I noticed this while experimenting with running two tasks on my single gpu. CPU usage of the GPU app increased from 5% to 60%. This made me to check the milkyway process with htop strace feature and I found that milkyway is performing approximately 100 calls to clock_gettime(CLOCK_MONOTONIC_RAW...) in 100ms every 300ms. This spam of sycalls also happens when running SINGLE milkyway task and nothing more on the machine.
I do not think the app really needs to know current time such often. Does the clock_gettime call enable communication with the gpu? I would expect some reads or ioctls.

I am talking about app name milkyway and milkyway_separation... plan_class opencl_nvidia obviously on Linux (weed) 4.2.0-1-amd64, nVidia GeF840M dirver 352.79.

The same I noticed on primegrid genefer. On the other hand Einsten and Asteroids tasks are quietly waiting on some FUTEX locks.