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!

 

[inf64]

The 13900KF is not faster than the standard 13900K we have plenty of geekbench entries to support that. It's just a 13900K without iGPU.

They could just bin it better and claim the highest ST clock and performance (with TVB when premium cooling is used) . It's a legit marketing tactic and frankly they will need every little thing they could milk from 13th gen.

 

[arcsign]

Most people do not seem to realize just how energy efficient Alder Lake is, if - and only if - it is locked at 125w TDP or less. This is presumably the range it was designed for before some marketing people decided to push it beyond sanity in order to win some benchmarks against the 5950X.

In CB23 MT the comparison is as follows:

• The 12900K uses 4.469 Joules per run, while the 5950X uses 4.149 Joules and the 5900X uses 5.274 Joules. So the 12900K is almost as energy efficient as the 5950X.
• In terms of performance it is also not that bad - only about 10% slower than the 5950X. But Intel seemingly could not live with that.
• In ST Alder Lake is the clear winner in terms of both performance and efficiency - hands down. This is due to the large energy overhead of the IFOP interconnect of Zen.
• To be fair the 5900X and 5950X have a PPT of 143w. Locked at 125w they would be a bit more efficient in MT as well.

In terms of RTL vs. Zen4 IMHO this is going to be a close call. If both are locked at 125w then RTL might win everything - efficiency and performance, ST and MT.

i9-12900K @ 125w


R9 5900X @ Stock


R9 5950X @ Stock (only one run, not 10 minutes)


P.S.
If you want to read more about the measurements above and a lot more numbers, please have a look at this thread: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...y-of-x86-cpu-architectures.2597905/?view=date

Boris, are these test done using internal power measurement numbers or external/whole system?

Would love to see more of the latter; I don’t particularly trust manufacturers to report accurate/consistent powers numbers from onboard sensors… and moreover, from what I understand, what the reported numbers actually encompass is different between manufacturers.

Seems also more relevant day-to-day to see what system power consumption is; board/chipset power consumption is obviously going to vary… but of course you can’t run the chip without all the supporting hardware, so I feel like it should be included when assessing power draw. (though this suggests, also, that one might need to test several mobos per chipset to get a complete picture)

 

[BorisTheBlade82]

@arcsign
These numbers are obtained from the internal package power sensors. The initial goal was to compare CPUs as isolated as possible. When you start to compare the whole system there is a universe of things having an impact. Even for the same CPU SKU the mileage may vary by a a huge margin.
As you mentioned, there might be differences between vendors. CnC also reported that while Intel's numbers seem to be accurate measurements, AMDs numbers seem to be calculated - although both appear to be more accurate as for example temperature sensors, where they have been applying offsets for decades.

I tried to create a rather tight scenario in order to make results gathered by a community as comparable as possible. But they need to be taken as what they are: Numbers, that might give some clues, but are in no way scientific.

 

[Just Benching]

At same wattage? Or both @ default settings, because if I remember correctly you prefer to compare efficiency when the power limit is equal?

Indeed I think that the only way to properly test efficiency is at equal wattage, but people here for some reason keep posting efficiency graphs with one cpu running at 240w and the other one at 140 watts. Using that method, the T series of intel cpus will be the most efficient. Of course when that happens people will stop comparing efficiency out of the box, but resort to ISO wattage, mark my words

Running at same wattage I think that only the 7950x can actually compete in that department. Yes zen 4 made a great leap forward compared to zen 3, but alderlake was so far ahead that it's still going to be close at the high end segment. At the low end there is no comparison to be even made, the 13600k will just clean house against the 7600x. Remember, efficiency isn't just measure at 100% heavy MT workloads. There are tons of productivity applications that are more mixed workloads, or being cache and memory bandwidth limited so not all cores can be simultaneously pegged to 100%.

 

[Just Benching]

@desrever
@Markfw
In the thread I linked there are also stock numbers for the 12900K. I just wanted to point out that the ADL architecture and Intel7 are much better than the insane 241w limit Intel applied make it seem. From a technical PoV this is highly interesting to me.

Yeap, I think the 12900k had no business competing against the 5950x. Those Pcores have to be pushed to extreme wattage to make up for the lack of big cores and HT that makes the whole CPU seem inefficient.

It already was the fastest cpu - by far - in ST performance. Did it really need to be the fastest in MT as well?

 

[Hans Gruber]

Two things. AMD needs a Zen 4-- 24 core 48 thread CPU and a 32 core 64 thread CPU. I would suggest not buying Zen 4 until the 3D v-cache is released. The last part is something the Intel fanboys cannot beat in gaming.

 

[nicalandia]

Two things. AMD needs a Zen 4-- 24 core 48 thread CPU and a 32 core 64 thread CPU. I would suggest not buying Zen 4 until the 3D v-cache is released.

