Modern Times wrote: ↑Fri Mar 12, 2021 3:15 am
I don't see anywhere in CuteChess GUI where you turn off castling, but this is an engine feature surely - if the engine wants to castle, then it will, nothing you can do to stop it. How can a GUI tell an engine not to castle ? You still use an opening book, or not ? Books may contain castling moves. I still see no point in this.
Here, I use PGN to not castling.
[pgn]1. Nf3 Nf6 2. Nc3 Nc6 3. Rg1 Rg8 4. Rb1 Rb8 5. Rh1 Rh8 6. Ra1 Ra8 7. Ng1
Ng8 8. Nb1 Nb8 *[/pgn]
They still will be playing under standard (Not by turning castling off, since Cutechess does NOT support it yet. The trick is on the fen for Normal standard chess that allows castling the fen looks like this
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
AND For Non Castling you simply remove the KQkq in between w and -01 like this
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w - - 0 1
So under Cutechess when you go to Tools ==>Settings ==> Games under the opening suite insert this Fen in the rectangle FEN string or you can create a directory with this FEN string rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w - - 0 1 and then you can browse it an select it when you want to set a tournament of Non castling games
Do NOT worry and be happy, we all live a short life
Modern Times wrote: ↑Thu Mar 11, 2021 2:00 am
... because testing GPU engines is a total pain in the arse. No concurrency, even if you had a machine with dozens of cores you can run only one match.
Hi,
Why is that? Do the GPU engines use all the available cores? I am thinking about getting a new PC with a GPU, that's what I am interested. I've seen in TCEC, Lc0 uses only 2 threads.
Modern Times wrote: ↑Thu Mar 11, 2021 2:00 am
... because testing GPU engines is a total pain in the arse. No concurrency, even if you had a machine with dozens of cores you can run only one match.
Hi,
Why is that? Do the GPU engines use all the available cores? I am thinking about getting a new PC with a GPU, that's what I am interested. I've seen in TCEC, Lc0 uses only 2 threads.
Regards,
Carlos
Those are cpu cores. The big nets hog lots of memory and resources on a gpu (or multiple gpus at tcec).
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".
Modern Times wrote: ↑Thu Mar 11, 2021 2:00 am
... because testing GPU engines is a total pain in the arse. No concurrency, even if you had a machine with dozens of cores you can run only one match.
Hi,
Why is that? Do the GPU engines use all the available cores? I am thinking about getting a new PC with a GPU, that's what I am interested. I've seen in TCEC, Lc0 uses only 2 threads.
Regards,
Carlos
Those are cpu cores. The big nets hog lots of memory and resources on a gpu (or multiple gpus at tcec).
So, you can't limit the amount of RAM and/or #cpus with a GPU engine then?
There is little need to limit CPU cores for GPU engines such as Lc0 and the like.
The general rule of thumb is threads should be the number of GPUs plus 1 only.
More threads may show faster nps, but playing strength does not improve and could get worse.
CPU RAM memory (nncache setting) is a manageable resource for GPU engines as large trees are kept in memory, but should only bothered with for very long searches. With Lc0 there is no hash, so that setting does not apply.
So, when doing timed games, one game per GPU can be run at a time, and no pondering unless 2 GPUs.
With fixed nodes per move, many games can be run at one time depending on the net size, GPU speed and VRAM capacity.
PS Just a reminder: SF-NNUE and related NNUE engines are CPU, not GPU.
brianr wrote: ↑Sat Mar 20, 2021 3:17 pm
So, when doing timed games, one game per GPU can be run at a time,
Yes that is what I was referring to.
Of course all the idle CPU cores can be used for something else, as log as they don't hit the GPU too much which would then affect the performance of the engine using the GPU. If that doesn't matter (e.g. you're just using the GPU engine for analysis purposes rather then engine matches where you want reliable results) then fine.
brianr wrote: ↑Sat Mar 20, 2021 3:17 pm
There is little need to limit CPU cores for GPU engines such as Lc0 and the like.
The general rule of thumb is threads should be the number of GPUs plus 1 only.
More threads may show faster nps, but playing strength does not improve and could get worse.
CPU RAM memory (nncache setting) is a manageable resource for GPU engines as large trees are kept in memory, but should only bothered with for very long searches. With Lc0 there is no hash, so that setting does not apply.
So, when doing timed games, one game per GPU can be run at a time, and no pondering unless 2 GPUs.
With fixed nodes per move, many games can be run at one time depending on the net size, GPU speed and VRAM capacity.
PS Just a reminder: SF-NNUE and related NNUE engines are CPU, not GPU.
In the case of single core tournament with ponder off, Lc0 should play one game at the time, other games could be played in parallel, like in a swiss tournament. For a round robin, I guess is adviceable to run a gauntlet for the GPU engine, while the other cores could be used to play the other games of the tournament.
