100 game tourney 3_1, 7 move HS book, 8 cores 4Ghz
SF163_SF16_3_1 2010
1 Stockfish 1.6.3 JA 64bit +21/=66/-13 54.00% 54.0/100
2 Stockfish 1.6 JA 64bit +13/=66/-21 46.00% 46.0/100
On 8 core system seems like almost +30 ELO
Stockfish 1.6 JA vs Stockfish 1.6.3 JA (8 core system)
Moderators: hgm, Rebel, chrisw
-
- Posts: 3707
- Joined: Thu Mar 16, 2006 7:10 pm
-
- Posts: 6442
- Joined: Tue Jan 09, 2007 12:31 am
- Location: PA USA
- Full name: Louis Zulli
Re: Stockfish 1.6 JA vs Stockfish 1.6.3 JA (8 core system)
This is not surprising to me. I assume you used default uci parameters, meaning 1.6 has Minimum Split Depth = 4 while 1.6.3 has Minimum Split Depth = 7. In my testing, the latter setting was much stronger than the former setting, at least in 1000 game matches at fast time controls.M ANSARI wrote:100 game tourney 3_1, 7 move HS book, 8 cores 4Ghz
SF163_SF16_3_1 2010
1 Stockfish 1.6.3 JA 64bit +21/=66/-13 54.00% 54.0/100
2 Stockfish 1.6 JA 64bit +13/=66/-21 46.00% 46.0/100
On 8 core system seems like almost +30 ELO
-
- Posts: 3707
- Joined: Thu Mar 16, 2006 7:10 pm
Re: Stockfish 1.6 JA vs Stockfish 1.6.3 JA (8 core system)
I still don't understand what split depth means. I do know that all 8 cores are at 100% now though, so it seems like the strength in ELO is coming from better usage of the 8 cores.
-
- Posts: 6442
- Joined: Tue Jan 09, 2007 12:31 am
- Location: PA USA
- Full name: Louis Zulli
Re: Stockfish 1.6 JA vs Stockfish 1.6.3 JA (8 core system)
In 1.6.2, I believe the Max Number of Threads was limited to 7, even if you had 8 cores. Now it is set to 8 if you have 8 cores available. That is probably the main change you are noticing.M ANSARI wrote:I still don't understand what split depth means. I do know that all 8 cores are at 100% now though, so it seems like the strength in ELO is coming from better usage of the 8 cores.