I made some tests with Sergio 512x40-50 net and the gotten results are disappointing:
512x40-50 net : J13B.2-188 net = 0 : 28 (12 draw) 40 games
512x40-50 net : Stockfish191002 = 1 : 22 (17 draw) 40 games
Max speed of nets:
J13B.2-188 net
max.nps = 28.0 Kn/s (Depth 19, Time 218.3 sec, Nodes 6.12 Mn)
512x40-50 net
max.nps = 7.56 Kn/s (Depth 14, Time 478 sec, Nodes 3.6 Mn)
I used my common hardware (2 x RTX 2060 OC, AMD 2950x 16x4.0 GHz with 6 men syzygy) and parameters (TC = 1min + 2sec/move with reversed colours, Leela on Multiplexing with 4 Threads, NNcacheSize = 10000000, PrefetchBatch = 64 and others are default) and my TestBook20 (20 opening positions).
Some short tests with Sergio 512x40 net
Moderators: hgm, Rebel, chrisw
-
corres
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
-
Dann Corbit
- Posts: 12541
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: Some short tests with Sergio 512x40 net
Try it with time control ten times slower
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
corres
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
Re: Some short tests with Sergio 512x40 net
If you have time for it, please try it.
But it is obvious to me it is effective if you enhance the move time of 512x40 net only what is not a correct thing or if you use the net with much more stronger GPU hardware against Stockfish like CCC.