Markus

A while back I translated the python-chess syzygy implementation to Chess.jl. It is not thoroughly tested because in the end I did not use it in my engine but it...

> * Try limiting XLA to use fewer threads > `export XLA_FLAGS="--xla_cpu_multi_thread_eigen=false` `intra_op_parallelism_threads=1"` Thank you! Unfortunately, setting the XLA_FLAGS did not change anything on my side.

I have set the flags with `export`, but even if I set them in the script like you suggested, I have the degrading performance. I attach a perf analysis computed...

I don't know why one cannot see the helpful comments of @spnixog anymore, but I added ```python import os os.environ["XLA_FLAGS"] = "--xla_cpu_multi_thread_eigen=false intra_op_parallelism_threads=1" os.environ["OMP_NUM_THREADS"] = "1" os.environ["OPENBLAS_NUM_THREADS"] = "1" os.environ["MKL_NUM_THREADS"]...

After ablation testing, I found that setting either one of the following flags resolves the issue - `os.environ["OMP_NUM_THREADS"] = "1"` - `os.environ["OPENBLAS_NUM_THREADS"] = "1"`