to DDR4 at 14 cycles. DDR5 has some big architectural improvements in regards to banks etc which allows more throughput at a given latency and speed. DDR5 spec

 

[Kocicak]

I think AMD needs to speed up their CPU release cadence.

WHAT? These are not trainers! These are extremely complex and expensive things to develop, and also to manufacture and buy! Users (at least the corporate users of servers) need to run them for years.

There is no reason at all to come with a new CPU each half a year!

 

[NostaSeronx]

A new CPU each half year?! Nah, we need browser style update cycles from CPUs.

If Zen109 isn't out in 12 years... I'll be sorely dissapoint.

 

[Thibsie]

This makes Zero sense ......

what is going on in this thread........

What did you expect ?
Always the usual suspects...

 

[leoneazzurro]

https://twitter.com/x/status/1628724904983134210

Ryzen 9 7945HX Geekbench 5 score

No info about the set power limit ( it'a an Asus ROG Zephyrus Duo 16 GX650PY_GX650PY in Performance mode)

 

[Det0x]

7950x3d GB5

As we can see, the 7950X3D scored 2271 points in the single-core test and 24727 points in multi-core. That’s indeed much better than previous scores. It is worth noting that the 7950X3D will not support overclocking, at least not in a traditional form. There will be no frequency or multiplier modifications allowed, but users will have access to memory overclocking, and Precision Boost Overdrive and undervolting. That said, the new highest score was definitely not achieved with overclocking.

AMD Ryzen 9 7950X3D beats Core i9-13900K in new Geekbench single-core test - VideoCardz.com

AMD Ryzen 9 7950X with 3D V-Cache faster than Core i9-13900K New benchmarks featuring flagship Ryzen 9 CPU. The 3D V-Cache is not expected to show its strength in most synthetic benchmarks such as Geekbench. Only cache-sensitive and gaming workloads will unlock the potential of AMD’s newest CPU...

videocardz.com

 

[Timmah!]

7950x3d GB5

View attachment 77068

AMD Ryzen 9 7950X3D beats Core i9-13900K in new Geekbench single-core test - VideoCardz.com

AMD Ryzen 9 7950X with 3D V-Cache faster than Core i9-13900K New benchmarks featuring flagship Ryzen 9 CPU. The 3D V-Cache is not expected to show its strength in most synthetic benchmarks such as Geekbench. Only cache-sensitive and gaming workloads will unlock the potential of AMD’s newest CPU...

videocardz.com

well, only one core showing 5,7ghz is bit weak to support the claim the cpu can boost up to 5,7. I mean, did not the same list of vanilla 7950x frequencies shown all the cores being able to do 5,7?

Anywy, its still just 200~300 MHz less - i wonder if one will be able to PBO/CO tune it to boost to the same speeds as vanilla one…

Monday cant come fast enough and some proper review showing the clocks, all around performance and its tuneability.

 

[deasd]

7950X3D, MT 24727

Gigabyte Technology Co., Ltd. X670E AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. X670E AORUS MASTER with an AMD Ryzen 9 7950X3D processor.

browser.geekbench.com

7950X, MT 24095

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 7950X processor.

browser.geekbench.com

While having similar MT score, 7950X3D has exactly 200-300 MHZ lower frequency than the 7950X.

 

[igor_kavinski]

Can't wait to see folks buy a $699 CPU, only to turn off the CCD without the V-cache

 

[igor_kavinski]

Gigabyte Technology Co., Ltd. X670E AORUS MASTER vs ASUS System Product Name - Geekbench

Machine learning, speech recognition and navigation MT workloads benefit from the V-cache while SQLite, physics, clang, gaussian blur and raytracing MT scores suffer. Can't have everything

 

[deasd]

Can't wait to see folks buy a $699 CPU, only to turn off the CCD without the V-cache

Since dual CCDs don't have any shared cache, like P and E cores, shutting down one CCD won't make any diference, except you can OC the V-cache CCD higher due to TDP and heat dissipating.

