Parallel search slowdown?

Discussion of chess software programming and technical issues.

Moderators: hgm, Rebel, chrisw

User avatar
Evert
Posts: 2929
Joined: Sat Jan 22, 2011 12:42 am
Location: NL

Re: Parallel search slowdown?

Post by Evert »

Ok, and two more bugs down:

The first one was a typo in the parallel search. Instead of searching with a window of (-1-alpha, -alpha) it was searching with (1-alpha, -alpha). Oops.

Fixing that the code crashed, which turned out to be due to me forgetting to copying a piece of global state.

With those fixed things now actually seem to work! Parallel speedup (measured as time-to-depth) seems to scale as 1.4^(cores/2) and in terms of playing strength the parallel version appears to not really be any stronger than the single core version (from looking at the games, far too few so far to make a quantitative assessment)..

So there's still clearly work to be done but things seem to be coming together nicely.