Hi there, I am trying to use:
sudo sh cuda_9.1.85_387.26_linux.run
(on Fedora 27 but question is independent) with a different gcc than system default v7.
I have a gcc v6 version called
cuda-gcc
but do not not how to feed this info to the runfile or how to set environment variables.
Surely, there is a way to use cuda-gcc v6 w/ runfile. Please advise. Thank you very much indeed!
Regards, Thomas
The easiest way to do this is to just make sure that the path to your cuda-gcc build appears in your PATH variable before the path to the system default gcc. You should also create an alias or symlink so that the gcc and g++ commands direct to your special build of gcc, rather than the system default.
Since the runfile is executed as root user, you’ll need to make sure this PATH is setup for root, as well.
I did this on Fedora 25 to support CUDA 8, with a special build of gcc 5.4 instead of the system default of 6.x
As I recall, it did not really matter for the runfile installer, but is important for building CUDA codes (e.g. building the CUDA samples)
Hi there, I did all that plus a few other measures but failed. The runfile installs everything and issues no errors: all fine. But after reboot, I get a blank screen. Looking at the log files, I suspect it is related to Wayland and except negativo17 probably not many have gotten the runfile to work. The policy of NVIDIA to lag behind by two versions on Fedora is not really customer friendly but there may be good reasons for it.