Hello,
I am installing the deb packages on TX1 and I get the following missing dependence:
dpkg: dependency problems prevent configuration of cuda-runtime-9-0:
cuda-runtime-9-0 depends on cuda-drivers (>= 384.00); however:
Package cuda-drivers is not installed.
the only driver package that came packed in:
cuda-repo-l4t-9-0-local_9.0.252-1_arm64.deb
is:
cuda-driver-dev-9-0_9.0.252-1_arm64.deb
Where can I download that missing package from ?
Thank you,
M
Normally CUDA 9 would be installed via JetPack3.2 running from a PC and talking to the Jetson over ethernet/ssh. You can run JetPack3.2 and disable all operations except for installing the CUDA package. A download URL for the package by itself is not available…if all fails there are still ways to extract this, but you’d still need the generated repository.json file from running JetPack to get the URL.
Thank you for your answer.
JetPack installation hangs indefinitely. So I had to flash-it by hand, following your instructions from January. Thank you very much, BTW!
I have the generated repository.json but there are is cuda-drivers url in it.
I was expecting it to be part of cuda-repo-l4t as it makes sense, but Nvidia obviously disagrees with that :-)
So, I am still looking for that package.
Tmx,
M
You need to start with the file which makes it possible to find CUDA packages. Look for the “cuda-repo-l4t-…” which applies the the system you are working with (JetPack understands packages for the host, TK1, TX1, and TX2s). Once this is installed and you run “apt update” you’ll see CUDA packages (these are actually installed to a local repository in “/var” rather than a remote server). After that install order may take some experimentation.
So be sure you got the right repo. Be sure you run “sudo apt update” before looking for other files for CUDA itself. Then you can locally run “apt search …” or “sudo apt-get install …”.
Thank you for your answer.
This is exactly what I did. I installed with dpkg the cuda-repo-l4t-9-0-local_9.0.252-1_arm64.deb package.
this created the folder /var/cuda-repo-9-0-local with lots of deb files
I installed successfully (lots of trial and error to find the right order :-) the following package from that directory:
cuda-core-9-0_9.0.252-1_arm64.deb
cuda-cublas-9-0_9.0.252-1_arm64.deb
cuda-cublas-dev-9-0_9.0.252-1_arm64.deb
cuda-cudart-9-0_9.0.252-1_arm64.deb
cuda-cudart-dev-9-0_9.0.252-1_arm64.deb
cuda-cufft-9-0_9.0.252-1_arm64.deb
cuda-cufft-dev-9-0_9.0.252-1_arm64.deb
cuda-curand-9-0_9.0.252-1_arm64.deb
cuda-curand-dev-9-0_9.0.252-1_arm64.deb
cuda-cusolver-9-0_9.0.252-1_arm64.deb
cuda-cusolver-dev-9-0_9.0.252-1_arm64.deb
cuda-cusparse-9-0_9.0.252-1_arm64.deb
cuda-cusparse-dev-9-0_9.0.252-1_arm64.deb
cuda-driver-dev-9-0_9.0.252-1_arm64.deb
cuda-gdb-src-9-0_9.0.252-1_arm64.deb
cuda-libraries-9-0_9.0.252-1_arm64.deb
cuda-libraries-dev-9-0_9.0.252-1_arm64.deb
cuda-license-9-0_9.0.252-1_arm64.deb
cuda-misc-headers-9-0_9.0.252-1_arm64.deb
cuda-npp-9-0_9.0.252-1_arm64.deb
cuda-npp-dev-9-0_9.0.252-1_arm64.deb
cuda-nvgraph-9-0_9.0.252-1_arm64.deb
cuda-nvgraph-dev-9-0_9.0.252-1_arm64.deb
cuda-nvrtc-9-0_9.0.252-1_arm64.deb
cuda-nvrtc-dev-9-0_9.0.252-1_arm64.deb
then it was time to install:
cuda-runtime-9-0_9.0.252-1_arm64.deb
and here is where I get:
“”"
dpkg: dependency problems prevent configuration of cuda-runtime-9-0:
cuda-runtime-9-0 depends on cuda-drivers (>= 384.00); however:
Package cuda-drivers is not installed.
dpkg: error processing package cuda-runtime-9-0 (–install):
dependency problems - leaving unconfigured
Errors were encountered while processing:
cuda-runtime-9-0
“”"
There is no cuda-drivers in that folder, nor does apt search find any. So, here is where I am stuck…
Thank you,
M
By forcing instalation of libxmu-dev I also got installed
cuda-samples-9-0 (9.0.252-1)
cuda-documentation-9-0 (9.0.252-1)
cuda-toolkit-9-0 (9.0.252-1)
cuda-command-line-tools-9-0 (9.0.252-1)
cuda-minimal-build-9-0 (9.0.252-1)
Now the only things left in that directory that is not installing due to cuda-drivers is:
cuda-runtime-9-0_9.0.252-1_arm64.deb
which does not seems to contain anything useful:
“”"
drwxr-xr-x root/root 0 2018-02-28 01:21 ./
drwxr-xr-x root/root 0 2018-02-28 01:21 ./usr/
drwxr-xr-x root/root 0 2018-02-28 01:21 ./usr/share/
drwxr-xr-x root/root 0 2018-02-28 01:21 ./usr/share/doc/
drwxr-xr-x root/root 0 2018-02-28 01:21 ./usr/share/doc/cuda-runtime-9-0/
-rw-r–r-- root/root 89 2018-02-28 01:20 ./usr/share/doc/cuda-runtime-9-0/copyright
-rw-r–r-- root/root 146 2018-02-28 01:20 ./usr/share/doc/cuda-runtime-9-0/changelog.Debian.gz
“”"
Is that really necessary ?
Do you know of a quick way to validate that cuda is working fine as it is ?
Thank you,
M
The “cuda-runtime-9-0_9.0.252-1_arm64.deb” package naming implies desktop PC (x86_64/amd64) runtime. Is it correct that you are installing this to host? If so, then you do need the NVIDIA video driver on the host before you can install CUDA runtime (the open source Nouveau driver won’t work). Is it correct that the files you mentioned are from a doc package and not “cuda-runtime-9-0_9.0.252-1_arm64.deb”? Do you have the NVIDIA video driver on the host?
Here is what I find, maybe somebody else will find it useful too.
The packages to be used are the arm64.deb files, the amd64.deb are for the host and should never be copied on the target.
The solution seems to be to just ignore the cuda-runtime package which anyhow, does not contain anything that needs to be used.
The apt-get must be used to install libxmu-dev package and then all the rest will install after playing around to figure out the order.
After that the CUDA and CUDNN samples are able to compile and run so I assume everything that is functional should be installed even without the troublesome cuda-runtime. If I find any problems I update this thread to let you guys know.
Tnx,
M