-
Notifications
You must be signed in to change notification settings - Fork 61
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Reduce even more GPU allocations #2394
Reduce even more GPU allocations #2394
Conversation
I tested this analogously to PR 2393 and also here, all my tests pass. For the specific case, however, the further reduction in memory consumption is quite small (from 111.5 MB to 109.8 MB). |
Well, that's expected as this reduction is conditional on certain commonly used things not being used. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
With many neuronal compartments, I just encountered very high main memory usage, but I guess that is a separate issue.
Regarding this PR, I don't have any further complaints.
The high memory use is a concern, yes, let's discuss that elsewhere, though. |
…' into mem/trim-gpu-allocation-even-more
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also here, the failing pipeline might have to be fixed or removed.
Bubbling up ion access information from modcc allows us to skip allocation of
Xi
andXo
iffno mechanism reads those values.
A minor problem: Sampling may try to touch data that doesn't exist; however who'd ask for
Xi
if they never use it?
Stacking this PR on top of #2393 the memory consumption drops by a further 5% down to 77%
if the original value.
Todo
Xi
/Xo