I’ve just published cuda-runtime-11-2-1804 as a content snap containing the CUDA runtime libraries from nVidia. The following slot is defined, which I would like to allow any consumer snap to plug automatically:
Now I’ve ironed the kinks out, a working example of a snap using the CUDA runtime snap is my “fakecam” app in beta and edge channels on amd64 (snap install fakecam --beta) revision number 110:
Curious to know: does the license allows to redistribute CUDA libs ? Asking because I have refrained myself from publishing that stuff previously as I had the need for CUDA 10.2 as well.
Fudge. Fudge. and more Fudge. Well that puts a downer on things. It’s worded as redistributable but as part of an application that adds material workings beyond the “sdk”. Anyone know someone at nVidia I can email?
If the distribution terms in this Agreement are not suitable for your organization, or for any questions regding this Agreement, please contact NVIDIA at nvidia-compute-license-questions@nvidia.com.
Nvidia do have a list of specific files that are redistributable, as Daniel says, provided they provide material value alongside other utility. libcudart is on this list and so generally speaking it’s allowed to ship CUDA embedded into snaps, but the definition of material value does complicate content snaps.
Hey @lucyllewy could you please confirm there are no licensing issues with cuda-runtime-11-2-1804? I think this is important to move fw with this request.
Unfortunately it looks like licensing is an issue.
If we can distribute it on the same basis that the non free Debian repository does it (I don’t know how they’re allowed to do it, either) then we can move forward, but not yet. Please take the request off your radar for now.
Aside: @popey mentioned there were discussions internally about this snap.