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!

 

[TESKATLIPOKA]

This thread is about Zen 4, please move your debate about storage to the correct section.

 

[igor_kavinski]

AMD EPYC 7773X "Milan-X" Performance Continues To Impress With Tremendous Opportunity For Large-Cache Server CPUs - Phoronix



If and when Intel matches this, AMD will have Genoa-X ready.

 

[DrMrLordX]

If and when Intel matches this, AMD will have Genoa-X ready.

Whenever that will be . . .

 

[soresu]

Whenever that will be . . .

.....only for Turin-X to roll up....

Intel: Why do I hear boss music? 😨

😂🤣

 

[eek2121]

Whenever that will be . . .

Dear Intel,

Do you need some aloe for that burn?

Thanks,
eek2121

 

[Saylick]

.....only for Turin-X to roll up....

Intel: Why do I hear boss music? 😨

😂🤣

 

[pakotlar]

Was talking to a friend about fall upgrades (theyre looking to build a desktop). Realized that even if he bought my 5950x, I wouldn’t be excited to spend the money on DDR5 + 7950x. 35% is underwhelming, especially given that it comes at 1.8x the PPT (rumored). Raptor Lake is even worse. Really wish/hope AMD gives us 32 or 24
cores on consumer motherboards. Oh, and they jacked up Threadripper prices by 1-2K For 32+ cores. Bleh. Corporate greed sucks.

 

[inf64]

Was talking to a friend about fall upgrades (theyre looking to build a desktop). Realized that even if he bought my 5950x, I wouldn’t be excited to spend the money on DDR5 + 7950x. 35% is underwhelming, especially given that it comes at 1.8x the PPT (rumored). Raptor Lake is even worse. Really wish/hope AMD gives us 32 or 24
cores on consumer motherboards. Oh, and they jacked up Threadripper prices by 1-2K For 32+ cores. Bleh. Corporate greed sucks.

How about you wait and see how Zen4 actually performs?

 

[pakotlar]

How about you wait and see how Zen4 actually performs?

Yeah, I will, and may still get the CPU, but it will cost a lot more than my current setup, and will not make a meaningful difference to my work, which is a bit sad. 5950x is an amazing chip; maybe there will be a 7990x or something that drives core count higher on AM5? Or maybe the 3D cache version will suffice. Here’s hoping

 

[Timmah!]

Was talking to a friend about fall upgrades (theyre looking to build a desktop). Realized that even if he bought my 5950x, I wouldn’t be excited to spend the money on DDR5 + 7950x. 35% is underwhelming, especially given that it comes at 1.8x the PPT (rumored). Raptor Lake is even worse. Really wish/hope AMD gives us 32 or 24
cores on consumer motherboards. Oh, and they jacked up Threadripper prices by 1-2K For 32+ cores. Bleh. Corporate greed sucks.

I feel your pain. I wanted 24C myself, since 16C does feel bit like a meh proposition, since i already own 14C processor. Especially, when i think how i 5 years ago got those 14 cores and replaced the 6-core CPU i owned before. Now that was exciting and tangible upgrade!

Anyway, regarding 7950x, in my case at least, the performance increase should be more like 2x, both ST and MT, so i am still considering it and probably will do it. Probably would not do it for just 35 percent though.

Definitely not keen on wasting 2500 on Threadripper.

 

[Makaveli]

Was talking to a friend about fall upgrades (theyre looking to build a desktop). Realized that even if he bought my 5950x, I wouldn’t be excited to spend the money on DDR5 + 7950x. 35% is underwhelming, especially given that it comes at 1.8x the PPT (rumored). Raptor Lake is even worse. Really wish/hope AMD gives us 32 or 24
cores on consumer motherboards. Oh, and they jacked up Threadripper prices by 1-2K For 32+ cores. Bleh. Corporate greed sucks.

What workloads do you have that require 24 or 32 cores?

 

[pakotlar]

What workloads do you have that require 24 or 32 cores?

Some MCMC optimization stuff (statistical model fitting).

 

[pakotlar]

I feel your pain. I wanted 24C myself, since 16C does feel bit like a meh proposition, since i already own 14C processor. Especially, when i think how i 5 years ago got those 14 cores and replaced the 6-core CPU i owned before. Now that was exciting and tangible upgrade!