And you can even shutting down everything you can like SMT and some unused cores....

 

[Kocicak]

Gigabyte Technology Co., Ltd. X670E AORUS MASTER vs ASUS System Product Name - Geekbench

Machine learning, speech recognition and navigation MT workloads benefit from the V-cache while SQLite, physics, clang, gaussian blur and raytracing MT scores suffer. Can't have everything

The improvements in 5 MT tasks out of 20 are better than 15%, overall in this task set the MT improvements are just 2,6 %.

 

[Timmah!]

7950X3D, MT 24727

Gigabyte Technology Co., Ltd. X670E AORUS MASTER - Geekbench

Benchmark results for a Gigabyte Technology Co., Ltd. X670E AORUS MASTER with an AMD Ryzen 9 7950X3D processor.

browser.geekbench.com

7950X, MT 24095

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 7950X processor.

browser.geekbench.com

While having similar MT score, 7950X3D has exactly 200-300 MHZ lower frequency than the 7950X.

Thats in Geekbench. How about Cinebench or even better Vray.

 

[leoneazzurro]

Supposed leaked AMD slides about 7950X3D in gaming

AMD Ryzen 9 7950X3D is 6% faster in gaming than Core i9-13900K according to leaked AMD review guide - VideoCardz.com

 

[stuff_me_good]

If true we seem to get diminishing returns out of bigger cache. I guess the internal cache is big enough for price performance sweet spot?

 

[Abwx]

https://twitter.com/x/status/1628724904983134210

Ryzen 9 7945HX Geekbench 5 score

No info about the set power limit ( it'a an Asus ROG Zephyrus Duo 16 GX650PY_GX650PY in Performance mode)

Power is about 75W.

It score 5% lower than a 7900X and according to Computerbase the 7950X@88W perform 4% better than a stock 7900X.

 

[Det0x]

More leaks

AMD Ryzen 9 7950X3D is 6% faster in gaming than Core i9-13900K according to leaked AMD review guide - VideoCardz.com

AMD Ryzen 9 7950X3D official performance leaks out HDTecnologia have leaked new data from confidential AMD Ryzen 7000X3D review guide. What one should know about review guides is that they tend to focus on games that are optimized for specific hardware. This is a reference for reviewers to...

videocardz.com

 

[utahraptor]

Supposed leaked AMD slides about 7950X3D in gaming

AMD Ryzen 9 7950X3D is 6% faster in gaming than Core i9-13900K according to leaked AMD review guide - VideoCardz.com

It looks like a pretty good showing for the 7950X3D. The slight defecit in the 2nd half of the synthetics is very minor.

The only whataboutisms I can envision are:

• Price difference
• Core i9-13900KS not tested
• Core i9-13900K was not allocated DDR5 8,000
• Mobo cost

I'm glad I waited for the 7950X3D.

 

[LightningZ71]

When the 5800x3d came out, it was compared against cache limited 11th gen Intel products and Alder Lake products with less cache than the current Raptor lake products. Current Raptor lake has 36MB of L3 and a lot of L2 coupled with a notably faster memory subsystem than previous processors. I wouldn't expect the large vCache to make a huge difference across the board any more, and only those few games that have moderate working sets that fall in the 50-100MB of size will strongly favor AMD. Anything larger will swing back to Intel.

 

[leoneazzurro]

It looks like a pretty good showing for the 7950X3D. The slight defecit in the 2nd half of the synthetics is very minor.

The only whataboutisms I can envision are:

• Price difference
• Core i9-13900KS not tested
• Core i9-13900K was not allocated DDR5 8,000
• Mobo cost

I'm glad I waited for the 7950X3D.

Yeah, considering that the worst showing in these slides is not even a game and that the X3D version is limited to 120W TDP, it seems not bad at all. Provided we can trust these numbers, which must be confirmed by third parties. A pity that there are no rumors about a mobile version of these processors...