Calculate Elo increased based on having 8, 16 or 32 cores ?
Moderators: hgm, Rebel, chrisw
-
- Posts: 5582
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Calculate Elo increased based on having 8, 16 or 32 cores ?
Using the same GHz speed can somebody calculate how much Elo is gained by Stockfish or Komodo by doubling the cores lets say from 4 to 8 or from 8 to 16 cores. ?
Do NOT worry and be happy, we all live a short life
-
- Posts: 5582
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Re: How much rating Komodo Dragon2 gain by doubling # of cores ?
I thought that Larry Kaufman already figured the rating increased of Komodo based on doubling the number of coresChessqueen wrote: ↑Wed Aug 25, 2021 11:07 pm Using the same GHz speed can somebody calculate how much Elo is gained by Stockfish or Komodo by doubling the cores lets say from 4 to 8 or from 8 to 16 cores. ?
Do NOT worry and be happy, we all live a short life
-
- Posts: 710
- Joined: Sat Dec 06, 2014 1:53 pm
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Calculate Elo increased based on having 8, 16 or 32 cores ?
Those results were pre-NNUE, so may be outdated now. Most likely doubling of threads will show smaller elo gains now with NNUE, since the level on one thread is much higher to start with. Also those results were at a very fast tc, not representative of play at time controls actually used by anyone, so probably too high for that reason as well. But they do show that benefit drops sharply beyond 16 threads.
Komodo rules!
-
- Posts: 5582
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Re: Calculate Elo increased based on having 8, 16 or 32 cores ?
That is what I thought that after 16 threads the benefit or Elo gain drops significantly.lkaufman wrote: ↑Thu Sep 02, 2021 12:09 amThose results were pre-NNUE, so may be outdated now. Most likely doubling of threads will show smaller elo gains now with NNUE, since the level on one thread is much higher to start with. Also those results were at a very fast TC, not representative of play at time controls actually used by anyone, so probably too high for that reason as well. But they do show that benefit drops sharply beyond 16 threads.
Do NOT worry and be happy, we all live a short life