Page 1 of 1
What if, SF and Leela were running on same hardware?
Posted: Fri Aug 16, 2019 9:54 am
by Nay Lin Tun
Now chess.com is doing 150/5 , 12 games match, SF vs Leela on CPU.
( Leela is 320x24, bigger network).
Speed ratio of nps = 1:100,000.
.
Results so far= 2 draws.
Re: What if, SF and Leela were running on same hardware?
Posted: Fri Aug 16, 2019 10:26 am
by crem
Nay Lin Tun wrote: ↑Fri Aug 16, 2019 9:54 am
Now chess.com is doing 150/5 , 12 games match, SF vs Leela on CPU.
( Leela is 320x24, bigger network).
Speed ratio of nps = 1:100,000.
.
Results so far= 2 draws.
nps ratio is ~120 000. Not sure whether comma is a decimal delimiter or thousand delimemter, but neither seem to be correct.
Re: What if, SF and Leela were running on same hardware?
Posted: Fri Aug 16, 2019 11:18 am
by jp
SF won the 3rd game with White (but they are now strangely calling that the 1st game). They're running this equal-hardware match "until the GPU is fixed" for CCC10 to resume.
SF's win was in the Berlin Defence, which has favored NN engines in the past. I can't see what book moves were used, but the game followed a Caruana-So game, which So won, until 10... f6.
Re: What if, SF and Leela were running on same hardware?
Posted: Fri Aug 16, 2019 11:27 am
by corres
crem wrote: ↑Fri Aug 16, 2019 10:26 am
Nay Lin Tun wrote: ↑Fri Aug 16, 2019 9:54 am
Now chess.com is doing 150/5 , 12 games match, SF vs Leela on CPU.
( Leela is 320x24, bigger network).
Speed ratio of nps = 1:100,000.
Results so far= 2 draws.
nps ratio is ~120 000. Not sure whether comma is a decimal delimiter or thousand delimemter, but neither seem to be correct.
A good book plus some search is against a fast calculator...