too_many_fds_error

[ma80mk]

When I try to use any kinds of multiprocessing calculations (with cpu/gpu and num_workers bigger than zero, or more than one gpu), I receive the "too many fds" error. I use the latest version of MACE v0.3.16, installed on the cluster by submitting PBS jobs. Does anybody have any ideas to solve this problem?

[ilyes319]

Hey can you post your full error so we can help.

[ma80mk]

Hi, Please find an example of the cluster output file in the attached zip file.
MACE_gpu.o20453200.zip