Search found 274 matches

by Sesse
Sun Dec 13, 2020 11:30 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Position Causes Stockfish and Komodo To Crash
Replies: 135
Views: 14293

Re: Position Causes Stockfish and Komodo To Crash

gbtami wrote:
Sun Dec 13, 2020 9:56 am
Sure. You can substitute current maintainers :wink:
Is there active resistance, or just “nobody has contributed a reasonable patch”?
by Sesse
Sun Dec 13, 2020 12:49 am
Forum: Computer Chess Club: General Topics
Topic: 5950X boost clocks
Replies: 27
Views: 3343

Re: 5950X boost clocks

To reiterate: I am not overclocking anything in normal operation. It is completely normal for a Zen 3 CPU (in fact, most modern CPUs from any manufacturer) to boost the clock of a specific one core if it has work to do and the others are idle. The degree to which it can reach these boost clocks is p...
by Sesse
Sat Dec 12, 2020 11:36 pm
Forum: Computer Chess Club: General Topics
Topic: 5950X boost clocks
Replies: 27
Views: 3343

Re: 5950X boost clocks

If you use Ryzen Master to get OC-ed CPU what you experienced these are natural phenomenons. I don't intend to overclock, no. I am not interested in OC-ed only one core. Well, I am. :-) I frequently do single-threaded work, and it's fine if that's 25% faster. "Zero time" is a poetic magnifying. Sur...
by Sesse
Sat Dec 12, 2020 6:25 pm
Forum: Computer Chess Club: General Topics
Topic: 5950X boost clocks
Replies: 27
Views: 3343

Re: 5950X boost clocks

Mm, OK. I haven't tried locking all-core, but I could give it a shot. I'm running at DDR4-3600 (XMP profile).
by Sesse
Sat Dec 12, 2020 6:02 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Can EGTB storage requirements be reduced using this scheme?
Replies: 29
Views: 4095

Re: Can EGTB storage requirements be reduced using this scheme?

BTW, Fishtest has revealed that the Syzygy 6-men (no data for 7-men) can help SF to gain about 13 Elo, a good amount but not really vital IMO. You may remove the EGTB then gain back about 5-10 Elo by improving the endgame code for SF. I think this is a case where it's easy to be a bit “Elo blind”. ...
by Sesse
Sat Dec 12, 2020 5:10 pm
Forum: Computer Chess Club: General Topics
Topic: 5950X boost clocks
Replies: 27
Views: 3343

Re: 5950X boost clocks

That's interesting. I'm getting ~62 Mnps on that benchmark, which is markedly worse than the ~72 others are getting, but I'm unsure if the other 5950X are overclocked? The slowest one is listed as “5950X @ 3.95 GHz”, and I'm unsure if that means the base clock has been raised from 3.4 GHz to 3.95 GH...
by Sesse
Sat Dec 12, 2020 4:18 pm
Forum: Computer Chess Club: General Topics
Topic: 5950X boost clocks
Replies: 27
Views: 3343

5950X boost clocks

Hi, I received my 5950X yesterday—primarily for compiling stuff at work (I work from home, like so many others during Covid-19), but also to evaluate Zen 3 for a possible Threadripper upgrade for the next iteration of analysis.sesse.net. I'm not getting the boost clocks I had hoped for, so I'd like ...
by Sesse
Mon Oct 19, 2020 11:04 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: BONA_PIECE_ZERO
Replies: 18
Views: 3817

Re: BONA_PIECE_ZERO

This is the BONA_PIECE_ZERO mystery. The +1 is a small waste of memory. It doesn't really hurt. Also, having a prime index makes it much harder to get into cache associativity issues! :-) (I doubt this was ever an intended reason, and it probably doesn't matter in this case, but it can sometimes be...
by Sesse
Tue Sep 08, 2020 10:16 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Neural network quantization
Replies: 11
Views: 2339

Re: Neural network quantization

The most common way of doing NN weight quantization is AFAIK bog-simple linear quantization. Since the weights are for the linear step, it's fine to have a 1e-5 weight go to 0, because it's effectively 0 anyway compared to that +64 or -57 weight. You store the scale factors per layer (sometimes also...
by Sesse
Sat Aug 08, 2020 10:47 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Idea #13 [experiment]
Replies: 14
Views: 3820

Re: Idea #13 [experiment]

For all the pieces that _didn't_ move, give a small positive bonus for the square they're on? Should fix the root pawn issue.