You can see the clocks dropping as soon as the GPU is active here:
$ nvidia-smi dmon
# gpu pwr temp sm mem enc dec mclk pclk
# Idx W C % % % % MHz MHz
0 - 40 0 0 0 0 2505 936
0 - 40 0 0 0 0 2505 936
0 - 40 0 0 0 0 2505 936
0 - 40 0 0 0 0 2505 936
0 - 40 1 0 0 0 2505 936
0 - 40 0 0 0 0 2505 936
0 - 40 100 46 0 0 405 135 <== glxgears started
0 - 40 100 58 0 0 405 135
0 - 40 100 58 0 0 405 135
0 - 41 100 57 0 0 405 135
0 - 40 100 56 0 0 405 135
0 - 41 100 58 0 0 405 135
0 - 41 100 57 0 0 405 135
0 - 41 100 58 0 0 405 135
0 - 41 50 8 0 0 2505 936 <== glxgears stopped
0 - 41 1 0 0 0 2505 936
0 - 41 0 0 0 0 2505 936
If I look at the clock throttle reasons while glxgears is running I see:
$ nvidia-smi -q | grep 'Power Cap' -C 5
Fan Speed : N/A
Performance State : P8
Clocks Throttle Reasons
Idle : Not Active
Applications Clocks Setting : Not Active
SW Power Cap : Active
HW Slowdown : Not Active
Sync Boost : Not Active
Unknown : Not Active
FB Memory Usage
Total : 4037 MiB
This is not specific to glxgears. It happens with everything I’ve tried so far. It is resulting in terrible performance!
This is on a new laptop; a HP Omen 15-ax000na. It has a GTX 965M. I’m running Ubuntu 16.04.1 64-bit with the nvidia-375 driver package from the graphics-drivers PPA (https://launchpad.net/~graphics-drivers/+archive/ubuntu/ppa). I’ve seen the same thing with all driver versions I have tried so far though: 340, 367, 370, 375.
It was working for a brief period yesterday with the 367 driver. I believe it started working after a resume when I plugged the laptop into AC power. I then foolishly upgraded to the 375 driver to see if it would work with that and it stopped working :(. After downgrading to 367 again it was still broken.
nvidia-bug-report.log.gz (213 KB)