trivia: ELO boost from a better compiler

Discussion of chess software programming and technical issues.

Moderators: hgm, Rebel, chrisw

brtzsnr
Posts: 433
Joined: Fri Jan 16, 2015 4:02 pm

trivia: ELO boost from a better compiler

Post by brtzsnr »

I've been testing the latest Go compiler in a 6000 games match between glarus (upcoming) and geneva (current) for three compiler versions.

Code: Select all

Rank Name                          ELO     +/-   Games   Score   Draws
   1 glarus.tip                    102      12    2000     64%     39%
   2 glarus.1.6                     71      12    2000     60%     41%
   3 glarus.1.5                     60      12    2000     59%     41%
   4 geneva.tip                    -51      12    2000     43%     36%
   5 geneva.1.5                    -88      12    2000     38%     36%
   6 geneva.1.6                    -95      12    2000     37%     36%
Compiled with golang 1.6 & 1.5 zurichess is about the same strength, but with golang at tip (which has a modern SSA backend) zurichess is about 30 ELO stronger in self-play. Given that huge ELO gains from search & eval are rarer and rarer I welcome this boost :).
spambanane
Posts: 22
Joined: Sun Jun 17, 2012 9:45 am

Re: trivia: ELO boost from a better compiler

Post by spambanane »

i too have seen improvements over the last couple of months. a lot of work went into the compiler/runtime recently (e.g. ssa rewrite, as you noted.). this big jumps speedwise is an indication that there is still a lot of room for improvement. exciting times ahead for go developers!