I have been trying to install CUDA toolkit 11.1.1 on my Ubuntu so I can then use the LTS PyTorch 1.8.2.
Following are the hardward and software specifications of my system:
Motherboard: ASUS TUF GAMING X570-PLUS (WI-FI) Rev X.0x
CPU: AMD Ryzen 7 5800X 8-Core Processor
GPU: NVIDIA GA106 [GeForce RTX 3060 Lite Hash Rate]
OS: x86_64 Ubuntu 20.04.1 LTS (Focal Fossa)
Driver: nvidia-driver-470 (proprietary) v470.103.01
Kernel: 5.4.0-42-generic, GCC: 9.3.0, GLIBC: 2.31
So I followed through with the download instructions and successfully installed cuda.
Upon restart, as soon as launching into Ubuntu, the monitor, which is directly hdmi-ed into the 3060, black screened, with no more display coming through. After relaunching into Ubuntu but in recovery mode this time, “nvidia-smi” in the root bash wielded “no devices were found” and “nvcc -V” wielded cuda not installed and suggested a sudo apt install of cuda 10.1, which I believe is not compatible with Ubuntu 20.04.
I have now reinstalled Ubuntu, and without installing cuda, “nvidia-smi” is able to see gpu and works just fine.
How do I correctly install cuda without running into previous issues?
*secure boot is turned off in the system bios
*there is a fresh Windows dual booted on my device