There's a misunderstanding. He used dev version of Stockfish extracted from gplay version of Droidfish it's complex but ok. What i meant before is that Droidfish has also a dev version of itself, which extracts a Stockfish dev version of different size. Wow it was hard to write.
Anyway i checked both 121014 binaries few minutes ago on intel. They run at same 770 knps. I will compare on exynos too.
Komodo and SF: small android tourney
Moderators: hgm, Rebel, chrisw
-
- Posts: 196
- Joined: Wed Oct 29, 2014 12:42 am
Re: Komodo and SF: small android tourney
My blog for Android users: http://chesstroid.blogspot.com
-
- Posts: 10948
- Joined: Wed Jul 26, 2006 10:21 pm
- Full name: Kai Laskos
Re: Komodo and SF: small android tourney
One more test. At 5 sec/move on my Exynos 4 cores, I played a round-robin 150 games total:
If taking as a guide this table for doubling nodes:
http://www.talkchess.com/forum/viewtopic.php?t=48733
With 1 core SF having about 200 knodes/sec, with 5 sec/move, then:
Effective Speedup 1->2 cores
2^(90.5/93) = 1.96
Effective Speedup 2->4 cores
2^(61.1/79) = 1.71
Effective Speedup 1->4 cores
1.96*1.71 = 3.35
Very similar to PC i5 or i7 scaling of SF on 1, 2, 4 cores at 5 sec/move.
Code: Select all
Program Table Points ELO
1 Stockfish 121014 4 cores : * 29.5 35.0 64.5/100 70.9
2 Stockfish 121014 2 cores : 20.5 * 31.5 52.0/100 9.8
3 Stockfish 121014 1 core : 15.0 18.5 * 33.5/100 -80.7
If taking as a guide this table for doubling nodes:
http://www.talkchess.com/forum/viewtopic.php?t=48733
With 1 core SF having about 200 knodes/sec, with 5 sec/move, then:
Effective Speedup 1->2 cores
2^(90.5/93) = 1.96
Effective Speedup 2->4 cores
2^(61.1/79) = 1.71
Effective Speedup 1->4 cores
1.96*1.71 = 3.35
Very similar to PC i5 or i7 scaling of SF on 1, 2, 4 cores at 5 sec/move.