ppigazzini
ppigazzini
https://github.com/official-stockfish/fishtest/blob/12981ffad22f5e44420d40ea9298b6ff83c0bfc2/worker/games.py#L1157-L1180 if I recall correctly `fast-chess` does not need stochastic rounding.
`cutechess` accepts only integers, but it's a pleasure to start blaming SF devs for the convoluted SPSA parameters rounding.
No way a legal department takes `can be` as a specification.
The PR has two problems: 1. it will be a regression for workers tested faster with `bmi2` than with `avx512` 2. `vnni512` and `vnni256` architectures have the same conditions on...
PROD updated, thank you @FauziAkram
Thank you @LightningThunder1 :)
@LightningThunder1 let me know your real name for the AUTHORS file, otherwise I will add the nickname.
https://github.com/official-stockfish/fishtest/blob/b204b37ec99e049481a1f3e955aaff23364c1775/server/fishtest/rundb.py#L681-L683 https://github.com/official-stockfish/fishtest/blob/b204b37ec99e049481a1f3e955aaff23364c1775/server/fishtest/rundb.py#L723-L732 It should queue only 4 requests and drop the other until a slot in the semaphore is released...
https://github.com/official-stockfish/fishtest/blob/b204b37ec99e049481a1f3e955aaff23364c1775/server/production.ini#L20-L36 Each of the 3 instances should have 4 threads, so the semaphore should never overtook.
PROD running with 10 threads per instance. I will check in the log for requests rejected.