Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Multi-Processor Non-Determinism
- - By likesforests (**) Date 2008-08-09 23:19 Edited 2008-08-09 23:21
A few people have alluded to the fact that multiprocessor Rybka is non-deterministic. I tested a position with a deep tactical solution:

1q6/r4pbk/1r1p2pp/B2Pn3/Q2NP3/1p3P2/6PP/1R3RK1 b - - 0 32


* Rybka MaxCPUs=1 finds the solution at depth=9 in 3/3 runs as expected.
* Rybka MaxCPUs=2 finds the solution at depth=11, 13, 16, and I-gave-up-at-18 on its tests runs--mp was slower in terms of depth and time!

My questions are:

   (1) Even given non-determinism, is multiprocessor expected to be much slower in deep tactical positions?
   (2) If yes, does that mean when we want to search for tactics, it's a good idea to set MaxCPUs=1 to ensure we find them?
Parent - By Vasik Rajlich (Silver) Date 2008-08-10 07:15
1) No. There could possibly be some tiny differences, so that if for example Rybka's effective speedup going from 4 cores to 8 cores is 1.6, her time-to-tactical-solution speedup might be 1.58 or 1.62. It will be negligible.

2) Definitely not.

Vas
Up Topic Rybka Support & Discussion / Rybka Discussion / Multi-Processor Non-Determinism

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill