when i am running this command : mpirun -np 1 ./build/all_reduce_perf -b 8 -e 128M -f 2 -g 2. I found this

NVIDIA / nccl-tests

NCCL Tests

BSD 3-Clause "New" or "Revised" License

809 stars 229 forks source link

when i am running this command : mpirun -np 1 ./build/all_reduce_perf -b 8 -e 128M -f 2 -g 2. I found this #163

Open james2v opened 1 year ago

james2v commented 1 year ago

3d84f23a07473a75ee29bb4520f627c

sjeaugey commented 1 year ago

This is usually due to your container not having enough shared memory space. You may want to check the NCCL user guide (troubleshooting section) to launch with more shared memory.

james2v commented 1 year ago

it works ty so much!