[math-fun] Deep Mind crushes chess
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess "Stockfish is the reigning TCEC computer chess champion, and while it failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!"
Seems they limited stock fish to 1GB hash. That seems low. Wonder why they did this. On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess
"Stockfish is the reigning TCEC computer chess champion, and while it failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!" _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ --
where did you read that, tom? just curious On Wed, Dec 6, 2017 at 11:58 AM Tomas Rokicki <rokicki@gmail.com> wrote:
Seems they limited stock fish to 1GB hash. That seems low. Wonder why they did this.
On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess
"Stockfish is the reigning TCEC computer chess champion, and while it failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!" _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
Page 5 of the paper. I would expect for 64 threads you'd use a bit more hash memory. https://cdn.chess24.com/GzFl-Z4-SVWO-mC9rL6XhQ/original/mastering-chess-and-... On Wed, Dec 6, 2017 at 3:37 PM, Thane Plambeck <tplambeck@gmail.com> wrote:
where did you read that, tom? just curious On Wed, Dec 6, 2017 at 11:58 AM Tomas Rokicki <rokicki@gmail.com> wrote:
Seems they limited stock fish to 1GB hash. That seems low. Wonder why they did this.
On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess
"Stockfish is the reigning TCEC computer chess champion, and while it failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!" _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ --
Unless they used a 64-core processor, they were using network-connected machines, so it may be at least 1GB per machine. However, a Stockfish support page lists the optimal hash size as (transaction-per-second x seconds per move)/100 MB. The Deep Mind paper shows Stockfish running 70,000 transactions per second and each player given one minute per move. That results in a recommendation of a 42GB hash, or 656MB per thread. Running a separate hash per thread at 1GB per hash gives a total hash size of 64GB. Does anyone here know about configuring or tuning Stockfish in this manner? My quick searches yielded no additional information. On Wed, Dec 6, 2017 at 6:41 PM, Tomas Rokicki <rokicki@gmail.com> wrote:
Page 5 of the paper. I would expect for 64 threads you'd use a bit more hash memory.
https://cdn.chess24.com/GzFl-Z4-SVWO-mC9rL6XhQ/original/mastering-chess-and-...
On Wed, Dec 6, 2017 at 3:37 PM, Thane Plambeck <tplambeck@gmail.com> wrote:
where did you read that, tom? just curious On Wed, Dec 6, 2017 at 11:58 AM Tomas Rokicki <rokicki@gmail.com> wrote:
Seems they limited stock fish to 1GB hash. That seems low. Wonder why they did this.
On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess
"Stockfish is the reigning TCEC computer chess champion, and while it failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!" _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
I would expect them to be using a two socket, 16 physical core per socket, two thread per physical core box as this is a pretty standard server configuration. No network involved. On Thu, Dec 7, 2017 at 5:22 PM Jeff Caldwell <jeffrey.d.caldwell@gmail.com> wrote:
Unless they used a 64-core processor, they were using network-connected machines, so it may be at least 1GB per machine. However, a Stockfish support page lists the optimal hash size as (transaction-per-second x seconds per move)/100 MB. The Deep Mind paper shows Stockfish running 70,000 transactions per second and each player given one minute per move. That results in a recommendation of a 42GB hash, or 656MB per thread. Running a separate hash per thread at 1GB per hash gives a total hash size of 64GB. Does anyone here know about configuring or tuning Stockfish in this manner? My quick searches yielded no additional information.
On Wed, Dec 6, 2017 at 6:41 PM, Tomas Rokicki <rokicki@gmail.com> wrote:
Page 5 of the paper. I would expect for 64 threads you'd use a bit more hash memory.
https://cdn.chess24.com/GzFl-Z4-SVWO-mC9rL6XhQ/original/mastering-chess-and-...
On Wed, Dec 6, 2017 at 3:37 PM, Thane Plambeck <tplambeck@gmail.com>
wrote:
where did you read that, tom? just curious On Wed, Dec 6, 2017 at 11:58 AM Tomas Rokicki <rokicki@gmail.com>
wrote:
Seems they limited stock fish to 1GB hash. That seems low. Wonder why they did this.
On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess
"Stockfish is the reigning TCEC computer chess champion, and while
it
failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!" _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ --
Thanks, good to know. Is there any reason to think the 1GB hash would or would not be per thread, even without a network/ On Thu, Dec 7, 2017 at 8:25 PM, Tomas Rokicki <rokicki@gmail.com> wrote:
I would expect them to be using a two socket, 16 physical core per socket, two thread per physical core box as this is a pretty standard server configuration. No network involved.
