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!

 

[JustViewing]

Not sure I follow. I thought 2 E cores ~= one P cores, yes ? (in performance, not power)

Are you saying one P-core ~= 4 E cores in power ?

Area is more like 4 P-core ~= 12 E-core. For AMD, zen 3 core is about double the size of E-Core, but also double the performance. So there won't be area efficiency for AMD to implement E-Cores

 

[tamz_msc]

7-Zip HWBot doesn't look that impressive against a PBO-enabled 5950X on AIO cooling. Only about 11% faster. 108 degrees is spicy.

5950X:

7950X:

 

[yuri69]

Sandra fits Zen 4 very well. 7zip or CPU-Z do not.

 

[igor_kavinski]

I trust Winrar more than 7-zip in benchmarks. 7-zip is free so less incentive to optimize it better.

 

[Carfax83]

If you are buying a 13900k/7950X you better have some work that requires heavily threaded performance.

I'm not sure about this. Despite the amount of cores in each CPU, they are still mainstream parts. If I do upgrade this year, I'll probably get either a 13900K or 7900x, and the most multithreaded workloads I ever do are gaming and encoding.

Professional rendering or encoding though typically use Xeon, Threadrippers or Epyc CPUs.

If you only have have work that is lightly threaded, how is it going to perform better on a hybrid design than a regular design, as it is only using a few cores?

I'm talking about performance per watt. The 12900K has better performance per watt than the 5950x in gaming, browsing and other lightly threaded applications as I recall.

 

[Carfax83]

I feel like I heard that in some movie

It's what all the young kids are saying these days

 

[tamz_msc]

Sandra fits Zen 4 very well. 7zip or CPU-Z do not.

Which is strange because LZMA is branch-misprediction heavy and the much larger uOP-cache in Zen 4 is supposed to help in such workloads. Could it be that we are seeing diminishing returns to increasing the uOP-cache size?

 

[igor_kavinski]

Which is strange because LZMA is branch-misprediction heavy and the much larger uOP-cache in Zen 4 is supposed to help in such workloads. Could it be that we are seeing diminishing returns to increasing the uOP-cache size?

Better to test that theory with something else, like LZMA Utils (tukaani.org). Could be that 7-zip code has some room for improvement left in it.

 

[deasd]

7z in hwbot is not a good benchmark, 12900k with LN2 @6.7Ghz could only catch up with PBO 5950x which score around 180000mips:

unityofsaints`s 7-Zip score: 181809 MIPS with a Core i9 12900K

The Core i9 12900K @ 6700MHzscores getScoreFormatted in the 7-Zip benchmark. unityofsaintsranks #null worldwide and #3 in the hardware class. Find out more at HWBOT.

hwbot.org

and some 6C result below show it is a mess, I even couldn't tell what 7z is testing, absolutely no affect by ST/MT performance......







OTOH the Cinebench result across different version R15 R20 R23, are better than I expected, R23 ~44% faster than a PBO 5950x! but I would rather see stock result cuz the temperature reach limit.

 

[tamz_msc]

7z in hwbot is not a good benchmark, 12900k with LN2 @6.7Ghz could only catch up with PBO 5950x which score around 180000mips

That's just Intel being traditionally much slower in 7-zip. Especially in decompression.

 

[DisEnchantment]

7z in hwbot is not a good benchmark, 12900k with LN2 @6.7Ghz could only catch up with PBO 5950x which score around 180000mips:

I believe it is because Zen architecture is exceptional in decompression performance. If you go all the way to Zen 2 you can see the advantage it had in 7z is very large.

 

[Just Benching]

When NDA Lifts and Reviews come in the entire Ryzen 7000 Line Up Will dominate current Intel Skus. No Questions about it.

What do you meant current? Alderlake? Yeah sure, with the exception of the 7600x getting pummeled by the 12600k, and probably the 7700x getting its ass handed to it by the 12700k, you are right. But they don't compete with alderlake, thry compete with Raptor

 

[Just Benching]

Pretty crazy perf./watt difference there, around 53% better for 7950X.

Uhm, efficiency comparisons should be done at iso wattage. Anything else is flawed.

 

[biostud]

Uhm, efficiency comparisons should be done at iso wattage. Anything else is flawed.

Not necessarily, as long as you specify the test conditions it is fine by me.

You can test at:
max performance no power limit
iso wattage
optimal performance/watt

All three will give different results and are to me valid test scenarios, but I agree that the first scenario does not make sense in most cases.

 

[DrMrLordX]

Also, we'll see if you're still singing the same song when Intel's mainstream CPUs eventually have 32 efficiency cores with the IPC of a Zen 3 core or better.

It is a very, very long wait 'til Arrow Lake. And that assumes no delays (see: Meteor Lake). Add a few more 'very's if Intel stumbles again. Raptor Lake has to anchor desktop for Intel for around two years if current schedules are to be believed. The fact that Raphael - and not even Raphael-X - seems to be positioned to dominate Raptor Lake in a large number of benchmarks is not a good look for Intel.

 

[PJVol]

iso wattage

What is that "iso wattage" ?

 

[Hans de Vries]

Ryzen 9 7950X: All Core 6.5 GHz:

https://twitter.com/x/status/1572861275834503168

 

[Hans de Vries]

Ryzen 9 7950X: Single Core 7.25 GHz:

https://twitter.com/x/status/1572858941490331648

 

[Atari2600]

Oh my....

Netburst has arrived!!

 

[PJVol]

I'm curious, is "RPL" officially used anywhere to refer to Intel 13-series cpus?
Saw this abbreviation posted before by the same person (APISAK) regarding raptorlake (just bored, sorry ) )

 

[DrMrLordX]

Netburst has arrived!!

Pentzenium 4?

 

[leoneazzurro]

View attachment 67947
I'm curious, is "RPL" officially used anywhere to refer to Intel 13-series cpus?
Saw this abbreviation posted before by the same person (APISAK) regarding raptorlake (just bored, sorry ) )

RaPtor Lake may indeed be shrinked to RPL, but so can be RaPhaeL. The other data in the screenshot are relative to Zen4.

 

[moinmoin]

Ryzen 9 7950X: All Core 6.5 GHz:

https://twitter.com/x/status/1572861275834503168

View attachment 67944

Ryzen 9 7950X: Single Core 7.25 GHz:

https://twitter.com/x/status/1572858941490331648

If true these would indicate that AMD finally managed to demolish the hard frequency wall as well as the v/f hockey stick curve of previous gens/nodes and support other reports that Zen 4 is essentially solely limited by temperature. Still personally wouldn't believe those numbers for now, feels way too high to be true.

 

[biostud]

What is that "iso wattage" ?

Same wattage

 

[nicalandia]

If true these would indicate that AMD finally managed to demolish the hard frequency wall as well as the v/f hockey stick curve of previous gens/nodes and support other reports that Zen 4 is essentially solely limited by temperature. Still personally wouldn't believe those numbers for now, feels way too high to be true.

APIKSA Has been delivering the Goods lately so I trust the source.