Search found 143 matches

by jkiliani
Tue Oct 02, 2018 10:12 am
Forum: Computer Chess Club: General Topics
Topic: Lc0 v18 rc2 for 2080 Syzygy implementation
Replies: 34
Views: 7252

Re: Lc0 v18 rc2 for 2080 Syzygy implementation

Thanks for finding a bug in the implementation. A proposed fix is available here: https://github.com/LeelaChessZero/lc0/pull/408 You can download a CUDA binary to try the fix here: https://ci.appveyor.com/api/buildjobs/g2ag25osq2sdbqti/artifacts/build%2Flc0.exe No, still misses wins as 3-fold rep. ...
by jkiliani
Mon Sep 24, 2018 12:41 pm
Forum: Computer Chess Club: General Topics
Topic: leela is official(?) better than sf9
Replies: 81
Views: 13943

Re: leela is official(?) better than sf9

AB engines are tested mainly for middle game. NN engines play learning games from start position so their knowledge decreases going further from start position. These facts cause that starting the games from a sort opening book or without opening book is favorable for NN engines and disadvantageous...
by jkiliani
Fri Sep 21, 2018 10:14 am
Forum: Computer Chess Club: General Topics
Topic: How good is the RTX 2080 Ti for Leela?
Replies: 48
Views: 12926

Re: How good is the RTX 2080 Ti for Leela?

How does that compare to a 1080 or 1080 ti? fp32 fp16 GTX 1080Ti: 8996 - Titan V: 13295 29379 RTX 2080: 9708 26678 RTX 2080Ti: 12208 32472 So an RTX 2080 will give around a factor 3 improvement to the GTX 1080 Ti when using fp16 which the GTX doesn't support. For the 2080Ti, the improvement is a fa...
by jkiliani
Fri Sep 21, 2018 8:36 am
Forum: Computer Chess Club: General Topics
Topic: How good is the RTX 2080 Ti for Leela?
Replies: 48
Views: 12926

Re: How good is the RTX 2080 Ti for Leela?

Presumably if we keep going down in precision there will be a penalty as the weights won't be as precise. Is the "backend=cudnn-fp16" command something which goes in the command line? What about for GUIs without one? The penalty for the case of going from fp32 to fp16 was found to be small enough t...
by jkiliani
Fri Sep 21, 2018 6:27 am
Forum: Computer Chess Club: General Topics
Topic: How good is the RTX 2080 Ti for Leela?
Replies: 48
Views: 12926

Re: How good is the RTX 2080 Ti for Leela?

Thank you ! Can you or someone else explain me what is fp32 and fp16 and how to set that? Is that also existing with other GPU such as the 1080Ti and or 1060 ? rgds Small update, Ankan added the 2080 to his benchmarks: with cudnn 7.3 and 411.63 driver available at nvidia.com minibatch-size=512, net...
by jkiliani
Thu Sep 20, 2018 8:10 pm
Forum: Computer Chess Club: General Topics
Topic: How good is the RTX 2080 Ti for Leela?
Replies: 48
Views: 12926

Re: How good is the RTX 2080 Ti for Leela?

Ankan posted Lc0 benchmarks for the RTX 2080 Ti on Leela Discord today, since nondisclosure clauses regarding benchmarks of those are no longer in force now that the hardware is released: with cudnn 7.3 and 411.63 driver available at nvidia.com minibatch-size=512, network id: 11250, go nodes 1000000...
by jkiliani
Mon Sep 17, 2018 2:33 pm
Forum: Computer Chess Club: General Topics
Topic: Chess.com 2018 computer chess championship
Replies: 279
Views: 56272

Re: Chess.com 2018 computer chess championship

They need to implement tablebase adjudication for this tournament to cut down o the number of times we end up watching engines messing about for 100+ moves of pointless crap. At least 6 man if not 7 man adjudication The problem is Lc0 is really bad at endgames. I know Alpha Zero adjudicated the gam...
by jkiliani
Sun Sep 09, 2018 1:08 pm
Forum: Computer Chess Club: General Topics
Topic: leela is official(?) better than sf9
Replies: 81
Views: 13943

Re: leela is official(?) better than sf9

In CCC1 Leela is drawing with many weak engines even with super hardware. Like draw vs Crafty with +8 evaluation. This confirms also AlphaZero was a hoax! Maybe after 2 years things change? This confirms absolutely nothing. AlphaZero most likely found solutions for the issues still affecting Leela,...
by jkiliani
Wed Aug 29, 2018 7:32 am
Forum: Computer Chess Club: General Topics
Topic: Komodo MCTS
Replies: 4
Views: 1906

Re: Komodo MCTS

Komodo MC was recently disqualified from CCCC for apparent stability issues, and Chess.com has an actually vested interest in seeing it succeed. Even while the CCCC tests were still running, it didn't reach anywhere close to the strength of standard Komodo. From this I consider it rather unlikely th...
by jkiliani
Sat Aug 25, 2018 10:30 pm
Forum: Computer Chess Club: General Topics
Topic: Something goes wrong with lc0 since yesterday?
Replies: 272
Views: 44041

Re: Something goes wrong with lc0 since yesterday?

I also think it's more plausible that we either have some sort of bug that hurts scaling, or simply haven't figured out some part of the AlphaZero approach properly yet. One experiment being tried now is changing the search parameters, i.e. PUCT, FPU reduction and futile move aversion constant. I'm...