Not logged inCSS-Forum
Forum CSS-Online Help Search Login
CSS-Shop Impressum Datenschutz
Up Topic Hauptforen / CSS-Forum / T1 distilled im Test
- By Lothar Jung Date 2023-05-28 11:48
Hier auf Discord:

**Match:** T1 distilled vs. older T60 320x24
**LC0 version:** v0.31.0-dev dag-update_9be9c1
**LC0 options:** Networks 606511 (320x24), t1-256x10-distilled-swa-2432500 (256x10), Backend=cuda-fp16, MinibatchSize=32 (320x24) / 20 (256x10), MoveOverheadMs=0, StrictTiming=true
**Hardware:** Ryzen 7 5800H (3.2-4.4GHz) + RTX 3070M
**Time control:** 30s/game+0.25s/move
**Book:** unbalanced 4-moves book (60-80cp)
**Tablebases:** 6-man
**Adjudication:** 6-man TBs + -resign movecount=3 score=300, -draw movenumber=1 movecount=4 score=5
**Software:** Cutechess-CLI, restart=on, timemargin=1000
**Speed:** 606511: 12348 nodes/move (median), t1-256x10: 15695 nodes/move (median)
**Comment:** Older T60 networks are compatible with not up-to-date backends such as DirectX or OpenCL. However, recent networks will probably be stronger even with a slower backend.
```
   # ENGINE                                     :  RATING  ERROR  CFS(%)   GAMES  DRAWS(%)
   1 lc0.net.t1-256x10-distilled-swa-2432500    :     0.0   ----   100.0     500      64.8
   2 lc0.net.606511                                         :   -73.4   17.0     ---     500      64.8

White advantage = 109.9 +/- 8.9
Draw rate (equal opponents) = 81.2 % +/- 2.4```

Könnte für Besitzer von schwächerer Hardware, RTX 2070 etc. interessant sein.
Up Topic Hauptforen / CSS-Forum / T1 distilled im Test

Powered by mwForum 2.29.3 © 1999-2014 Markus Wichitill