Bugfix for SMP workloads on high core-count machines.

[cosmobobak]

No description provided.

[AndyGrant]

I assume the issue here is you have a big box like 256 threads, and you are running say 4 thread games, and then 64 cutechess copies is still too much?

We've updated to Fastchess now. I was actually intending to drop the -N sockets option entirely, and possible entierly remove the ability to run multiple tournament managers at once, since the MULTIPLE distribution type for SPSA is just not worth doing.

[cosmobobak]

Yes, precisely. If fastchess is totally stable and working, that's great, this can be ignored.

[AndyGrant]

Fastchess has been running on grantnet for a few days now, and we've not encountered any issues other than the initial setup for a Window's user.

If your framework wants to catch up to master, you could be the first test case for using a very high concurrency fastchess, with -N 1 despite having a big machine.