Hi, I follow this instructions for installing CUDA: https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions
and installed with deb(network). Im using Ubuntu 16.04.
Now Im performing the post-installations actions. I have exported the PATH variable like the instructions said and when I do echo to that variable it shows:
/home/jetson/bin:/home/jetson/.local/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
so its ok.
Now Im not sure if I have to export the LD_LIBRARY_PATH too, but i did it anyway and whn I echo its shows:
/usr/local/cuda-10.0/lib64
si its ok too I think.
Then I run the nvcc -V command and tell me that the program nvcc its not installed and to install it i have to execute sudo apt-get install nvidia-cuda-toolkit.
Why is this happening? I think I followed the instructions well. I have installed GCC 5.4.0.
Are you installing this on a Jetson?
I don’t know what you mean by “I follow this instructions” You haven’t indicated what you mean by “this instructions”
Since you indicate
/usr/local/cuda-10.0/lib64
as your LD_LIBRARY_PATH, I assume your CUDA install path is
/usr/local/cuda-10.0
If that is the case, and you are looking for nvcc from cuda 10.0, then something like:
/usr/local/cuda-10.0/bin
needs to appear in your PATH output. This is indicated in the post install actions in the linux install guide. Since that doesn’t appear in the PATH output you have shown, I’m not surprised that nvcc cannot be found.
Sorry, I forgot to post the link to the instructions. Is this: https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions.
The 7.1.1 Environment setup says that I need to add to the PATH the path to bin of cuda 10. This is clear but then it says: In addition, when using the runfile installation method which i didnt use because I use the deb(network) package installation, you have to add tho the LD_LIBRARY_PATH variable the following:
export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
I have not test it with the path you recommends, but I just want to explain more clear my problem. When I test that path I will see if nvcc -V works and reply with the results.
Thank you for the help and your time.
you didn’t answer my question about jetson
and you did not follow the instructions correctly for the PATH setup
the nvcc not installed problem has nothing to do with your LD_LIBRARY_PATH setup
No im not installing on a Jetson. Now I see my error, I dont know why I thought I have exported the PATH variable correctly when I clearly didnt. Now nvcc -V shows up some information.
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130
Thank you for your help and time!
Hello I have some problem intalling CUDA 10-0-0 on Ubuntu 18.0.4
I have installed cuda 10-0 but i can’t find its directory
Now there is usr/local/cuda link but this is linked with usr/lib/cuda and its version is 9.1
And with nvcc -V it shows version 9.1
And with nvidia-smi shows version 10.0
I am going to install tensorflow-gpu version 1.14
What is wrong and how can i fix this problem?
I want to find cuda-10 directory and want to work with nvcc -V
If you installed CUDA 10.0 you need to follow the post install instructions in the cuda linux install guide.
I had this error with CUDA 11. I think the issue is that the instructions in the Linux install guide only lead to a temporary change to PATH.
I just added the the line export PATH=/usr/local/cuda-11.0/bin${PATH:+:${PATH}} to my .profile, and it seems to be working.
1 Like