Anyway, regarding 7950x, in my case at least, the performance increase should be more like 2x, both ST and MT, so i am still considering it and probably will do it. Probably would not do it for just 35 percent though.

Definitely not keen on wasting 2500 on Threadripper.

2x due to AVX512, or something very memory bandwidth bound for your work?

 

[igor_kavinski]

Some MCMC optimization stuff (statistical model fitting).

Not possible to do that with a Beowulf cluster?

 

[tomatosummit]

2x due to AVX512, or something very memory bandwidth bound for your work?

Because he's got an older 14core intel something.
Few people actually look to upgrade every season.

 

[pakotlar]

Not possible to do that with a Beowulf cluster?

All possible, but distributed compute carries a lot of complexity overhead, so we (humanity) should go back to thinking closely about dense servers (local first, distributed second); at least that's the direction I'm taking. Hopefully will be productive.

 

[MadRat]

Sounds like you would want AVX512 sitting off the convergence of memory bandwidth, and is not in any way cache dependent. Would that be at the Northbridge?

 

[carrotmania]

Was talking to a friend about fall upgrades (theyre looking to build a desktop). Realized that even if he bought my 5950x, I wouldn’t be excited to spend the money on DDR5 + 7950x. 35% is underwhelming, especially given that it comes at 1.8x the PPT (rumored). Raptor Lake is even worse. Really wish/hope AMD gives us 32 or 24
cores on consumer motherboards. Oh, and they jacked up Threadripper prices by 1-2K For 32+ cores. Bleh. Corporate greed sucks.

£10 says you'll have Raptor Lake in the first week, claiming it's amazing, possibly while all reviews show Zen4 matching it at lower TDP at least. We all understand the part you are playing in this game.

 

[pakotlar]

Sounds like you would want AVX512 sitting off the convergence of memory bandwidth, and is not in any way cache dependent. Would that be at the Northbridge?

Not sure if directed at me, but I would want density; maybe AVX512, not sure yet. If I could get access to 512 cores in a 2P, with a few TB/s of HBM3 (say 64GB and a 200GB/s pool of DDR5), I’d be happy to give up AVX, since I’m running GPUs anyway. The biggest advantages of local execution are 1/1000th latency for ram and disk, and vastly simpler fault tolerance. For latency sensitive algorithms, a dense server can outperform a much larger cluster. Even 1 thread can be better than a cluster, see Feank McSherry’s Scalability: At What Cost http://www.frankmcsherry.org/graph/scalability/cost/2015/01/15/COST.html

 

[pakotlar]

£10 says you'll have Raptor Lake in the first week, claiming it's amazing, possibly while all reviews show Zen4 matching it at lower TDP at least. We all understand the part you are playing in this game.

I have a 5950X, and there is 0 chance I go to Raptor Lake. 10% that I go to Sapphire Rapids X or Genoa. I have no want of E cores.

 

[Markfw]

I have a 5950X, and there is 0 chance I go to Raptor Lake. 10% that I go to Sapphire Rapids X or Genoa. I have no want of E cores.

Well, we agree on that. I hate e-cores. They are useless to me.

 

[biostud]

The consensus is that the higher frequency, is the main reason for expected higher power draw, but could it mainly be limited to situations where something like AVX512 is in use?

 

[leoneazzurro]

The consensus is that the higher frequency, is the main reason for expected higher power draw, but could it mainly be limited to situations where something like AVX512 is in use?

Well if a CPU can withstand a certain amount of power when using AVX512, it most probably can use the same power limit for all the other workloads. It is quite probable it is an AMD choice whether allowing it in all usage scenarios or not. To me, it seems that the 170W power limit will be used in all multithreading loads, as it is proably the most limiting factor in those cases.

 

[RTX2080]

What's funny is someone found a mistake in PPT slides that AMD officially described "Cinebench R23 1T 35% improvement" in the foot note:

https://twitter.com/x/status/1547434913862803461

,,,,,,,if it turns out to be real. or it would have lawsuit lol.

 

[tomatosummit]

What's funny is someone found a mistake in PPT slides that AMD officially described "Cinebench R23 1T 35% improvement" in the foot note:

https://twitter.com/x/status/1547434913862803461

,,,,,,,if it turns out to be real. or it would have lawsuit lol.

It doesn't state anything of the sort. It's still NT as 35% and just states "single thread performance data was gathered in R23-1T"
All it does is disassociate that reference from the slide and only semiaccurate will care.