I agree that it would be good to be able to suspend specific GPUs
- manually
- when an exclusive app is running
- when computer is in use.
However, this would be a large and complex code change:
1) GPU suspension currently applies to all GPUs,
so there's only one copy the data structures.
We'd need to add copies of the data structures for each GPU
(and keep the current one).
2) Any change to CPU/GPU scheduling must be reflected in the work fetch logic.
For example, if a particular GPU were suspended,
the current work fetch code would see it as an idle device
and try to fetch work.
3) Designing and developing a GUI for it would be a lot of work.
Bottom line: I don't think we can do this any time soon.
I added it as a feature request on GitHub:
https://github.com/BOINC/boinc/issues/1459
Note: this feature will be easier to implement when we switch to
modeling each GPU as a separate resource.
-- David
On 12/30/2015 1:10 AM, Jord van der Elst wrote:
Hi David,
The past couple of days we've had a person try to suspend individual GPUs in
https://boinc.berkeley.edu/dev/forum_thread.php?id=10698
When he games, he wants his Nvidia GPU to suspend and his AMD GPU to continue.
I'm thinking this could be solved by adding a couple of exclusive_gpu switches,
, and ,
where when these are set and the exclusion program is in memory all of the
AMD/Nvidia/Intel GPUs get suspended, but those not specifically mention continue.
Is that a doable thing?
Is that something we'd want to add? Or is that thought to be too fiddly?
-- Jord van der Elst.
___
boinc_dev mailing list
boinc_dev@ssl.berkeley.edu
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
