I can run some examples from cuda8.0 ,But Cuda 9.0 examples can’t. It said cuda version is error I can’t understand it .Nvcc -V show that it’s cuda9.0 9.0.252. Help !How to solve it?
Please remember that there are some dependencies between the OS and packages.
You cannot install JetPack3.2.1 packages on the OS flashed by another JetPack version.
Could you try to set up all the requirements from the JetPack3.2.1 and run the CUDA sample again?
We have verified the CUDA sample with JetPack3.2.1 and it works correctly.
Now my environment is error. So How can I recover to the Initialization environment?Before I set my jetpack in my tx2
It’s recommended to re-flash your device with JetPack3.2.1 to get everything work.
AastaLLL, can you share which dependencies CUDA 9.0 libraries have on JetPack 3.2/3.2.1 system? I need to install CUDA 9.0 on the system originally flashed with JetPack 3.1. I cannot re-flash it with Jetpack 3.2.x, but I can try to manually install packages to satisfy dependencies.
This is probably not recommended, but I use it all the time to install packages on the TX2 from JetPack.
You can run the installer and look for the individual .deb files that you’re interested in and install them manually.
Step 1: Download Jetpack from https://developer.nvidia.com/embedded/jetpack
Step 2: Extract contents of installer
bash JetPack-L4T-X.Y-linux-x64.run --noexec
X: major version, Y: minor version
Step 3: Run Chooser to generate the repository.json file
Look inside the contents of the repository.json file and you can “wget” and install them using “dpkg”. You should see missing dependencies if there are any and install them accordingly.
Not a very neat solution, but works for me.
singularity7, thanks! I was thinking of trying to do it, so your experience is very encouraging!
Did you ever try to install CUDA 9 packages on JetPack 3.1 system and whether it worked for you?
We’ve got many TX2 devices here running different versions of things, so I can’t be absolutely sure if we had JetPack3.1 + CUDA 9.0, but I vaguely remember having this configuration for a while and it working correctly. It’s a short install procedure so give it a shot. You can always revert back to 8.0 if it doesn’t work as planned. Don’t forget to update your bashrc and recompile your samples (if you need them) once you get 9.0 installed!
Great, thanks! I’ll give it a shot.
Happy to report that now we have an old Jetpack 3.1-based system running the latest CUDA/cuDNN/TensorRT libraries along with nvidia_drivers (all extracted from Jetpack 3.2.1). Everything seems to be working well, including ZED SDK 2.4.1 and DeepStream 1.5. So apparently there’s no tight relationship between kernel and userland on Jetson and reflashing JetPack can be avoided.
singularity7, thanks a bunch for inspiration!
Glad that you’ve got it running now. Good luck!