TalkChess.com

I think quality of hardware can be measured by time to be able to do a specific chess task with the best software that is possible to write.

Possible tasks:

1)calculating perft function(number of possible games of n plies from a given position).
2)finding the shortest mate.

I read that gpu is relatively good for 1 and the question if it is also relatively good for 2.

If gpu is also good for 2 then the conclusion is that lc0 use a better hardware relative to stockfish in games.

Uri Blass wrote: ↑Sun Mar 03, 2019 9:49 am I think quality of hardware can be measured by time to be able to do a specific chess task with the best software that is possible to write.

Possible tasks:

1)calculating perft function(number of possible games of n plies from a given position).
2)finding the shortest mate.

I read that gpu is relatively good for 1 and the question if it is also relatively good for 2.

If gpu is also good for 2 then the conclusion is that lc0 use a better hardware relative to stockfish in games.

There is no method to make comparison between chess power of CPU and chess power of GPU without subjectivizm.
Moreover a Leela fun is not interested in the objective comparison. The essence is the win of Leela -
even if it was without opening book, what is an evident incorrect benefit for Leela.

Or simply 3) perform a parallel alphabeta search on gpu,

Zeta v099k with 11 to max 35 Mnps on Nvidia RTX Titan:

https://github.com/smatovic/Zeta/blob/m ... esults.txt

Regardless of all modesty, i am not aware of any other chess engine that is able
to utilize a gpu in such a way, but it would be fun to see how a team of fishy
eggheads give it a try

--
Srdja

Okay, another try,

let us assume the best software is those which makes best use of the transistors
given on a specific chip.

So let us compare the transistor count and the clock frequency of TCEC hardware:

Nvidia RTX 2080 TI:
18.6 billion transistors clocked with max 1.545 GHz
Nvidia RTX 2080:
13.6 billion transistors clocked with max 1.710 GHz

2*Intel Xeon E5-2699v4:
2*7.2 billion transistors clocked with max 3.6 GHz


Even with the best black magic bit twiddling hacks you will be limited by
transistor count and frequency, now do the maths and tell me where the
hardware advantage is...

--
Srdja

Tell me more about Zeta GPU?

j.korhonen wrote: ↑Sun Mar 03, 2019 6:46 pm Tell me more about Zeta GPU?

In short, it is a port of Zeta Dva engine to OpenCL to run on a GPU.

https://www.chessprogramming.org/Zeta

https://zeta-chess.app26.de/

--
Srdja

corres wrote: ↑Sun Mar 03, 2019 10:12 am

Uri Blass wrote: ↑Sun Mar 03, 2019 9:49 am I think quality of hardware can be measured by time to be able to do a specific chess task with the best software that is possible to write.

Possible tasks:

1)calculating perft function(number of possible games of n plies from a given position).
2)finding the shortest mate.

I read that gpu is relatively good for 1 and the question if it is also relatively good for 2.

If gpu is also good for 2 then the conclusion is that lc0 use a better hardware relative to stockfish in games.

There is no method to make comparison between chess power of CPU and chess power of GPU without subjectivizm.
Moreover a Leela fun is not interested in the objective comparison. The essence is the win of Leela -
even if it was without opening book, what is an evident incorrect benefit for Leela.

You are a bean counter.

If all the hardware is available for any engine to use, and it's just the dev's decision to take advantage of it or not, then there's no such thing as "advantage". That's exactly the situation at TCEC.

corres wrote: ↑Sun Mar 03, 2019 10:12 am

Uri Blass wrote: ↑Sun Mar 03, 2019 9:49 am I think quality of hardware can be measured by time to be able to do a specific chess task with the best software that is possible to write.

Possible tasks:

1)calculating perft function(number of possible games of n plies from a given position).
2)finding the shortest mate.

I read that gpu is relatively good for 1 and the question if it is also relatively good for 2.

If gpu is also good for 2 then the conclusion is that lc0 use a better hardware relative to stockfish in games.

There is no method to make comparison between chess power of CPU and chess power of GPU without subjectivizm.
Moreover a Leela fun is not interested in the objective comparison. The essence is the win of Leela -
even if it was without opening book, what is an evident incorrect benefit for Leela.

+1

All I think we can do is let the Elo system do the heavy lifting. That mean on my system Stockfish on a 2950x is better then Lc0 4xxxx on 2080 ti with the latest nn. I see results on 4 and 8 core system were Lc0 is better.

This data will get to us to were we want to be.

I will be testing with 6 move book. I will not handicap CPU engines. To me Lc0 must win with my standard testing protacals.

And that means Lc0 will have the benefit of EGTB.

smatovic wrote: ↑Sun Mar 03, 2019 7:06 pm

j.korhonen wrote: ↑Sun Mar 03, 2019 6:46 pm Tell me more about Zeta GPU?

In short, it is a port of Zeta Dva engine to OpenCL to run on a GPU.

https://www.chessprogramming.org/Zeta

https://zeta-chess.app26.de/

--
Srdja

classic parallel Alpha-Beta on GPU and open source. WOW. Why you not in TCEC?