I have cuda-drivers and cuda-toolkit-12-8 installed. I also need a parallel installation of cuda 11.8. The latter can be a user install or a system install. How can I do that? Where can I find a tarball for cuda 11.8?
Thanks,
Jeff (http: //engineering.purdue.edu/~qobi)
You can find past releases here.
Bypass or opt out of installing the driver that comes along with it, read the Install Guide carefully.
There are no cuda-11.x packages for Debian-12 that I’m guessing you need it for based on your other posts. You may try temporarily adding Debian-11 repo (the same as the one pointed by the doc linked by @rs277) and install from there, but I’m not sure if it will work: it may mess your system up, so do test it in some disposable container/vm first.
Also as pointed elsewhere, it’s probably better to install cuda-<version> metapackages rather than cuda-toolkit-<version> directly.
Thanks. Yup. I need to recompile the old C/C++/CUDA code that came with Torch 7 (Lua Torch, not PyTorch) because I made a fork of that code and have modified it. I can’t easily migrate my code to ATen. CUDA 11.8 compiles that Torch 7 code. CUDA 12.8 does not. I discovered that CUDA 12.3 does. I haven’t checked versions between 12.3 and 12.8. It would be great if NVidia didn’t break old major code like that. Debian 12 has CUDA 11.8 in its nvidia-cuda-toolkit.