• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

64 core EPYC, first benchmark leaked

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Tuna-Fish

Golden Member
Mar 4, 2011
1,069
675
136
Where did you find that? That's ~30% faster per core per clock than what a dual socket EPYC 7601 system (with a base clock of 2.2) can do. Realistically, some of that is probably turbo over the 1.8GHz base so the IPC improvement won't be that drastic. Still, that's quite encouraging.
 

Abwx

Diamond Member
Apr 2, 2011
9,154
1,001
126
Where did you find that? That's ~30% faster per core per clock than what a dual socket EPYC 7601 system (with a base clock of 2.2) can do. Realistically, some of that is probably turbo over the 1.8GHz base so the IPC improvement won't be that drastic. Still, that's quite encouraging.
This amount to 100.5pts/core, a R7 1700 does 177pts/core at 3.2GHz, wich is 100pts at 1.8GHz, the delta in IPC could explain the fact that it would amount to 100% scaling with a Zen 1 uarch.
 

lightmanek

Senior member
Feb 19, 2017
289
528
136
While 1st pic of that 128C Epyc seems well done, 2nd pic give it away. Cinebech starts new rendering threads clockwise and there are disjointed tiles on the bottom right of that pic where there should be none.
I call it a fake ...
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,217
1,822
136
While 1st pic of that 128C Epyc seems well done, 2nd pic give it away. Cinebech starts new rendering threads clockwise and there are disjointed tiles on the bottom right of that pic where there should be none.
I call it a fake ...
whats wrong with the second one ? looks to me like its captured multiple screen refreshes thats why the center of the render is bright and the outside of it is dull.

edit: if you watch this slowly you can see that what i say above aligns with this render in terms of what it renders when:
R15 V4.5 Quad Xeon Platinum 8180
 
  • Like
Reactions: lightmanek

lightmanek

Senior member
Feb 19, 2017
289
528
136
whats wrong with the second one ? looks to me like its captured multiple screen refreshes thats why the center of the render is bright and the outside of it is dull.

edit: if you watch this slowly you can see that what i say above aligns with this render in terms of what it renders when:
R15 V4.5 Quad Xeon Platinum 8180
That would be an interesting theory catching two refreshes with a single pucture, but probability of it happening is low. Other than someone recorded video of that run, compressed it and then took pictures of said video played back.

I must say, numbers alighn with what Zen could and should do, just that I tend to be suspicious about these early leaks.
 

Abwx

Diamond Member
Apr 2, 2011
9,154
1,001
126
Is there 256 tiles in the second pic..?

If there s more then more than a frame has been captured, although that wouldnt validate the thing, if there s less, well, it should be fake.
 

Mopetar

Diamond Member
Jan 31, 2011
5,962
2,784
136
That would be an interesting theory catching two refreshes with a single pucture, but probability of it happening is low. Other than someone recorded video of that run, compressed it and then took pictures of said video played back.

I must say, numbers alighn with what Zen could and should do, just that I tend to be suspicious about these early leaks.
Even if the numbers are reasonable, I don't think that means it's real. A good fake should have an air of authenticity about it. If there's anything else that looks sketchy, don't let the numbers appearing to be reasonably cause you to allay your suspicion.
 

bobhumplick

Junior Member
Jun 29, 2018
8
1
41
you cant compare a 2990wx to an epyc because the epyc has twice the mem channels. the 2990wx is memory starved. you need to compare the 64 core epyc to the 32 core epyc or better yet a dual socket 32 core epyc. the 64 core will do a bit better thant he dual socket maybe because of the lower power per watt. but then again all that heat was spread over 2 sockets instead of one so maybe they will be about the same. but you also have to take into account the fact that the latency between the 2 sockets wont be present (or at least as bad) on the 64 core so yeah a bit faster than a dual socket 32 core epyc.

and single 32 core epyc at the same clocks as your threadripper at 4ghz will have quite a bit higher score than a 2990wx at 4ghz because of the memory channels
 

ASK THE COMMUNITY