[Question] Training DQN in Atari Env occupy multiple threads

[PaladinEE15]

Training DQN in atari env will occupy 9-10 threads, is it a normal phenomenon?
(Platform： Nvidia Titan XP + Xeon E5-2630 v4)

[araffin]

Hello,
this is normal, PyTorch is using threads to parallelize matrix computation.
You can change that using:

• th.set_num_threads() see doc (and optionally th.set_num_interop_threads())
• export OMP_NUM_THREADS=1 (to set OpenMP threads to 1)

[PaladinEE15]

Hello,
this is normal, PyTorch is using threads to parallelize matrix computation.
You can change that using:

• th.set_num_threads() see doc (and optionally th.set_num_interop_threads())
• export OMP_NUM_THREADS=1 (to set OpenMP threads to 1)

Thank you very much!