Tesla K20m - Clock Throttling - HW Slowdown

Hi,

I am working with a Tesla K20m card. The clock throttling on this card seems to be always on.

The relevant nvidia-smi output is

Clocks Throttle Reasons

Idle                        : Active
Applications Clocks Setting : Not Active
SW Power Cap                : Not Active
HW Slowdown                 : Active
Sync Boost                  : Not Active
Unknown                     : Not Active

I have tried the simple steps like GPU Reset, Reboot the system and they have not helped. Google hasn’t been of a lot of help with this problem.

I have seen the 4 reasons that could lead to “HW Slowdown” from http://developer.download.nvidia.com/compute/cuda/6_0/rel/gdk/nvidia-smi.331.38.pdf.

  1. The temperature is under 50
  2. PState is constant when I check the status
  3. Power draw can't be too high as the measured power is around 17W (P8 power state)
  4. Since the other reason is related to power supply, I have tested the card with two different power supplies. However the problem remains even when I use alternate power supplies

The system has Ubuntu 16.04 OS and I have CUDA 8.0 and driver version 375.66 installed. Earlier I had driver version 375.57 installed and the problem exists in both driver versions.

Any suggestions will be very helpful.

Hi, I’ve encountered the same problem with P4 Ubuntu 16.04 CUDA 9.0 and driver version 384.145. Did you solve the problem? Any insight?

Please enable the nvidia-persistenced to start on boot, make sure it is continuously running and check if that resolves the issue.