You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
You can use gds_kernel_loopback_latency, which uses one GPU and one NIC to emulate both server and client.
Another option is to mpirun -n 2 <your-script>. Then, the script determines if it is process 0 or 1. The script then launches ./gds_kernel_latency -d <mlx5_0 or mlx5_1> -G <0 or 1>. This option is more complex but will allow you to use 2 GPUs and 2 NICs.
Activity
pakmarkthub commentedon Jan 10, 2023
Hi @hongbilu,
You can use
gds_kernel_loopback_latency, which uses one GPU and one NIC to emulate both server and client.Another option is to
mpirun -n 2 <your-script>. Then, the script determines if it is process 0 or 1. The script then launches./gds_kernel_latency -d <mlx5_0 or mlx5_1> -G <0 or 1>. This option is more complex but will allow you to use 2 GPUs and 2 NICs.protoss1235 commentedon Jan 30, 2023
thanks, I'll try it then