Search found 34 matches

by grahamj
Thu Jan 17, 2019 2:01 pm
Forum: Computer Chess Club: General Topics
Topic: Can somebody explain what makes Leela as strong as Komodo?
Replies: 50
Views: 6214

Re: Can somebody explain what makes Leela as strong as Komodo?

... But NN just seem to be an "natural" fit for gpus. Srdja Right and this is the essence. If NNs are such a natural fit for GPUs, why is NVIDIA's NN library so big? The Windows DLL is 280Mb, about 100x the size of the SF and LC0 binaries put together. It appears that NVIDIA have spent a huge amoun...
by grahamj
Thu Jan 17, 2019 9:18 am
Forum: Computer Chess Club: General Topics
Topic: Can somebody explain what makes Leela as strong as Komodo?
Replies: 50
Views: 6214

Re: Can somebody explain what makes Leela as strong as Komodo?

Like lkaufman, I think it's mainly a successful exploitation of hardware. I look at the AlphaZero paper in two quite different ways. As a scientific contribution, the important thing is the generality of the algorithm. The importance of the `zero' approach is not to avoid human bias, it is to demons...
by grahamj
Sun Jan 13, 2019 8:07 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Writing bugs
Replies: 66
Views: 9708

Re: Writing bugs

mar wrote:
Sat Jan 12, 2019 7:46 pm
You mean something like this? :)
:)
by grahamj
Sat Jan 12, 2019 4:59 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Writing bugs
Replies: 66
Views: 9708

Re: Writing bugs

This is no excuse for using a bad PRNG. It isn't "bad", it's at worst unsuited for certain applications. A very long time ago, probably last century, I wanted to make a random texture for artistic purposes. I knew that Microsoft's rand() was bad, but I thought it would be good enough just to make a...
by grahamj
Fri Jan 11, 2019 11:39 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Writing bugs
Replies: 66
Views: 9708

Re: Writing bugs

Yes, it's Python parallel/delayed code which then calls up (8) C-threads, each thread plays out 1024 games (not chess). At the beginning of each game the piece square table is varied a bit, each entry has a small-ish randint based on rand() added to it.. The speed up is near linear, multiple parall...
by grahamj
Sat Jan 05, 2019 8:45 am
Forum: Computer Chess Club: General Topics
Topic: LCZero FAQ is missing one important fact
Replies: 9
Views: 1532

Re: LCZero FAQ is missing one important fact

Why are just GPUs fast with NN engines? Please explain it to me! Before AlphaZero even came out including its paper, there was already a thread here by someone asking about harnessing GPU power because this expensive piece of hardware was just uselessly sitting down inside our machines during that ...
by grahamj
Tue Jan 01, 2019 2:15 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Generate EGTB with graphics cards?
Replies: 29
Views: 4587

Re: Generate EGTB with graphics cards?

I am learning how to program GPUs, in particular, NVIDIA GPUs. I think other GPUs are similar, but I'll stick to what I (half) know. On terminology, a single NVIDIA GPU has 100s or 1000s of 'CUDA cores'. 1. You will be launching LOTS of threads. It depends on the calculation, but to fully utilise th...
by grahamj
Fri Dec 21, 2018 9:19 pm
Forum: Computer Chess Club: General Topics
Topic: Komodo 12.3 is out
Replies: 76
Views: 11883

Re: Komodo 12.3 is out

I've read 5 or 6 online projects of putting MCTS on a GPU. I can't vouch for their success but people are trying. Here's one: https://pdfs.semanticscholar.org/fe90/c1f9955ba1f06f5ef26bde100bcc5c7a3327.pdf There is MCTS and MCTS. The so-called MCTS used by AlphaZero and LC0 is a deterministic algori...
by grahamj
Sun Nov 25, 2018 4:03 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: CCR board representation
Replies: 7
Views: 1261

Re: CCR board representation

The author claims the benefits over both array and bitboard (regardless of x32/x64 arch) representation assuming assembly language is used(but also high level languages with optimization flags are fine according to author). I very much doubt this can beat bitboards. Today almost everyone uses magic...
by grahamj
Sat Nov 17, 2018 1:00 pm
Forum: Computer Chess Club: General Topics
Topic: Houston: We have lift off ...
Replies: 55
Views: 7136

Re: Houston: We have lift off ...

250 Elo points below 10xxx run with large 20x256 net. But these 20xxx and 30xxx runs with large nets run for months with only 100 Elo points better results than small net. 20xxx is IIRC already abandoned after several LR droppings Not abandoned, but put on lower priority. 30xxx is of course no long...