Not logged inCSS-Forum
Forum CSS-Online Help Search Login
CSS-Shop Impressum Datenschutz
Up Topic Hauptforen / CSS-Forum / BT4 Netztest
- - By Lothar Jung Date 2024-08-14 12:41 Upvotes 1
**Match:** BT4-100 vs BT4-1130                               
**LC0-version:** lc0-v0.31.0-dag+git.692b821b-master, cuda, default, mbs=80, (both)
**Hardware:** ASUS TUF RTX 4070 Ti + CPU AMD Ryzen 9 7900
**Software:** cutechess-cli
**Time control:** 2000 Nodes/move
**Book:** sufi_22 to 26, sequential, color reversed
**Tablebases:** Syzygy 5-men
**Adjudication:** -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
**Test Duration:**  ~3h 
**Comment:**  BT4 ->  https://storage.lczero.org/files/networks-contrib/big-transformers/.  
dag+git.692b821b-master ->   https://ci.appveyor.com/project/Ergodice/lc0/builds/50366238

```
# PLAYER   : RATING  ERROR  POINTS  PLAYED    W    L    D  D(%)  CFS(%)
1 BT4-1130 :   55.4   25.6   283.5     500  150   83  267    53     100
2 BT4-100  :    0.0   ----   216.5     500   83  150  267    53     ---

White advantage = 142.73 +/- 13.46
Draw rate (equal opponents) = 68.16 % +/- 2.38
```
Parent - - By Max Siegfried Date 2024-08-14 13:33
Das Ergebnis sieht sehr gut aus.
Worin unterscheiden sich die beiden BT4 Netze?
Gibt es Neuigkeiten zu BT5?
Parent - By Lothar Jung Date 2024-08-14 19:35 Upvotes 1
T4-1024x15x32h-swa-6315000.pb.gz                  09-Mar-2024 21:06           335885561
BT4-1130.pb.gz                                                 12-Aug-2024 05:16           382645316
Parent - By Jörg Oster Date 2024-08-14 16:13 Upvotes 1
Na ja, genau wie bei AB Engines sind solche fixed nodes Tests mit Vorsicht zu betrachten.
Just my 2 cents ...
- - By Lothar Jung Date 2024-08-16 17:51 Upvotes 1
**Match:** BT4-6147500 vs BT4-1740                               
**LC0-version:** dag-5350a2e0 (Sufi 26) , cuda, default, mbs=80, (both)
**Hardware:** ASUS TUF RTX 4070 Ti + CPU AMD Ryzen 9 7900
**Software:** cutechess-cli
**Time control:** 3000 Nodes/move
**Book:** sufi_24 to 26, sequential, color reversed
**Tablebases:** Syzygy 5-men
**Adjudication:** -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
**Test Duration:**  ~2h 
**Comment:**  

```
# PLAYER      : RATING  ERROR  POINTS  PLAYED    W    L    D  D(%)  CFS(%)
1 BT4-1740    :   32.2   32.5   162.0     300   77   53  170    57      97
2 BT4-6147500 :    0.0   ----   138.0     300   53   77  170    57     ---

White advantage = 130.45 +/- 16.57
Draw rate (equal opponents) = 70.03 % +/- 3.01
```
https://storage.lczero.org/files/networks-contrib/big-transformers/
Parent - - By Peter Martan Date 2024-08-16 18:10 Upvotes 1
Lothar Jung schrieb:

**Time control:** 3000 Nodes/move
...
# PLAYER      : RATING  ERROR  POINTS  PLAYED    W    L    D  D(%)  CFS(%)
1 BT4-1740    :   32.2   32.5   162.0     300   77   53  170    57      97
2 BT4-6147500 :    0.0   ----   138.0     300   53   77  170    57     ---


Abgesehen davon, dass es wieder fixed nodes sind, hätte man sich dann wenigstens etwas mehr als 300 Partien gönnen können, um ordentlich aus der error bar zu kommen.
Parent - By Lothar Jung Date 2024-08-16 20:17 Upvotes 1
Ist doch nur ein Hinweis auf die Spielstärkensteigerung der neueren BT4 Netze.
Up Topic Hauptforen / CSS-Forum / BT4 Netztest

Powered by mwForum 2.29.3 © 1999-2014 Markus Wichitill