GPU tasks slow down macOS UI
16 Jan 2018 9:19:42 UTC
Topic 212727
(moderation:
Hello, everybody
I'm experiencing a problem when running CPU+GPU tasks (they're reported as 1 CPU + 1 NVIDIA GPU), namely, FGRPopencl-nvidia-mav. They're severely slow down UI response time of my macOS: window animations, content scrolling, font rendering when typing, Terminal output etc.
CPU consumption of the GPU-ed hsgamma process is 1.5-2.0%. Everything is back to normal when I suspend the GPU task.
Mid 2014 MacBook Pro with NVIDIA 750M, macOS 10.13.2 High Sierra
CUDA Driver 387.99
GPU Driver 10.28.10
BOINC 7.8.4
Thanks
ilya_11 wrote:Hello,
)
This is fairly normal but should not be a major problem as Boinc will slow itself down the more you use your laptop during that session. There are some settings under Options, Computing Preferences, then under the Computing tab called "When to Suspend", if you adjust those settings Boinc will actually stop crunching while you are physically using the laptop and then start crunching again after you are done. Be sure to click the "Save" box at the bottom of that page to save your settings.
Thanks, I checked the
)
Thanks, I checked the settings.
But could you please explain the root of the problem? It looks like a bus (FSB/PCI) saturation of some sort, and I wonder why it doesn't happen when I play gpu-intesive games, for example.
Regards,
Ilya
ilya_11 wrote:Thanks, I
)
I don't play games so have no clue, but do know that they are trying to optimize the thru put as much as they can get so they can analyze more data quicker. Each cpu unit now takes about 850mb of ram while a gpu unit takes about the same.
ilya_11 wrote:Thanks, I
)
CPUs have interrupts so if there is a queue of work in the pipeline and you start typing then keyboard strokes interrupts the other work to come 1st ahead of the queue. User response remains smooth and quick. GPUs are designed to process a video stream in near real time and do not have interrupts. So if the queue is full and you do something to cause a screen to redraw it stays at the end of the queue until the rest has processed. That ends up being screen lag when the GPU is fully utilized at 98-100%. From my own experience, lower 90ish percent GPU utilization is fine for web browsing. Playing videos may need more and has some stuttering depending on the BOINC project and its work load.
Thank you, guys.
)
Thank you, guys.