Search found 189 matches

by dragontamer5788
Tue Dec 10, 2019 4:46 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: PEXT/PDEP are even slower than you think on Zen
Replies: 12
Views: 2197

Re: PEXT/PDEP are even slower than you think on Zen

Pext/Pdep is not only very handy for sliding move generation, in the evaluation function you can also make good use of it, for instance to look at pawn configurations. Could you elaborate on that? Most of the pawn-configuration ideas (backwards pawn, isolated pawn, etc. etc.) seem like simple masks...
by dragontamer5788
Tue Dec 10, 2019 4:45 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Misprediction-poor looping
Replies: 9
Views: 1345

Re: Misprediction-poor looping

GCC still needs the #pragma omp simd to see that this loop is vectorizable for some reason why? the code you posted is auto-vectorized by gcc regardless of any pragma, clang even unrolls the loop as well as msc Interesting. I stand corrected. I tested it earlier, but maybe I herp-derp'd and messed ...
by dragontamer5788
Tue Dec 10, 2019 3:05 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Misprediction-poor looping
Replies: 9
Views: 1345

Re: Misprediction-poor looping

Yes, but only if you assume that Action() is trivial without side effects and the compiler can see its guts. Indeed, I used the function calls only as a means to write compact pseudo-code; they were supposed to be simple code sections. BTW, all calls to Action are done in the same order as they wou...
by dragontamer5788
Mon Dec 09, 2019 11:42 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: PEXT/PDEP are even slower than you think on Zen
Replies: 12
Views: 2197

Re: PEXT/PDEP are even slower than you think on Zen

Since I don't have access to a PEXT / PDEP machine, I figure I might as well publish my basic idea here, and maybe someone else will want to run with it. The basic idea is that sliding-piece attacks , that is rook, queen, and bishop, may be optimized even further than magic-bitboards, using PEXT / P...
by dragontamer5788
Mon Dec 09, 2019 10:25 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Misprediction-poor looping
Replies: 9
Views: 1345

Re: Misprediction-poor looping

Comment: For some reason, I couldn't grok your code in the first pass. My brain understands the following more: for(i=0; i<MAX; i++) { todo[n] = i; n += (Condition(i) ? 1 : 0); } Condition(i) ? 1 : 0 will probably compile into a cmov instruction, so you won't have any branches either. In fact, you p...
by dragontamer5788
Mon Dec 09, 2019 8:10 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: PEXT/PDEP are even slower than you think on Zen
Replies: 12
Views: 2197

Re: PEXT/PDEP are even slower than you think on Zen

Very unfortunate. PEXT / PDEP seem like they have a huge opportunity for sliding piece attacks in bitboards. Bishops, Queens, and Rook movement clearly can be calculated more efficiently with PEXT / PDEP. I have a Zen processor though, so I have to basically use other methodologies. Intel having a 1...
by dragontamer5788
Sun Dec 08, 2019 6:48 pm
Forum: Computer Chess Club: General Topics
Topic: Time per gigatype RAM with NN engines to exhaustion
Replies: 23
Views: 2537

Re: Time per gigatype RAM with NN engines to exhaustion

This sort of traversal would minimize the hits to the hard drive, while still providing a MCTS traversal of some kind. SSD Flash storage might be fast enough to provide the ~50,000 positions/second that LeelaZero can provide, but hard drives are just way too slow But I read that SSDs will be harmed...
by dragontamer5788
Fri Dec 06, 2019 9:26 pm
Forum: Computer Chess Club: General Topics
Topic: Time per gigatype RAM with NN engines to exhaustion
Replies: 23
Views: 2537

Re: Time per gigatype RAM with NN engines to exhaustion

LC0 is still WIP, make an proposal on GitHub or Discord. Afaik Percival (dragontamer) came up with this idea too. I believe it is possible to do a MCTS traversal with virtual-loss on the hard drive or SSDs, but I haven't fully discussed with anyone the full scope of my idea yet. I'm kind of busy wi...
by dragontamer5788
Thu Dec 05, 2019 2:05 am
Forum: Computer Chess Club: General Topics
Topic: AMD Ryzen R9 3950X new build
Replies: 55
Views: 6313

Re: AMD Ryzen Threadripper 3960X new build

As a side note: I purchased a 3960x this morning. The MB I'm looking at is the ASROCK TRX40 Creator. It has 3 NVMe slots and 4x PCIe Gen. 4 X16 slots. So it can support up to 4 GPU's of your choice! :D :D :shock: :shock: :D :D Regards, Zenmastur 4x physical x16 slots, but its a 16+8+16+8 configurat...
by dragontamer5788
Wed Dec 04, 2019 11:50 pm
Forum: Computer Chess Club: General Topics
Topic: Time per gigatype RAM with NN engines to exhaustion
Replies: 23
Views: 2537

Re: Time per gigatype RAM with NN engines to exhaustion

With an RTX 2080, how fast does it consume RAM? I have found that my 1080TI eats 32 GB in an hour, and the analysis slows to a crawl (and the machine starts to get balky). If you have 2x RTX2080, what is the maximum time you can analyze before RAM is exhausted with 64 GB and with 128 GB? Has anyone...