Chess engine: Clover 3.1 NNUE


Clover
- UCI chess engine
Author: Luca Metehau Rating JCER=3146

3.1 
Clover 3.1 is now available and it's about 200 elo stronger than 3.0 in self-play. Around half of it comes from using simd code and the other one comes from search improvements and new net arch.

Some info about new net:
trained on 500 mil positions generated from 3.0 and some 3.1 development version
arch is: 768x512x1
training code can be found here: https://github.com/lucametehau/neuralnet
Note: the avx2 binary was heavily recommended for use, so I decided to attach it only. (Sorry if you don't have avx2 on your cpu)

Individual statistics: Clover 3.1 - 52 games CEDR (3'+3") +13,= 27,- 12, 51.0 %

Schooner 2.2                  :   2 (+  0,=  2,-  0), 50.0 %
Laser 1.7                     :   4 (+  1,=  3,-  0), 62.5 %
Fritz 17                      :   4 (+  1,=  2,-  1), 50.0 %
Xiphos 0.6.1                  :   4 (+  3,=  1,-  0), 87.5 %
Wasp 5.50                     :   4 (+  0,=  3,-  1), 37.5 %
Komodo 13.02                  :   4 (+  0,=  3,-  1), 37.5 %
Gogobello 3.0                 :   3 (+  1,=  2,-  0), 66.7 %
Toga IV                       :   3 (+  3,=  0,-  0), 100.0 %
Tucano 10.00                  :   3 (+  0,=  3,-  0), 50.0 %
RofChade 3.0                  :   3 (+  0,=  0,-  3),  0.0 %
Nemorino 6.00                 :   4 (+  0,=  2,-  2), 25.0 %
RubiChess 20220223            :   2 (+  0,=  0,-  2),  0.0 %
Booot 6.5                     :   4 (+  0,=  4,-  0), 50.0 %
Defenchess 2.3                :   4 (+  2,=  0,-  2), 50.0 %
Roc 2020Aug05                 :   4 (+  2,=  2,-  0), 75.0 %

Comments

Post a Comment

Popular posts from this blog

We already have the new Fritz 20 program - what does it look like and how does it work?

Stockfish 250702 wins Android Chess Engines Tournament, 2025.08.03-04