Real or hoax? Threadripper 3990wx

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Dann Corbit
Posts: 9894
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Real or hoax? Threadripper 3990wx

Post by Dann Corbit » Mon Sep 09, 2019 11:26 pm

Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

MikeB
Posts: 3267
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Real or hoax? Threadripper 3990wx

Post by MikeB » Tue Sep 10, 2019 2:44 am

Don't know about that - but if you get two of these on one motherboard , you are looking at 240,000,000 nps (SF) at a minimum.

https://browser.geekbench.com/v4/cpu/14554041

Dann Corbit
Posts: 9894
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Real or hoax? Threadripper 3990wx

Post by Dann Corbit » Tue Sep 10, 2019 6:43 pm

I think that number is high.
I guess it would be 30% more due to instruction efficiency than two AMD 7601 32 core CPUs, which get 127,342,246 NPS, which is: 165,544,920 NPS
And then the loss for CPU speed of 2.7 GHz moving down to to 2.0 GHz gives 122,625,867 NPS
Pretty nice score, but I would be very surprised at 240,000,000 NPS. Now, with 2 of the 7742 chips, 241,185,105 was obtained.
https://openbenchmarking.org/result/190 ... XEPYC77437

I am considering the 7702p, but I am waiting to see the cost and performance of the threadripper first.
If it is only 32 cores, then I will be disappointed. That is why I was excited about the original link. 64 cores is listed as max core count.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

MikeB
Posts: 3267
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Real or hoax? Threadripper 3990wx

Post by MikeB » Wed Sep 11, 2019 3:04 am

Dann Corbit wrote:
Tue Sep 10, 2019 6:43 pm
I think that number is high.
I guess it would be 30% more due to instruction efficiency than two AMD 7601 32 core CPUs, which get 127,342,246 NPS, which is: 165,544,920 NPS
And then the loss for CPU speed of 2.7 GHz moving down to to 2.0 GHz gives 122,625,867 NPS
Pretty nice score, but I would be very surprised at 240,000,000 NPS. Now, with 2 of the 7742 chips, 241,185,105 was obtained.
https://openbenchmarking.org/result/190 ... XEPYC77437

I am considering the 7702p, but I am waiting to see the cost and performance of the threadripper first.
If it is only 32 cores, then I will be disappointed. That is why I was excited about the original link. 64 cores is listed as max core count.
I'm not sure about that , but you may be correct.

7702 with 64 cores https://browser.geekbench.com/v4/cpu/14490973

7742 with 128 cores https://browser.geekbench.com/v4/cpu/14486381

the most highly correlated benchmark for Stockfish nps is the Integer Score ( although may not be as correlated for Intel vs AMD, but for the same processor brand, it is very correlated when the processors are very similar as they are here)

7742 Multi-Core Performance with 128 cores

Code: Select all

Multi-Core Score	191125	
Crypto Score	64768	
Integer Score	282056	
Floating Point Score	197032	
Memory Score	9261
7702P Multi-Core Performance with 64 cores

Code: Select all

Multi-Core Score	134681	
Crypto Score	62208	
Integer Score	198539	
Floating Point Score	136764	
Memory Score	5992
obviously ,there could be something wrong with the way the benchmark was run, but here , the same user ran both. I guess we'll have to wait and see. The price difference between the 7742 and the 7702P is not very much - about $500-$600 difference - one worked think there would not be a huge difference, but with the 7742 higher pricing , one would think that it would be a tad faster.
We'll find out soon enough.

Dann Corbit
Posts: 9894
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Real or hoax? Threadripper 3990wx

Post by Dann Corbit » Wed Sep 11, 2019 10:02 am

The 7742 is run in pairs for the 240M NPS bench.
Extremely doubtful that one 7702p can approach that.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

MikeB
Posts: 3267
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Real or hoax? Threadripper 3990wx

Post by MikeB » Wed Sep 11, 2019 9:50 pm

Dann Corbit wrote:
Wed Sep 11, 2019 10:02 am
The 7742 is run in pairs for the 240M NPS bench.
Extremely doubtful that one 7702p can approach that.
Agree 100% ,and I was not suggesting that in my original post.
MikeB wrote:
Tue Sep 10, 2019 2:44 am

