NN-testrun of Lc0 0.26.3 PhoenixStein 1.2 net finished.
https://www.sp-cc.de
(Perhaps you have to clear your browsercache or reload the website)
SPCC: Testrun of Lc0 PhoenixStein 1.2 finished
Moderators: hgm, Rebel, chrisw
-
pohl4711
- Posts: 2435
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
-
jjoshua2
- Posts: 99
- Joined: Sat Mar 10, 2018 6:16 am
Re: SPCC: Testrun of Lc0 PhoenixStein 1.2 finished
Intersesting. I see you underclock your GPU more now that lc0 26.3 is faster. Did you test that it was faster with the size of LS or PS? I think the speedup mostly comes from nets with lots of filters, especially the biggest 384 filters. Probably speedup is half for the small nets, although the next release that is tested at TCEC should help the small nets a lot more than the big ones...
-
pohl4711
- Posts: 2435
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl
Re: SPCC: Testrun of Lc0 PhoenixStein 1.2 finished
I measure the LeelaRatio always like it should be done, as it was introduced: Lc0 with an old T10 net (20x256). And Stockfish 8 for the measuring of Stockfish.jjoshua2 wrote: ↑Sat Jan 30, 2021 12:47 am Intersesting. I see you underclock your GPU more now that lc0 26.3 is faster. Did you test that it was faster with the size of LS or PS? I think the speedup mostly comes from nets with lots of filters, especially the biggest 384 filters. Probably speedup is half for the small nets, although the next release that is tested at TCEC should help the small nets a lot more than the big ones...