This effort augments present methods builders have CUDA entry by letting them acquire the software program wanted in a single location. Further distributors are coming quickly.
Every distribution platform that redistributes CUDA will embody:
-
Constant CUDA Toolkit naming: Third-party packages will match NVIDIA naming conventions to keep away from confusion in documentation and tutorials.
-
Well timed CUDA updates: Third-party packages will probably be up to date in a well timed method after NVIDIA official releases to make sure compatibility and cut back QA overhead.
-
Continued free entry: CUDA itself will stay freely obtainable—even when packaged in paid software program. Distributors could cost for entry to their packages or software program however won’t monetize CUDA particularly.
-
Complete assist choices: You possibly can entry assist through distributors and also can discover assist through NVIDIA boards or NVIDIA’s developer web site, similar to at all times.
Acquiring CUDA software program from NVIDIA is free, and present avenues to get CUDA stay (they embody downloading the CUDA Toolkit, pulling the CUDA container, putting in for Python utilizing pip or conda).
“However the potential for distribution platforms to bundle CUDA inside bigger enterprise deployments and software program purposes permits us to make sure your expertise as a developer is straightforward. You obtain and set up your software, and underneath the covers, the right CUDA model is put in, as effectively,” the corporate stated in a weblog by NVIDIA’s Jonathan Bentz, Rob Armstrong and Anshuman Bhat. “Working with the NVIDIA ecosystem on this method is a major milestone in our mission to scale back friction in GPU software program deployment. By collaborating with key gamers throughout the OS and bundle administration panorama, NVIDIA is making certain that CUDA stays accessible, constant, and straightforward to make use of—irrespective of the place or how builders select to construct.”
