FFe: Sync nvidia-cuda-toolkit 10.1.105-1 (multiverse) from Debian experimental (non-free)
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
nvidia-cuda-toolkit (Debian) |
Fix Released
|
Unknown
|
|||
nvidia-cuda-toolkit (Ubuntu) |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
Please sync nvidia-cuda-toolkit 10.1.105-1 (multiverse) from Debian experimental (non-free)
Explanation of FeatureFreeze exception:
New upstream version with support for the latest GPU devices.
* Support added for GCC 8 and Clang 7
The packages contain pre-compiled binaries so there is no opportunity to backport fixes from future versions, so shipping the newest possible version in Ubuntu is desirable.
Note:
This has not been uploaded to Experimental yet (it was only released at the end of February, 2019) and will need to clear Debian NEW, so I may end up uploading 10.1.105-0ubuntu1 directly (but using the same orig.tar.xz) .
I have already tested snapshots from Debian's git in a PPA for amd64 and ppc64el, and rebuilding all reverse-
pycuda and starpu-contrib. Only caffe-contrib and starpu-contrib
required minor changes to switch from gcc-7/g++-7 to the default 8.
Changed in nvidia-cuda-toolkit (Debian): | |
status: | Unknown → New |
Changed in nvidia-cuda-toolkit (Debian): | |
status: | New → Fix Released |
Sounds like a good idea.
> so I may end up uploading 10.1.105-0ubuntu1 directly (but using the same orig.tar.xz)
I'd suggest doing that sooner rather than later, to get this stabilised in disco ASAP.