Größer ist nicht immer besser:
**Match**: <@!351220241243832322> 's test 256x20-787, 384x30-386, 512x40-175 <
https://www.comp.nus.edu.sg/~sergio-v/test/> vs 256x20-T40-1541 (SV strongest 20b net)
**LC0 version:** 0.23.2
**LC0 options:** cudnn-fp16, *1 thread, cpuct=1.45, fpu=0.24, pst=1.85, cpuct-base=5000*
**Time control:** 1Kn/move
**Hardware:** RTX 2070
**Book:** SuperGM_4mvs_500 book, in sequence, reversed color
**Tablebases:** 6-man TB
**Adjudication**: 6-man TB, -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
**Speed:**: From pgn files
**Software:**cutechess-cli
**Comments:** The 20b net 256x20-787 is quite strong, only -82 elo less than 256x20-T40-1541. The big nets 512x40 is surprisingly weaker in fixed nodes (-185 elo): the weakest as well the slowest.
```diff
1 thread, vs 256x20-t40-1541 at 1 Knpm
cpuct=1.45, fpu=0.24, pst=1.85, cpuct-base=5000
# PLAYER : RATING ERROR POINTS PLAYED (%) CFS(%) W D L Nps
- 1 lc0.net.256x20-t40-1541 : 0 ---- 1921.0 3000 64.0 100 1079 1684 237 16.6
+ 2 lc0.net.384x30-386 : -46 13 435.0 1000 43.5 100 111 648 241 5.8
+ 3 lc0.net.256x20-787 : -82 13 385.0 1000 38.5 100 90 590 320 16.5
+ 4 lc0.net.512x40-175 : -185 16 259.0 1000 25.9 --- 36 446 518 3.0
```