Has NVIDIA made public a policy on support for older versions of CUDA? When I file bug reports, it seems that the immediate response is “have you tried with the latest version of CUDA?” Since CUDA upgrades generally involve driver upgrades too this is not always a simple check to perform (especially when I don’t always have root access to the relevant machines). I know that (in theory) so long as the driver is kept up-to-date, I can keep multiple versions of CUDA itself around. However, that only solves part of the problem. If there’s an official statement from NVIDIA about support, it would greatly help planning.