Search found 404 matches

by brianr
Wed Jul 01, 2020 2:53 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: knodes/s halved after improving eval
Replies: 6
Views: 421

Re: knodes/s halved after improving eval

Yes, thanks for pointing this out. Perhaps I've been spending too much time with Leela where nets of the same size can use fixed nodes. Of course, if the new eval terms might take longer to score than the knowledge is worth, then timed games should be used. For A/B engines with IID surprisingly fast...
by brianr
Tue Jun 30, 2020 11:10 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: knodes/s halved after improving eval
Replies: 6
Views: 421

Re: knodes/s halved after improving eval

Yes, it is expected that doing more in the eval will slow things down quite a lot. What counts is how well it plays. Mobility with computing moves and looking at squares attacked is quite expensive. There were some posts at one point about the relative value of various mobility terms. The baseline (...
by brianr
Sun Jun 28, 2020 10:11 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Testing Search and Evaluation
Replies: 6
Views: 579

Re: Testing Search and Evaluation

The "Texel type" tuning gained about 100-150 Elo for tinker, IIRC. While that was great, it was a one time improvement. Tinker's eval went from very poor to mediocre. After that, smaller changes exceeded my patience to measure given my hardware at the time. If your eval is already pretty good, impro...
by brianr
Thu Jun 11, 2020 10:20 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Perft speed and depth questions
Replies: 43
Views: 4374

Re: Perft speed and depth questions

While correctly working code is most important, 14M nps seems a little on the slow side. Tinker is far from the fastest at perft, but it is about 95M nps on a 3.6GHz CPU. Move generation in Tinker is done with magic bitboards, BTW. 95M nps without bulk counting is very fast. 14 Mnps looks like a go...
by brianr
Thu Jun 11, 2020 6:19 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Perft speed and depth questions
Replies: 43
Views: 4374

Re: Perft speed and depth questions

While correctly working code is most important, 14M nps seems a little on the slow side.
Tinker is far from the fastest at perft, but it is about 95M nps on a 3.6GHz CPU.
Move generation in Tinker is done with magic bitboards, BTW.
by brianr
Wed Jun 10, 2020 11:36 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Lc0 + CUDA 11 --- 50% loss of speed?
Replies: 3
Views: 827

Re: Lc0 + CUDA 11 --- 50% loss of speed?

This was mentioned on the Leela Discord also.
It was pointed out that this is a "release candidate" level release, so more to be done.
by brianr
Wed Jun 03, 2020 1:41 am
Forum: Computer Chess Club: General Topics
Topic: Rise of the mighty NN
Replies: 45
Views: 3573

Re: Rise of the mighty NN

There is no question that significant work was being done with NN engines for Go before Giraffe. Nonetheless, the crossover from Go back to Chess using both NNs and the "zero" approach re-energized interest in computer chess. While Alpha Zero for chess would have happened anyway, I for one am gratef...
by brianr
Wed Jun 03, 2020 12:42 am
Forum: Computer Chess Club: General Topics
Topic: Rise of the mighty NN
Replies: 45
Views: 3573

Re: Rise of the mighty NN

The overall impact of Giraffe should not be overlooked. Firstly, it demonstrated that NN engines could work reasonably well, even without powerful GPUs. This in turn was exciting enough to rekindle interest in computer chess, which had become a rather torpid field by 2015. This resurgent interest in...
by brianr
Tue Jun 02, 2020 12:51 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: How to work with batch size in neural network
Replies: 6
Views: 637

Re: How to work with batch size in neural network

Perhaps I am not understanding the issue, but generally the batch size used for training a net has nothing to do with the batch size used when playing a game (inference only). Moreover, the batch size has little to do with the number of features, assuming features means things addressed by a traditi...
by brianr
Sat May 23, 2020 1:31 am
Forum: Computer Chess Club: General Topics
Topic: Leela - Dance of the knights
Replies: 16
Views: 1124

Re: Leela - Dance of the knights

The speed difference will also depend on the net.
3010 is a very large net, so it might be that 1 CPU is enough to keep the GPU busy.
The difference may well become larger with smaller net sizes.

In any case, nice to see Leela being used and tested more and more.