Search found 2086 matches

by Michel
Tue Jan 07, 2020 11:23 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

It is only the fixed depth tests that I prefer v4, but there is no reason not to stop the search if one thread reaches the depth limit. It turns out Stockfish can be made stronger by simply repeating a single-threaded search at a given depth a number of times. In fact, it seems a single-threaded se...
by Michel
Mon Jan 06, 2020 7:45 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

Vondele did an interesting experiment : I implemented 5 versions (drafts: https://github.com/vondele/Stockfish/commits/threadDepth) v1: Master version: all threads search irrespective of Limits.depth, mainThread stops search when it reaches limits depth. v2: First thread to reach Limits.depth stops...
by Michel
Sun Jan 05, 2020 8:58 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

// Iterative deepening loop until requested to stop or the target depth is reached while ( (rootDepth += ONE_PLY) < DEPTH_MAX && !Threads.stop && !(Limits.depth && mainThread && rootDepth / ONE_PLY > Limits.depth)) Perhaps this is the culprit? If I read it correctly then the depth limit is only res...
by Michel
Sun Jan 05, 2020 12:01 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

I modified this logic to make it never exceed the fixed depth limit and ran 8 threads vs 1 thread again at fixed depth 10. The result was +42.5 elo after 5536 games. Thanks. Interesting! This points indeed to a bit of widening. But I would argue that the elo difference is quite small for 3 thread d...
by Michel
Sun Jan 05, 2020 11:52 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

// Iterative deepening loop until requested to stop or the target depth is reached while ( (rootDepth += ONE_PLY) < DEPTH_MAX && !Threads.stop && !(Limits.depth && mainThread && rootDepth / ONE_PLY > Limits.depth)) Perhaps this is the culprit? If I read it correctly then the depth limit is only res...
by Michel
Sun Jan 05, 2020 10:34 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

You ran fixed depth games. Stockfish with 8 threads on 1 core will take much more time than Stockfish with 8 cores to reach the given depth, but will have a rather similar search tree in the end. Hence, having similar results with 1 or 8 cores using fixed depth for 8 threads is expected, though som...
by Michel
Sun Jan 05, 2020 5:37 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

It looks "ridiclous theory number 1" could be the winner :) I redid the test at fixed depth=8 but this time constraining all 8 threads to 1 cores using "taskset -c 0 ./cutechess-cli", and i can see only 100% cpu usage when the previous test was 800% cpu usage. So both the single thread and multi-th...
by Michel
Sun Jan 05, 2020 12:07 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

The story for Scorpio is the same at depth=14 too Score of scorpio-sht-8-thread vs scorpio-sht-1-thread: 78 - 90 - 132 [0.480] 300 BUT for stockfish, I was shocked to find out some +170 elo at fixed depth=8. Then I decided to investigate further by doing fixed depth tests starting from depth=1 Dept...
by Michel
Sat Jan 04, 2020 10:02 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

Occam's razor says that we should first look for the simplest possible explanation. And the simplest possible explanation is that Lazy has almost zero overhead. The overhead is so small, that (combined with a few primitive tricks to avoid that threads do too much double work) it wins against (some?...
by Michel
Fri Jan 03, 2020 9:53 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: RMO - Randomized Move Order - yet another Lazy SMP derivate
Replies: 114
Views: 24358

Re: RMO - Randomized Move Order - yet another Lazy SMP derivate

No that is not true. Threads work at two depths: d and d+1 simultaneously. While this separation may result in reduced overhead, each group of threads working on the same depth sure going to result in a ton of overhead! When I said "overhead" I meant "administrative overhead" (like in YBW). Lazy tr...