Page 3 of 3

Re: Lazy SMP and 44 cores

Posted: Sat Dec 01, 2018 1:36 pm
by syzygy
AndrewGrant wrote:
Wed Aug 08, 2018 5:23 am
I've tried doing the "pick best thread" type stuff that Stockfish does -- and it never works for me. Perhaps it just takes too many cores to test, or maybe I've done the logic wrong.
Preferring a thread that returns a higher score makes no sense to me. It is just a way of trying to fool yourself.

Preferring a higher depth does make sense to me, provided the main thread is not treated differently than other threads (whether it works or not is another question).

I doubt that letting threads start at different depths really makes a difference (perhaps at shorter TCs it reduces some contention?).

Re: Lazy SMP and 44 cores

Posted: Sat Dec 01, 2018 1:43 pm
by syzygy
sandermvdb wrote:
Wed Nov 21, 2018 1:20 pm
The only cause I can think of is my hashtable implementation or the way I spawn threads which is now about the same as other engines. Any ideas/suggestions?
Is there a problem?

If there is a problem, then make sure your hashtable functions correctly and your replacement scheme is not trying to do things differently than what other engines do. Lazy smp is fully dependent on the hashtable.