Cuda for nvidia jetson nano

[paulgheorghe94]

Hello,

I wanted to install scikit-cuda to accelerate FFT and it complained about not finding cuda.h. I followed this tutorial Installing CUDA on Nvidia Jetson Nano - JFrog Connect and after fixing errors, I managed to pip install scikit-cuda, but it doesn’t work.

nvidia-smi says NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

Importing pycuda yields
Traceback (most recent call last):
File “”, line 1, in
File “/home/paul/.local/lib/python3.8/site-packages/pycuda/autoinit.py”, line 5, in
cuda.init()
pycuda._driver.Error: cuInit failed: unknown error

I have nvcc.
nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Thu_Jun_11_22:26:42_PDT_2020
Cuda compilation tools, release 11.0, V11.0.194
Build cuda_11.0_bu.TC445_37.28540450_0

In bashrc I put:
export PATH=“/usr/local/cuda/bin:${PATH}”
export LD_LIBRARY_PATH=“/usr/local/cuda/lib64:${LD_LIBRARY_PATH}”
export CPATH=$CPATH:/usr/local/cuda-11.0/targets/sbsa-linux/include
export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/cuda-11.0/targets/sbsa-linux/lib

Is there something I can do to repair this? Thanks!

LE: I installed CUDA with SDK Manager after some battles :D
nvidia-smi doesn’t exist anymore, but the pip package is working. This can be closed I think.

[linuxdev]

Note that Jetsons have an integrated GPU (iGPU), and it is wired directly to the memory controller. Desktop PCs have a discrete GPU (dGPU), and this is accessed through PCI. The nvidia-smi only works with a dGPU and has no possibility of functioning on a Jetson. Did you install a CUDA or video driver via anything other than JetPack/SDK Manager? If so, then you might need to flash the system again. Mixing the two is in general a "bad idea"™.

Also, I see this is CUDA 11. Nano is incompatible with CUDA version 11, and uses version 10. Many people try to install version 11, and then run into this before discovering the iGPU cannot use 11. To use 11 you’d need an Xavier or newer (Orin).

Note: Xavier has an NX/Nano form factor, but you’ve posted in the original Nano forum. That’s a TX1.