I wanted to download tensorflow with cuda and i don’t know where to start and which versions to be downloaded for jetson orin nano 8 GB developer kit with GPU: 1024-core NVIDIA Ampere architecture? and what nvidia driver should i need to install ?
can anyone please guide me to download them and send few links if possible ?
You have to start by knowing which release of software is flashed to the Jetson. “L4T” (Linux for Tegra) is just Ubuntu plus NVIDIA drivers, and you can find the L4T release via:
head -n 1 /etc/nv_tegra_release
From there, you find the JetPack/SDK Manager which matches this. JetPack and L4T releases are tied together, so if you pick one, you’ve picked the other. You can go to the specific release page through either of these links:
The documentation will be specific to your release (and often to the hardware). You would install CUDA via JetPack/SDK Manager running on an Ubuntu host PC (in this case Ubuntu 20.04 is recommended).
Note that most GPUs which you find drivers for are “discrete” GPUs (dGPU). This means they are on the PCI bus. Jetsons have an integrated GPU (iGPU) in which the GPU is directly wired to the memory controller without the PCI getting involved. This in turn means that your generic drivers you find in the wild (even if arm64/aarch64) won’t work (and in fact will likely break something). The version you find in JetPack/SDKM is the one intended to work with the Jetson.
It is important to note that when you start
sdkmanager it has options for flashing, and that those require the Jetson to be in recovery mode. However, you don’t need to flash; you can uncheck everything except putting CUDA on your Jetson, and keep the Jetson fully booted (then it uses
ssh to install software).
apt repository was set up, then you can also install on command line without JetPack/SDKM via a standard “
sudo apt-get install ...” command if you know the name of the package. You can for example search via:
apt search cuda
(it might be a long list and it might also add users of cuda, not just drivers)
Anyone know the exact package name for command line CUDA install on the 8 GB Orin Nano?
I had already installed jetpack 5.1.1 and flashed it and for some reason on the sdkmanager even though i had placed my jetson orin nano on recovery mode it is not detected in the sdkmanager. So, can i directly install it from the browser. In the link below i have seen the compatible versions of TensorRT, CuDNN, CUDA for jetpack 5.1.1
i want to confirm that instead of using sdkmanager can i directly download them from browser. And i’m confused about the drivers, so we need drivers to be downloaded only if GPU is not integrated with memory installer.
For reference, L4T is what actually gets flashed (Ubuntu plus NVIDIA drivers), and JetPack/SDK Manager are front end tools to the flash. If you pick one release, you’ve tied to the other’s release. You have a Jetson Orin Nano developer’s kit (I assume it is not using a third party carrier board), so this is what you must check release compatibility against. See:
I don’t see the Orin Nano listed yet. Can someone from NVIDIA comment on the above URL for Orin Nano?
From what I know though, R35.3.1 or newer will work with your Orin Nano (which is what JetPack 5.1.1 produces).
Do note though that you can flash on command line, but USB issues during flash are common. However, the stage of adding optional packages does not use recovery mode; if the unit had been flashed, and then it had moved on to optional packages, the unit would have rebooted, you would have completed the user account first boot setup, and then networking would be used for adding those components over
Is it correct to say that the unit fully boots and you can log in? If you can, then any network can be used. The default is to try to use the wired virtual ethernet over USB. The address for that is
192.168.55.1, but if you want to use wired (or if Wi-Fi is set up, but that’s a whole topic in itself), then you can enter that IP address. If the virtual wired ethernet over USB does not work, then there are a lot of reasons why that might occur (for example, the host might block this as a security issue until you ok it).
If you are using a VM, then there is a very high rate of failures due to VM setup.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.