Hier der Test von 2 Netzen:
**Time control:** t1-512x15: 2.22s/game+0.037s/move, t1-256x10: 3.06s/game+0.051s/move
**Speed:** t1-512x15: 3195 nodes/move (median), t1-256x10: 13268 nodes/move (median)
**Comment:** Different time control for both networks to simulate BLAS DNNL backend test conditions (6 threads, TC: 12min/game+3s/move) where the speed ratio is different in favor of the smaller t1-256x10 network. At >3Knodes/move t1-512x15 is stronger on CPU (rapid TC on most systems).
```
# ENGINE : RATING ERROR POINTS GAMES DRAWS(%) CFS(%)
1 lc0.net.t1-512x15x8h-distilled-swa-3395000 : 0.0 ---- 761.0 1500 63.7 87.3
2 lc0.net.t1-256x10-distilled-swa-2432500 : -5.7 9.8 739.0 1500 63.7 ---
White advantage = 111.6 +/- 5.2
Draw rate (equal opponents) = 76.9 % +/- 1.6```
Hier der Link auf die Netze:
https://discord.com/channels/425419482568196106/430695662108278784/1124256643799076864