Hi,
I responded to several threads already o nthe subject but no clear answers has been made besides the fact of using a specific program in the exclusion list.
But I'm not under windows, I'm under fedora 32 and I would like to understand where does it coming from that gpugrid computing can't stop the gpu from working while on all my windows host it's working properly.
So please give me an answer to that and be specfici. IS it the responsability of fedora, of gpugrid or of boinc?
It's really really annoying, and again it's working fine underwidnows hosts.
the gpu is nvidia by the way.


Thanks in advance

It can't be the responsibility of GPUGrid as it is just the science application. Only BOINC determines how BOINC works. And it is possible that something about Fedora is preventing BOINC from working normally.
The GPUGrid app runs at nice 10 normally and it is possible that Fedora might be changing that to a higher nice level.
Or it is possible that you are overcommitting the cpu and preventing the cpu thread that is servicing the gpu task from responding promptly to the in use detection.
The GPUGrid application requires a full cpu thread for each task. It takes a while for BOINC to notice the in use detection.

For example in my very heavily committed processor running 30 of 32 threads, it takes quite a while for BOINC cpu processes to end after quitting BOINC.

There is another thing as well: security.

Under Windows we measure idle time with a dedicated small program that constantly runs in the background (boinctray.exe) for your user account. And for any other account you gave permission to control BOINC. BOINC Manager itself also has code on board to check for idle time. Since the mouse and keyboard processes run under your user account in Windows, it's easy to check these.

Under Linux that's a difficult thing to do, because mouse and keyboard run under a different process than BOINC or even your user account. Last I heard we measure idle time by checking X server, but that doesn't work the same under all Linux distributions. And once the developers made it working for one distro, others will fail again. There's no single easy check or API to fall back to, because every distro handles idle time differently and to go code If strings for all different distros is undoable.

---

Fedora, though being well aged for a Linux distribution, is not one of the most common operating systems either.
It's based on Redhat, which also CentOS is. I wonder if those 2 are also affected by it?

I've never really played around with RPM based linux versions, but most of the Debian and Ubuntu based operating systems seem to work quite well.

thanks for all those complete explanations. That's way much clearer when you have explained all the consequences and causes of those checker processes.
Yeah so definitely i think it's fedora the problem since twe are still under this batttle between wayland use and xorg use. xorg used only if you install the nvidia proprietary drivers by a certain way or an other.
Plus as docker's users are experiencing, there are also the problem about cgroups security which hasn't been resolved.

But then that means it won't be fixed because there are too much things up in the stream to fix yet.
Could we think about putting it in the wiki please as a warning for fedora users?

by the way it's not the same process than detect if the computer is busy because the option about cpu ressources being taken more than x% works.