The Last Survivor of AB engine, Stockfish! What position will Stockfish stand?

[Leo]

I guess SF will win,but I don't know the hardware

Dann, how do you feel about NN chess engines. If you are thinking about an Epyc machine it dawned on me that you haven't been caught up in the NN craze. Have you used LC0 ?

[Ozymandias]

I think my dream machine will be the Ryzen 3000 series Threadripper with 16 cores

Reality will probably be cheaper, Ryzen 9 is bringing 16 cores for about $500 to market.

[Leo]

I think my dream machine will be the Ryzen 3000 series Threadripper with 16 cores

Reality will probably be cheaper, Ryzen 9 is bringing 16 cores for about $500 to market.

Sounds good. So I wont need the thrreadripper.

[Dann Corbit]

I have lots of GPUs.
Mostly, I use them for weird things like folding proteins.
I have tried LC0.

My intention is to buy some top end GPUs and put them in the box also.
I am interested in GPU programming, and in the evolution of the GPU programming model.

[Leo]

I have lots of GPUs.
Mostly, I use them for weird things like folding proteins.
I have tried LC0.

My intention is to buy some top end GPUs and put them in the box also.
I am interested in GPU programming, and in the evolution of the GPU programming model.

OK. I see.

[Leo]

I think my dream machine will be the Ryzen 3000 series Threadripper with 16 cores

Reality will probably be cheaper, Ryzen 9 is bringing 16 cores for about $500 to market.

I am going to look into that for sure.

[Ozymandias]

I think my dream machine will be the Ryzen 3000 series Threadripper with 16 cores

Reality will probably be cheaper, Ryzen 9 is bringing 16 cores for about $500 to market.

I am going to look into that for sure.

For chess, the real pickle is to choose the GPU. RTX cards are the only way to go, but they're 12nm and consume a lot of power. Not knowing when nVidia may go for a node reduction makes things confusing.

[Leo]

I think my dream machine will be the Ryzen 3000 series Threadripper with 16 cores

Reality will probably be cheaper, Ryzen 9 is bringing 16 cores for about $500 to market.

I am going to look into that for sure.

For chess, the real pickle is to choose the GPU. RTX cards are the only way to go, but they're 12nm and consume a lot of power. Not knowing when nVidia may go for a node reduction makes things confusing.

I need the cheapest GPU. I am not into LCO yet. Maybe in the future.

[Nay Lin Tun]

Update, after 176+ games (out of 600)!
Leela still maintain No 1.

[jd1]

Update, after 176+ games (out of 600)!
Leela still maintain No 1.

To summarize (wins only):

Lc0 vs SF 7-5
Lc0 vs Leelenastein 8-1
Lc0 vs Antifish 17-2

SF vs Lc0 5-7
SF vs Leelenstien 12-2
SF vs Antifish 12-7

For me the interesting points are:
A) The inclusion of Antifish seems to help Lc0 significantly, which is not entirely surprising as it was made to beat SF and not care about its performance against Lc0 much. I wonder if it is possible to make an engine to target Lc0 and lose to SF. This is perhaps a good reason why multiple versions like this shouldn't be allowed in a more serious tournament, otherwise SF should be looking to make an anti-Lc0 version which can take points away from Lc0 while losing badly to SF itself.

B) Although Antifish is doing well against SF, it seems that regular Lc0 is just better, even at that. I've been out of things for a while, are there any tests with a larger number of games to see whether Antifish does better than Lc0 against SF, or at least, better than expected?