Is it useful to use more than one core for stockfish?

[Uri Blass]

https://ccrl.chessdom.com/ccrl/4040/rat ... t_all.html

Stockfish 15 64-bit 3534 +20 −20 72.4% −143.3 54.1% 772
Stockfish 15 64-bit 4CPU 3531 +23 −23 63.9% −80.2 71.7% 506

Note that the leader is
Stockfish 13 64-bit 4CPU 3537 +17 −17 75.7% −168.0 48.3% 1172

[Uri Blass]

Some nice results by Dragon3 4 cpu against Stockfish engines

15 − 15(+0−0=30) against Stockfish13 4 cpu
10 − 10(+0−0=20) against Stockfish15 4 cpu
25 − 25(+0−0=50) against Stockfish14.1 1 cpu


https://ccrl.chessdom.com/ccrl/4040/cgi ... 4-bit_4CPU

[Guenther]

https://ccrl.chessdom.com/ccrl/4040/rat ... t_all.html

Stockfish 15 64-bit 3534 +20 −20 72.4% −143.3 54.1% 772
Stockfish 15 64-bit 4CPU 3531 +23 −23 63.9% −80.2 71.7% 506

Note that the leader is
Stockfish 13 64-bit 4CPU 3537 +17 −17 75.7% −168.0 48.3% 1172

I see zero cross-testing between 4CPU and single core results, this way it is
useless to compare the ratings as they seem to be two completely different pools
anyway?
As long as 4CPU versions never (or nearly never? - may be done loooong ago in the past?)
play 1CPU versions a real 'rating' comparison is impossible between them.

[Graham Banks]

I see zero cross-testing between 4CPU and single core results, this way it is
useless to compare the ratings as they seem to be two completely different pools
anyway?
As long as 4CPU versions never (or nearly never? - may be done loooong ago in the past?)
play 1CPU versions a real 'rating' comparison is impossible between them.

They have and they will.

[Uri Blass]

https://ccrl.chessdom.com/ccrl/4040/rat ... t_all.html

Stockfish 15 64-bit 3534 +20 −20 72.4% −143.3 54.1% 772
Stockfish 15 64-bit 4CPU 3531 +23 −23 63.9% −80.2 71.7% 506

Note that the leader is
Stockfish 13 64-bit 4CPU 3537 +17 −17 75.7% −168.0 48.3% 1172

I see zero cross-testing between 4CPU and single core results, this way it is
useless to compare the ratings as they seem to be two completely different pools
anyway?
As long as 4CPU versions never (or nearly never? - may be done loooong ago in the past?)
play 1CPU versions a real 'rating' comparison is impossible between them.

Big majority of games of 4 CPU are against 4 CPU but there were 50 draws between Dragon3 4 cores and Stockfish14.1 single core.

Dragon3 is not something long ago in the past.

https://ccrl.chessdom.com/ccrl/4040/cgi ... 4-bit_4CPU

[Graham Banks]

https://ccrl.chessdom.com/ccrl/4040/rat ... t_all.html

Stockfish 15 64-bit 3534 +20 −20 72.4% −143.3 54.1% 772
Stockfish 15 64-bit 4CPU 3531 +23 −23 63.9% −80.2 71.7% 506

Note that the leader is
Stockfish 13 64-bit 4CPU 3537 +17 −17 75.7% −168.0 48.3% 1172

I see zero cross-testing between 4CPU and single core results, this way it is
useless to compare the ratings as they seem to be two completely different pools
anyway?
As long as 4CPU versions never (or nearly never? - may be done loooong ago in the past?)
play 1CPU versions a real 'rating' comparison is impossible between them.

Big majority of games of 4 CPU are against 4 CPU but there were 50 draws between Dragon3 4 cores and Stockfish14.1 single core.

Dragon3 is not something long ago in the past.

https://ccrl.chessdom.com/ccrl/4040/cgi ... 4-bit_4CPU

More 4CPU v 1CPU games are planned.
I'm to blame for not running as many as we should, as I keep forgetting to add a 1CPU opponent into my 4CPU gauntlets.

[Guenther]

I see zero cross-testing between 4CPU and single core results, this way it is
useless to compare the ratings as they seem to be two completely different pools
anyway?
As long as 4CPU versions never (or nearly never? - may be done loooong ago in the past?)
play 1CPU versions a real 'rating' comparison is impossible between them.

They have and they will.

Well, but you have to admit the linking between both pools is incredible tiny by now (and inexistent for newer versions).
I wonder if Ordo even could compute the full database for all players (no idea how Bayeselo addresses this) by now.

Anyhow SF15 4CPU has no 1CPU opponent and SF15 1CPU has no 4CPU opponent so far and you need to go back to SF13 4 CPU or so
to find a pairing with a single 1CPU opponent.
BTW you shouldn't see this as critiscm, but I just to want enlighten that you should not make conclusions between the
very loose connected pools of MP and SP versions.

Code: Select all

Legend: #1C-Opp means number of SP 1 core opponents
        #G games with SP 1 core opponents

Version       #1C-Opp  #G

SF15    4CPU  0        0  
SF14.1  4CPU  0        0
SF14    4CPU  0        0
SF13    4CPU  1      (48)
SF12    4CPU  0        0
SF11    4CPU  0        0
SF10    4CPU  1      (52)

[Modern Times]

I wonder if Ordo even could compute the full database for all players (no idea how Bayeselo addresses this) by now.

As recently as 1 or 2 weeks ago Ordo processes it perfectly as one connected group. But you're right, far too few recent 1CPU Vs 4CPU games.

[Modern Times]

Just ran Ordo on 7th May database:

Group connectivity: **PASSED**
All players are connected into only one group.

As Graham says though, he needs to pay more attention to 1CPU vs 4CPU.

[pedrox]

ryzen 7 5800H 8 cores / 16 threads
40 moves in 20 seconds

Score of stockfish_14.1_win_x64_avx2 (1 thread) vs stockfish_14.1_win_x64_avx2 (6 threads): 0 - 110 - 890 [0.445]
... stockfish_14.1_win_x64_avx2 (1) playing White: 0 - 14 - 486 [0.486] 500
... stockfish_14.1_win_x64_avx2 (1) playing Black: 0 - 96 - 404 [0.404] 500
... White vs Black: 96 - 14 - 890 [0.541] 1000
Elo difference: -38.4 +/- 6.8, LOS: 0.0 %, DrawRatio: 89.0 %
1000 of 1000 games finished.

Score of stockfish_14.1_win_x64_avx2 (6 threads) vs stockfish_14.1_win_x64_avx2 (12 threads): 16 - 23 - 961 [0.496]
... stockfish_14.1_win_x64_avx2 (6) playing White: 15 - 1 - 484 [0.514] 500
... stockfish_14.1_win_x64_avx2 (6) playing Black: 1 - 22 - 477 [0.479] 500
... White vs Black: 37 - 2 - 961 [0.517] 1000
Elo difference: -2.4 +/- 4.2, LOS: 13.1 %, DrawRatio: 96.1 %
1000 of 1000 games finished.