Just noticed that my Linux jobs (SETI@home v8 8.22 opencl_intel_gpu_sah) claim to require 0.00641 CPUs = 1 Intel GPU, but actually continuously consume 100% of a CPU.
(Equally oddly, but "better" for me, is that a collatz GPU job claims to need 0.886 CPUs, but actually only uses 1.7% of one).

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
32365 boinc     30  10  482356 321100  60340 R 100.3   4.0   2:24.35 setiathome_8.22

straceing the process show it doing this, very quickly...

sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8e20) = 0
futex(0x1f341f0, FUTEX_WAKE_PRIVATE, 1) = 1
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8b20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8b20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8b20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8e20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8b20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8820) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8820) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8820) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8820) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8b20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8b20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8e20) = 0
ioctl(8, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffcd63f8e20) = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
sched_yield()                           = 0
.....

You have fallen into the common misconception that the task identifier showing some xx percent of a cpu is what is needed to support the gpu task. It is not. It is value used only for scheduling of work sent to a host by the scheduler. Only the application itself determines how much of a cpu thread is needed to support the gpu task.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

You have fallen into the common misconception that the task identifier showing some xx percent of a cpu is what is needed to support the gpu task. It is not. It is value used only for scheduling of work sent to a host by the scheduler.

Which ignores it, from what I can see.

Only the application itself determines how much of a cpu thread is needed to support the gpu task.

OK.
But running at 100% still seems wrong when the vast majority of syscalls as just sched_yield().

You have fallen into the common misconception that the task identifier showing some xx percent of a cpu is what is needed to support the gpu task. It is not. It is value used only for scheduling of work sent to a host by the scheduler. Only the application itself determines how much of a cpu thread is needed to support the gpu task.

Doesn't the display quoted originate from Linux, rather than from BOINC? Linux itself doesn't pay any attention to BOINC's estimates, internally. It simple measures what it sees.

Doesn't the display quoted originate from Linux, rather than from BOINC? Linux itself doesn't pay any attention to BOINC's estimates, internally. It simple measures what it sees.

BOINC isn't measuring anything. It's stating that its requirement for CPU usage is very low, whereas in fact it is very high.
There is no way that this is going to help any scheduler.