If the piece of code is really meant to use two threads even though one is requested, it is dishonest. Perhaps it could be called cheating since it would give the engine an advantage in single-threaded tournaments.
May I ask which engine it is from?
PS. A better solution would probably be to have two different builds. One which uses A/B and one which uses MCTS and has the minimum amount of threads set to two by default. That way there won't be any dishonesty.
niel5946 wrote: ↑Sat Jul 31, 2021 6:51 pm
If the piece of code is really meant to use two threads even though one is requested, it is dishonest. Perhaps it could be called cheating since it would give the engine an advantage in single-threaded tournaments.
May I ask which engine it is from?
PS. A better solution would probably be to have two different builds. One which uses A/B and one which uses MCTS and has the minimum amount of threads set to two by default. That way there won't be any dishonesty.
CMCanavessi wrote: ↑Sat Jul 31, 2021 6:43 pmNow, is that really valid/honest?
Engine designs that use N+1 threads are common e.g. if one thread is a dedicated input thread that sleeps as long as there is no input to process, which is why such an input thread is not counted against threads in terms of UCI.
Another possibility is to have two threads that can do chess-wise calculations, but only one of them is active at each point in time because they have some interlock mechanism on application level. This would still count as one UCI thread. Such a design may make sense if you have both an A/B and an MCTS search and want to keep them cleanly separated.
However, if this engine actually uses two calculation threads that are active at the same time, that's cheating.
The number of threads being used is announced by Fire in the UCI GUI at game start, so I don't think there should be any confusion.
In the meantime, I suggest only using MCTS mode in a tour that is 2 or more threads...
or if you prefer
download the 1st release of NN.MC binaries that are available from 7/27/21 Fire 8.NN.MC which do not contain the hack
and you can see the broken 1 thread mcts
this is something fairly simple I just haven't had time to properly address it.
I'll try to address it soon
Norm