TalkChess.com

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...

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...

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.

Or you can use Clang, which has 64-bit ASan (also on Windows).

Ideally, I guess just generate the good moves first, instead of generating and then sorting? E.g., if you sort good captures first, generate them before anything else. The canonical structure for this would be the binary heap, but it suffers from extreme branch misprediction. (A winner tree is bette...

I am still running Chess.js on the depth 6 results…

nrbbnk1r/pp2pppq/8/2pp3p/3P2P1/1N6/PPP1PP1P/1RBBNKQR w HBhb - 0 9. Chess.js perf depth 6 on my machine: 2526 seconds. Stockfish 11: Below 9 seconds! Clearly, there is some room for improvement here. =)

My Chess.js fork seemingly handles this situation correctly, although I have no idea why. =)

Late to the party, but I've verified most of these in my Chess960 Chess.js fork. (Raw node counts only. It takes forever to get to depth 7, so I'm not sure if I ran all of them that far.)

Dann Corbit wrote: ↑

Sun May 03, 2020 11:11 pm

https://www.trentonsystems.com/products/bam-board

12 memory sticks per CPU?
I never saw a configuration that was not a power of 2

Skylake-SP has six memory channels per CPU, so 12 memory sticks per CPU would be expected. (Not saying anything about the rest of the page.)

If you have debugging information, at least perf will de-inline it for you when showing the profile; pass the --inline flag. (It's not perfect, but it's a lot better than nothing.) This is assuming Rust's tool chain gives out full DWARF, which I'd assume it does.