Don't know about that - but if you get two of these on one motherboard , you are looking at 240,000,000 nps (SF) at a minimum.

https://browser.geekbench.com/v4/cpu/14554041

Cumnor
Posts: 58
Joined: Fri Jan 27, 2012 3:23 pm
Location: Cumnor, Oxford, UK
Contact:

Re: Real or hoax? Threadripper 3990wx

Post by Cumnor » Wed Sep 11, 2019 10:15 pm

Another source for Chess bench https://www.servethehome.com/wp-content ... chmark.jpg

AMD EPYC 7402P Review 24 Cores of Performance and Value https://www.servethehome.com/amd-epyc-7 ... d-value/2/
Moderator of Rybka forum
Admin of Infinitychess playing server

Dann Corbit
Posts: 9894
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Real or hoax? Threadripper 3990wx

Post by Dann Corbit » Wed Sep 11, 2019 10:24 pm

MikeB wrote:
Wed Sep 11, 2019 9:50 pm
Dann Corbit wrote:
Wed Sep 11, 2019 10:02 am
The 7742 is run in pairs for the 240M NPS bench.
Extremely doubtful that one 7702p can approach that.
Agree 100% ,and I was not suggesting that in my original post.
MikeB wrote:
Tue Sep 10, 2019 2:44 am

Don't know about that - but if you get two of these on one motherboard , you are looking at 240,000,000 nps (SF) at a minimum.

https://browser.geekbench.com/v4/cpu/14554041
I see.
However, the processors that end with a P designation cannot be run in pairs.
They require a single socket motherboard
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

MikeB
Posts: 3267
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Real or hoax? Threadripper 3990wx

Post by MikeB » Wed Sep 11, 2019 10:47 pm

Dann Corbit wrote:
Wed Sep 11, 2019 10:24 pm
MikeB wrote:
Wed Sep 11, 2019 9:50 pm
Dann Corbit wrote:
Wed Sep 11, 2019 10:02 am
The 7742 is run in pairs for the 240M NPS bench.
Extremely doubtful that one 7702p can approach that.
Agree 100% ,and I was not suggesting that in my original post.
MikeB wrote:
Tue Sep 10, 2019 2:44 am

Don't know about that - but if you get two of these on one motherboard , you are looking at 240,000,000 nps (SF) at a minimum.

https://browser.geekbench.com/v4/cpu/14554041
I see.
However, the processors that end with a P designation cannot be run in pairs.
They require a single socket motherboard
I knew they had single and dual socket configuration - I assumed that the "P" flavors would be run in pairs, my bad, you are correct.

MikeB
Posts: 3267
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Real or hoax? Threadripper 3990wx

Post by MikeB » Thu Sep 12, 2019 4:58 pm

MikeB wrote:
Wed Sep 11, 2019 10:47 pm
Dann Corbit wrote:
Wed Sep 11, 2019 10:24 pm
MikeB wrote:
Wed Sep 11, 2019 9:50 pm
Dann Corbit wrote:
Wed Sep 11, 2019 10:02 am
The 7742 is run in pairs for the 240M NPS bench.
Extremely doubtful that one 7702p can approach that.
Agree 100% ,and I was not suggesting that in my original post.
MikeB wrote:
Tue Sep 10, 2019 2:44 am

Don't know about that - but if you get two of these on one motherboard , you are looking at 240,000,000 nps (SF) at a minimum.

https://browser.geekbench.com/v4/cpu/14554041
I see.
However, the processors that end with a P designation cannot be run in pairs.
They require a single socket motherboard
I knew they had single and dual socket configuration - I assumed that the "P" flavors would be run in pairs, my bad, you are correct.
The pricing premium for the non p version over the p version is ~ 40% , factor of in the additional cost for the dual socket motherboard and you are looking at quite bit more than 2x premium to go from 64 to 128 cores. Currently pricing a 7702p configuration with a custom builder today. Would anyone like to guess the price - Ubunta OS , 1 TB SSD drive , assume a $750 video card and 64 GB ram. Guess the final price without delivery charges and as well the price you would be willing to pay if you had the financial resources to buy one without any undue hardship. I haven’t committed yet - still waiting for the quote. My best guess is that it will do the SF bench at 120M nps. Has anyone seen a bench from the 7702P yet?

Post Reply