Hi,
I did some testing and with the DAC plugged in, temperatures are roughly 6 degrees celsius higher and obviously power usage is more. This is even if I disable the Mellanox.
meanaverage@dgx2:~$ for dev in 0000:01:00.0 0000:01:00.1 0002:01:00.0 0002:01:00.1; do
echo “$dev” | sudo tee /sys/bus/pci/drivers/mlx5_core/unbind
done
0000:01:00.0
0000:01:00.1
0002:01:00.0
0002:01:00.1
meanaverage@dgx2:~$ for dev in 0000:01:00.0 0000:01:00.1 0002:01:00.0 0002:01:00.1; do
echo 1 | sudo tee /sys/bus/pci/devices/$dev/remove
done
1
1
1
1
Is there a command that can truly override CX7 power control? Nothing but actually ejecting the DAC brings the temperatures way down.
(immediately upon DAC pull)
Often I’m processing jobs on multiple DGX’s and I don’t always need 200GbE between them for parallel workflows, and have redundant paths to the NAS and each other via 10GbE… so I’d love a way to just nip off the CX7 when it’s not needed during these long, hot runs.
Thanks