Check the clock every 1024 nodes #1482
Conversation
|
You have my vote :) |
|
Congratulations Stephane!!! |
|
The non regression SPRT[-3..1] at LTC passed: And I have also the results given in #1471 (speed-up and Elo gain against master), four tests called "Estimate the Elo cost of calling now() every N nodes", with N in {256,1024} and time control in {STC,LTC}. Tests were 40000 games with auto-purge off against master. http://tests.stockfishchess.org/tests/view/5aa4f37c0ebc59029780ff2d
|
This patch checks the clock every 1024, instead of 4096 in current master. This is a step towards a solution to alleviate the problem reported by Gian-Carlo Pascutto in pull request official-stockfish#1471, about a situation where SF would sometimes lose on time in endgames at time control 1min+1sec when using Syzygy EGTB on a spinning drive. We made four tests called "Estimate the Elo cost of calling now() every N nodes", with N in {256,1024} and time control in {STC,LTC}. Each test was 40000 games with auto-purge off, against master branch. http://tests.stockfishchess.org/tests/view/5aa4f37c0ebc59029780ff2d http://tests.stockfishchess.org/tests/view/5aa4f3080ebc59029780ff2b http://tests.stockfishchess.org/tests/view/5aa5cbb50ebc59029780ffce http://tests.stockfishchess.org/tests/view/5aa5cb730ebc59029780ffcc Here is a summary of the results (speed-up and Elo gain against master): | Time | N=256 | N=1024 | |:--------:|:-----------------:|:----------------:| | Speed-up | 0.1% slower | 0.2% faster | | STC | -0.33 +-2.2 Elo | 1.33 +-2.2 Elo | | LTC | 1.08 +-2.3 Elo | 2.34 +-2.2 Elo | Based on these results, we tested N=1024 as a non-regression against master at LTC: LLR: 2.96 (-2.94,2.94) [-3.00,1.00] Total: 80746 W: 12430 L: 12399 D: 55917 http://tests.stockfishchess.org/tests/view/5aa7ba660ebc590297810116 Closes official-stockfish#1482 No functional change.
Check the clock every 1024 nodes
This patch checks the clock every 1024, instead of 4096 in current master.
This is a step towards a solution to alleviate the problem reported by
Gian-Carlo Pascutto in pull request #1471, about a situation where SF
would sometimes lose on time in endgames at time control 1min+1sec when
using Syzygy EGTB on a spinning drive.
We made four tests called "Estimate the Elo cost of calling now() every
N nodes", with N in {256,1024} and time control in {STC,LTC}. Each test
was 40000 games with auto-purge off, against master branch.
http://tests.stockfishchess.org/tests/view/5aa4f37c0ebc59029780ff2d
http://tests.stockfishchess.org/tests/view/5aa4f3080ebc59029780ff2b
http://tests.stockfishchess.org/tests/view/5aa5cbb50ebc59029780ffce
http://tests.stockfishchess.org/tests/view/5aa5cb730ebc59029780ffcc
Here is a summary of the results (speed-up and Elo gain against master):
Based on these results, we tested N=1024 as a non-regression against
master at LTC:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 80746 W: 12430 L: 12399 D: 55917
http://tests.stockfishchess.org/tests/view/5aa7ba660ebc590297810116
Closes #1482
No functional change.