Why does it need those? Zen3 Threadripper PRO is enough for serious Workstation jobs.

Do you see any Alternative from Intel that will motivate AMD to release a such a CPU soon?

I mean yes, both the 13900K/KS and 7950X will make very potent Workstation CPUs. But they lack many of the features that a true Workstation Class CPU.

 

[nicalandia]

Latest Geekbench 5 Results for the 7950X

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

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

browser.geekbench.com

 

[nicalandia]

Is this MB for You?

https://twitter.com/x/status/1569844805299052544

 

[MangoX]

Not sure if this was mentioned, but maybe this is AMDs planned SKU eventually:

7600X
7700X
7800XT
7900X
7900XT
7950X
7990XTX

All XTx series are 3DV cache, pure gaming CPUs. Nothing Intel can really counter with.

What do ya'll think?

 

[Makaveli]

Looks like a list of GCN Ati Radeons

 

[nicalandia]

Not sure if this was mentioned, but maybe this is AMDs planned SKU eventually:

7600X
7700X
7800XT
7900X
7900XT
7950X
7990XTX

All XTx series are 3DV cache, pure gaming CPUs. Nothing Intel can really counter with.

What do ya'll think?

X3DX its their designation following the 5800X3D release

 

[Henry swagger]

When are review embargo's lifted ? For zen 4

 

[BorisTheBlade82]

When are review embargo's lifted ? For zen 4

AFAIR September the 20th.

 

[tamz_msc]

Latest Geekbench 5 Results for the 7950X

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

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

browser.geekbench.com

View attachment 67513

Still can't break 2000 int score it seems.

 

[Saylick]

AFAIR September the 20th.

Allegedly pushed back to the 26th... Launch still the 27th though. Seems there's some small tweaks and kinks to work out.

https://twitter.com/x/status/1569685791131172866

 

[Saylick]

Raichu posted some interesting IPC comparison tables. Thought I'd share. Basically, Zen 4 is very comparable to Golden/Raptor Cove, so if you get a 7950X you are essentially getting the equivalent to Intel offering a hypothetical SKU with 16 P cores.

https://twitter.com/x/status/1569904688933531649

 

[deasd]

Another AMD Marketing Devices leak from Userbbbbbenchmark...

https://twitter.com/x/status/1569871008227815424

 

[Hans Gruber]

Why does it need those? Zen3 Threadripper PRO is enough for serious Workstation jobs.

Do you see any Alternative from Intel that will motivate AMD to release a such a CPU soon?

I mean yes, both the 13900K/KS and 7950X will make very potent Workstation CPUs. But they lack many of the features that a true Workstation Class CPU.

Because Intel is marketing the e-cores as cores in their marketing. Even if they are half a core those e-cores are still bringing additional performance. When AMD was on GloFlo, they had 8 core CPU's. Then 7nm brought 16 core CPU's to the consumer line. 5nm should bring at least a 24 core if not a 32 core simply to counter Intel.

The other problem I see is the need for 3D V-cache. Another problem is the DDR5 sticks have very high latency. It will take some time for DDR5 to bring lower latency sticks to the market.

I will be watching from the sidelines for the next 6 months.

 

[inf64]

Raichu posted some interesting IPC comparison tables. Thought I'd share. Basically, Zen 4 is very comparable to Golden/Raptor Cove, so if you get a 7950X you are essentially getting the equivalent to Intel offering a hypothetical SKU with 16 P cores.

https://twitter.com/x/status/1569904688933531649

Excellent IPC numbers, basically confirms what AMD stated in their presentation.

Another AMD Marketing Devices leak from Userbbbbbenchmark...

https://twitter.com/x/status/1569871008227815424

View attachment 67518

Pretty brutal score, much better than the AMD Hype Train Zen4 which was a solid number one until now . I wonder when will the idiot who runs the website capitulate?

 

[igor_kavinski]

@arcsign
although both appear to be more accurate as for example temperature sensors, where they have been applying offsets for decades.

How much of an offset are we talking about? +/- 5 degrees?

 

[Timmah!]

Is this MB for You?

https://twitter.com/x/status/1569844805299052544

View attachment 67514

Nope. If the bottom PCIEX4 was PCIEX8 to allow 2 GPUs, then maybe.

 

[eek2121]

Is this MB for You?

https://twitter.com/x/status/1569844805299052544

View attachment 67514

Yep. That’s my job. I copy files between the drives inside my computer all day. That is all I do. All day. 80. hours. a week. Please send help.

(j/k of course)

 

[DrMrLordX]

Please send help.

We will send help in the form of PCIe 5.0 NVMe drives. Well not really, that would cost money . . .

 

[DisEnchantment]

Yep. That’s my job. I copy files between the drives inside my computer all day. That is all I do. All day. 80. hours. a week. Please send help.

That's the only job of a build server or CI/CD pipe. Read and write millions of files per day.