128 CPU Stockfish Cluster Comments.
Moderators: hgm, Rebel, chrisw
-
- Posts: 1080
- Joined: Fri Sep 16, 2016 6:55 pm
- Location: USA/Minnesota
- Full name: Leo Anger
128 CPU Stockfish Cluster Comments.
Sorry if I missed a thread on this. I was taken aback by the 4 billion nps of this machine. Its found here:http://www.ipmanchess.yolasite.com/amd- ... -bench.php Of course deep blue was at 200 million nps in the 1990s. What would this entity be capable of? I think chess engines have become better at using more cores in the last few years but still need improvement using all these cores. It seems like it could take the chess world where no man has gone before.
Advanced Micro Devices fan.
-
- Posts: 134
- Joined: Thu Jul 24, 2014 2:49 pm
Re: 128 CPU Stockfish Cluster Comments.
Here are some numbers for the cluster branch against a single 32 thread machine. Would love to see this monster in action
https://github.com/official-stockfish/S ... /pull/1931
https://github.com/official-stockfish/S ... /pull/1931
-
- Posts: 3707
- Joined: Thu Mar 16, 2006 7:10 pm
Re: 128 CPU Stockfish Cluster Comments.
With all these threads available maybe some of those cores would be better utilized for Monte Carlo search. You reach a point of diminishing returns and maybe those resources are better used via a different form factor.
-
- Posts: 30
- Joined: Mon Jan 14, 2019 7:55 am
- Full name: Henk Verbaasdonk
Re: 128 CPU Stockfish Cluster Comments.
If some billionaire was willing to waste all his money on dedicated Stockfish 10 ASICs.
Then the cluster could easily be blown away.
4 billion nps is still very impressive though.
Then the cluster could easily be blown away.
4 billion nps is still very impressive though.
-
- Posts: 1080
- Joined: Fri Sep 16, 2016 6:55 pm
- Location: USA/Minnesota
- Full name: Leo Anger
Re: 128 CPU Stockfish Cluster Comments.
In those tests the cluster did whomp on the 32 core SF. It looks like it maxed out at 130 Elo better.pferd wrote: ↑Sun Feb 03, 2019 8:07 pm Here are some numbers for the cluster branch against a single 32 thread machine. Would love to see this monster in action
https://github.com/official-stockfish/S ... /pull/1931
Advanced Micro Devices fan.
-
- Posts: 1080
- Joined: Fri Sep 16, 2016 6:55 pm
- Location: USA/Minnesota
- Full name: Leo Anger
Re: 128 CPU Stockfish Cluster Comments.
There are diminishing returns. Yes, I would like to see Monte Carlo tried out on it.
Advanced Micro Devices fan.
-
- Posts: 12541
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: 128 CPU Stockfish Cluster Comments.
That billionaire would learn the lesson of Hydra.
A dump truck full of money will make an engine that will win for 4 years.
Ten years later, a commodity PC will give it a spanking.
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.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 3291
- Joined: Wed Mar 08, 2006 8:15 pm
Re: 128 CPU Stockfish Cluster Comments.
This +130 gives nice 3700 ELO! Maybe the upper limit for chess?
Jouni
-
- Posts: 12541
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: 128 CPU Stockfish Cluster Comments.
I doubt it, but possibly an upper limit for shared hash table alpha-beta.
On the other hand, the evaluation of LC0 is clearly vastly superior to the evaluation of Stockfish (due to less than 0.1% as many nodes during search giving an answer that is about equal in quality).
So suppose that the evaluation of Stockfish were tuned to equal the evaluation of LC0.
Perhaps the limit is 5000 Elo.
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.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 1796
- Joined: Thu Sep 18, 2008 10:24 pm
Re: 128 CPU Stockfish Cluster Comments.
No, I doubt any entity will get over 4000 elo. I think we'll start to see the draw rate dramatically rise as this level of play is reached. But I'd love to be proved wrong.Dann Corbit wrote: ↑Tue Feb 05, 2019 10:47 pmI doubt it, but possibly an upper limit for shared hash table alpha-beta.
On the other hand, the evaluation of LC0 is clearly vastly superior to the evaluation of Stockfish (due to less than 0.1% as many nodes during search giving an answer that is about equal in quality).
So suppose that the evaluation of Stockfish were tuned to equal the evaluation of LC0.
Perhaps the limit is 5000 Elo.