On Thu, Dec 7, 2017 at 5:22 PM Jeff Caldwell <jeffrey.d.caldwell@gmail.com> wrote:
Unless they used a 64-core processor, they were using network-connected machines, so it may be at least 1GB per machine. However, a Stockfish support page lists the optimal hash size as (transaction-per-second x seconds per move)/100 MB. The Deep Mind paper shows Stockfish running 70,000 transactions per second and each player given one minute per move. That results in a recommendation of a 42GB hash, or 656MB per thread. Running a separate hash per thread at 1GB per hash gives a total hash size of 64GB. Does anyone here know about configuring or tuning Stockfish in this manner? My quick searches yielded no additional information.
On Wed, Dec 6, 2017 at 6:41 PM, Tomas Rokicki <rokicki@gmail.com> wrote:
Page 5 of the paper. I would expect for 64 threads you'd use a bit more hash memory.
https://cdn.chess24.com/GzFl-Z4-SVWO-mC9rL6XhQ/original/mastering-chess-and-...
On Wed, Dec 6, 2017 at 3:37 PM, Thane Plambeck <tplambeck@gmail.com>
wrote:
where did you read that, tom? just curious On Wed, Dec 6, 2017 at 11:58 AM Tomas Rokicki <rokicki@gmail.com>
wrote:
Seems they limited stock fish to 1GB hash. That seems low. Wonder why they did this.
On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess
"Stockfish is the reigning TCEC computer chess champion, and while
it
failed to make the final this year it went unbeaten in 51 games. In a match with the chess-trained AlphaZero, though, it lost 28 games and won none, with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. It turns out the starting move is really important after all!" _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
I believe the hash is shared and configured by total size. On Thu, Dec 7, 2017 at 5:29 PM Jeff Caldwell <jeffrey.d.caldwell@gmail.com> wrote:
Thanks, good to know.
Is there any reason to think the 1GB hash would or would not be per thread, even without a network/
On Thu, Dec 7, 2017 at 8:25 PM, Tomas Rokicki <rokicki@gmail.com> wrote:
I would expect them to be using a two socket, 16 physical core per socket, two thread per physical core box as this is a pretty standard server configuration. No network involved.
On Thu, Dec 7, 2017 at 5:22 PM Jeff Caldwell < jeffrey.d.caldwell@gmail.com> wrote:
Unless they used a 64-core processor, they were using network-connected machines, so it may be at least 1GB per machine. However, a Stockfish support page lists the optimal hash size as (transaction-per-second x seconds per move)/100 MB. The Deep Mind paper shows Stockfish running 70,000 transactions per second and each player given one minute per move. That results in a recommendation of a 42GB hash, or 656MB per thread. Running a separate hash per thread at 1GB per hash gives a total hash size of 64GB. Does anyone here know about configuring or tuning Stockfish in this manner? My quick searches yielded no additional information.
On Wed, Dec 6, 2017 at 6:41 PM, Tomas Rokicki <rokicki@gmail.com> wrote:
Page 5 of the paper. I would expect for 64 threads you'd use a bit more hash memory.
https://cdn.chess24.com/GzFl-Z4-SVWO-mC9rL6XhQ/original/mastering-chess-and-...
On Wed, Dec 6, 2017 at 3:37 PM, Thane Plambeck <tplambeck@gmail.com>
wrote:
where did you read that, tom? just curious On Wed, Dec 6, 2017 at 11:58 AM Tomas Rokicki <rokicki@gmail.com>
wrote:
Seems they limited stock fish to 1GB hash. That seems low. Wonder
why
they
did this.
On Wed, Dec 6, 2017 at 11:42 AM Hans Havermann <gladhobo@bell.net> wrote:
> https://chess24.com/en/read/news/deepmind-s-alphazero-crushes-chess > > "Stockfish is the reigning TCEC computer chess champion, and while it > failed to make the final this year it went unbeaten in 51 games. In a match > with the chess-trained AlphaZero, though, it lost 28 games and won none, > with the remaining 72 drawn. With White AlphaZero scored a phenomenal 25 > wins and 25 draws, while with Black it 'merely' scored 3 wins and 47 draws. > It turns out the starting move is really important after all!" > _______________________________________________ > math-fun mailing list > math-fun@mailman.xmission.com > https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun > -- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ -- _______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun
-- -- http://cube20.org/ -- http://golly.sf.net/ --
participants (4)
-
Hans Havermann -
Jeff Caldwell -
Thane Plambeck -
Tomas Rokicki