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!

 

[Geddagod]

That was to be expected.
The most impressive comparison from the tweet above IMHO is this:
EPYC 9754 128c - 132.5k
EPYC 9654 96c - 94k
That's +41% for only around 70% 5nm area at the same TDP.

What's the difference between the 9654p and 9654? Because the 9654p scores 108k, nearly 15% higher. TDP seems to be the same as well between the two parts, at least according to TPU.

 

[jpiniero]

What's the difference between the 9654p and 9654? Because the 9654p scores 108k, nearly 15% higher. TDP seems to be the same as well between the two parts, at least according to TPU.

The P you don't get dual processor support.

 

[BorisTheBlade82]

What's the difference between the 9654p and 9654? Because the 9654p scores 108k, nearly 15% higher. TDP seems to be the same as well between the two parts, at least according to TPU.

I see what you mean. The 9654P result seems more plausible. 41% from only 33% more cores would have been too good to be true. Although it is not impossible, given TDP constraints and a much better Perf/W.
But even so, it is 23% more performance.

 

[Geddagod]

It's not that surprising. The last 100 MHz always eats up a lot more power than the first 100 MHz.

The extra frequency isn't the killer, but the voltage necessary to drive the chip at that clock speeds while keeping it stable is what does it.

Power = CV[sup]2[/sup]F, where V is the voltage and F is the frequency.

If you only raise or drop the frequency it just increases the power proportional to that. However, the voltage usually rises or falls along with the maximum frequency so a 10% reduction in both voltage and frequency is actually a .9[sup]3[/sup] or ~37% reduction in power. Add in 50% more cores and you're only using 10% more power than originally.

I actually didn't believe that last sentence until I checked the 7950x vs the 7900x, the 7950x consumes ~15% more energy than the 7900x while clocking very similarly, and having 50% more cores (16 vs 12). The more you know ig.

 

[nicalandia]

The Performance/Area is Off the Charts here.

 

[DrMrLordX]

Interesting that 1P Bergamo is beating 2P Genoa in anything. Is the 2P system suffering from NUMA penalties?

 

[nicalandia]

Interesting that 1P Bergamo is beating 2P Genoa in anything. Is the 2P system suffering from NUMA penalties?

Most apps including V-Ray, Geekbench, Cinebench suffer latency issues along with NUMA issues when they have to jump to the second socket.

 

[nicalandia]

I have found Benchmarks results for Bergamos and Genoa-X, just give me a few minutes to post them and compare it to base Genoa

 

[nicalandia]

EPYC 9684X(Genoa-X) vs 9654(Genoa), The 3D Cache show great gains in FPU performance

 

[Exist50]

The Performance/Area is Off the Charts here.

View attachment 81739

Clearly not a good comparison. All things considered, 128c Genoa should be at worst comparable to 128c Bergamo, and significantly better in some cases. Looks like there's no multi-socket scaling, for one thing, so the "2x" is misleading, and throws into question the rest as well.

 

[coercitiv]

Looks like there's no multi-socket scaling, for one thing, so the "2x" is misleading, and throws into question the rest as well.

The 9554 is a 64 core CPU, wccftech is just not good at presenting information. We get shiny graphs though, isn't that more important?

 

[Saylick]

The 9554 is a 64 core CPU, wccftech is just not good at presenting information. We get shiny graphs though, isn't that more important?

View attachment 81764

Lol... Good ol' WCCFTech. Their graphs are an accurate reflection of their ethos: looks good on the surface, but dig a little deeper and you realize they can't get even the basic facts straight.

 

[nicalandia]

Lol... Good ol' WCCFTech. Their graphs are an accurate reflection of their ethos: looks good on the surface, but dig a little deeper and you realize they can't get even the basic facts straight.

Whats wrong with the Graph? It show a single Bergamo with total of 128 cores slighty ahead of two 64 cores(total of 128 cores)

 

[Exist50]

Whats wrong with the Graph? It show a single Bergamo with total of 128 cores slighty ahead of two 64 cores(total of 128 cores)

You would more typically represent that as 2x64.

 

[Saylick]

Whats wrong with the Graph? It show a single Bergamo with total of 128 cores slighty ahead of two 64 cores(total of 128 cores)

It's what @Exist50 said.

