-
Notifications
You must be signed in to change notification settings - Fork 27.5k
CUDNN version in nightly pytorch 2.10.0 builds #167242
Copy link
Copy link
Closed
Labels
module: binariesAnything related to official binaries that we release to usersAnything related to official binaries that we release to usersmodule: cudnnRelated to torch.backends.cudnn, and CuDNN supportRelated to torch.backends.cudnn, and CuDNN supporttriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
Milestone
Metadata
Metadata
Assignees
Labels
module: binariesAnything related to official binaries that we release to usersAnything related to official binaries that we release to usersmodule: cudnnRelated to torch.backends.cudnn, and CuDNN supportRelated to torch.backends.cudnn, and CuDNN supporttriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
Hi, I mainly use pytorch with ComfyUI. I know there is an issue with pytorch and CUDNN for which there have been made workarounds in ComfyUI code.
I have seen here #166122 that CUDNN 9.15 solves the problem (from I can understand, as I'm not a developer). Checking today's torch nightly 2.10.0+cu130 for Windows it shows, if I'm not mistaken, CUDNN version 9.12:
My question is: when is to be expected to see CUDNN v 9.15 in the nightly 2.10.0+cu130 builds?
And another question is: seeing that today CUDNN 9.15 is available from nvidia (in fact is already downloaded and installed on my computer) is there a way to use 9.15 in the current torch build, as this comment suggests?
#166122 (comment)
My apologies, if I ask, maybe, trivial questions.
cc @seemethere @malfet @atalman @csarofeen @ptrblck @xwang233 @eqy