WCCFTech are notorious for getting the finer details wrong, often times doing so in the process of regurgitating the primary source.

 

[coercitiv]

Whats wrong with the Graph? It show a single Bergamo with total of 128 cores slighty ahead of two 64 cores(total of 128 cores)

There are other ways to arrange the information, some arguably easier to parse:

• 2x EPYC 9554 128-core (Genoa)
• 2x EPYC 9554 (Genoa), 128 cores
• 2x EPYC 9554 (Genoa, 64 core)
• 2x 64-core EPYC 9554 (Genoa)

 

[Abwx]

Comparison of RMB and Phoenix iGPUs at computerbase :

iGPUs von AMD im Vergleich: Radeon 780M vs. Z1 Extreme (ROG Ally) vs. Radeon 680M

Wie schnell sind die iGPUs des Z1 Extreme im Asus ROG Ally, die Radeon 780M sowie die Radeon 680M im Vergleich?

www.computerbase.de

 

[jpiniero]

AMD Ryzen 5 7500F to launch this week, the first AM5 CPU without integrated RDNA2 graphics - VideoCardz.com

AMD Ryzen 5 7500F has no integrated graphics AMD set to launch a new 6-core CPU this week, claims Korean retailer. AMD Ryzen 5 7500F CPU, Source: 신성조 It appears AMD will soon release a new 6-core processor that may lack one of the most common features of AM5 CPUs. The Ryzen 5 7500F shall […]

videocardz.com

Apparently AMD is going to release the 7500F soon. Guess that's where they will dump the IO dies with busted IGP. Doesn't sound like it will be much cheaper than the 7600.

 

[Asterox]

I dont remember any earlier rumors, that vapor chamber IHS was a option for Zen 4.Very good video, apparently AMD can hide every test design detail if it so chooses.

 

[jpiniero]

AMD 6-core Ryzen 5 7500F trades places with 7600X in first Geekbench test - VideoCardz.com

AMD Ryzen 5 7500F CPU without integrated graphics tested with first benchmark The company is set to launch a new Ryzen 7000 CPU that will lack integrated GPU support. The AMD Ryzen 5 7500F stands out as the first AM5 processor without integrated graphics support. This suggests the presence of a...

videocardz.com

7500F confirmed to be Raphael based and also has performance of roughly the 7600X. Not sure if it's worth giving up the IGP for 10 bucks but seems like you aren't losing much by going with it versus the 7600X.

 

[SteinFG]

AMD 6-core Ryzen 5 7500F trades places with 7600X in first Geekbench test - VideoCardz.com

AMD Ryzen 5 7500F CPU without integrated graphics tested with first benchmark The company is set to launch a new Ryzen 7000 CPU that will lack integrated GPU support. The AMD Ryzen 5 7500F stands out as the first AM5 processor without integrated graphics support. This suggests the presence of a...

videocardz.com

7500F confirmed to be Raphael based and also has performance of roughly the 7600X. Not sure if it's worth giving up the IGP for 10 bucks but seems like you aren't losing much by going with it versus the 7600X.

Don't believe it's better than 7600 or 7600X, this may be just an OCed chip

 

[adroc_thurston]

https://www.amd.com/en/newsroom/press-releases/2023-7-11-amd-epyc-cpus-now-power-sap-applications-hosted-on.html

and the last bastion falls.
itsover.png

What are even the leftover onprem dinosaur workloads left that are still not on EPYC?

 

[Ajay]

https://www.amd.com/en/newsroom/press-releases/2023-7-11-amd-epyc-cpus-now-power-sap-applications-hosted-on.html

and the last bastion falls.
itsover.png

What are even the leftover onprem dinosaur workloads left that are still not on EPYC?

Some older z-series IBM mainframe stuff. Oracle running on Power(x) at SomeBigCompany with SomeStupidJavaClient.

 

[adroc_thurston]

Some older z-series IBM mainframe stuff. Oracle running on Power(x) at SomeBigCompany with SomeStupidJavaClient.

Not even Intel managed to dislodge those and they were trying real hard with Xeon E7.
But there's always a chance kek.

 

[biostud]

and having 50% more cores (16 vs 12). The more you know ig.

That is only 33% more cores