No driver version after reflashig Jetson AGX Orin developer kit

Hi,
I reflashed a Jetson AGX Orin 64 Gb developer kit with a 22.04 Ubuntu version. After the flashing was over, i ran “nvidia-smi” in the terminal (cf. picture).
There is no driver version (N/A) on the screen. I don’t understand because there was no error during the flashing.
I have reflashed another time 1 month ago but with a 20.04 Ubuntu version and I had the driver version.
Is the Ubuntu version the cause of this problem?

nvidia-smi is for discrete GPUs (dGPU) on the PCI bus. Jetsons have an integrated GPU (iGPU) wired directly to the memory controller (nvidia-smi cannot find the GPU because there is none on the PCI bus). The drivers and software which come with JetPack/SDK Manager are compatible with the iGPU. I couldn’t answer about specific software, but if you are trying to make some specific software work, then you should ask here and someone familiar with it could tell you about how to get there.

Ok, thank you for your answer.
I understand for nvidia-smi for not display all data about the GPU but does that mean there is a graphic driver installed? When I go to additional drivers in settings, there is no additional driver and no selected driver.
I want to run some AI scripts on this Jetson with pytorch but when I print torch.cuda.is_available(), I have false.
I precise I have installed the compatible cuda version.

There is a graphic driver installed, but nvidia-smi is incapable of seeing it (only PCI devices can be seen and there are no PCI GPUs; it isn’t about finding the driver, it is about finding a driver attached to hardware).

The program which requires nvidia-smi is what must change. This of course can be difficult, but there are often substitutes for a given GPU-based program on an iGPU via JetPack. These are alternate versions that work with iGPU. If you could describe what program needs this, then there might be some advice possible. Without knowing anything except your program relies on nvidia-smi the only answer is that the driver and GPU won’t be visible.

There is no program which need nvidia-smi, I used this to see the data about the GPU.
My need is to use the GPU of the AGX ORIN for running computer vision scripts.
If I understand, the procedure for running AI script on classic GPU and iGPU is different? How can I use iGPU with pytorch?

nvidia-smi will not give you a valid answer on a Jetson. If you are interested in finding out if there is a hardware accelerated driver for OpenGL (or OpenGLES?), then you might install mesa-utils (“sudo apt-get install mesa-utils”), and run this command from the GUI:
glxinfo | egrep -i '(nvidia|version)'

Running hardware accelerated graphical applications is no different on an iGPU versus a dGPU. What differs is running a CUDA program or other program which talks to the GPU and are not graphical. APIs like OpenGL and Vulkan are handled by the libraries that are installed, and this will abstract the GPU hardware (the end application won’t care if it is only software rendered, rendered by an NVIDIA iGPU or dGPU, or another brand of GPU). CUDA and programs which directly use the GPU are different.

Incidentally, the X server is not actually a graphics program the way people think it is. The X server itself is an API and ABI that abstracts a buffer in a defined way. It just happens that the buffer is usually a video card’s memory and the other APIs (such as OpenGL and display managers) are what make this useful as a graphical interface. There are ways which some CUDA applications run through the X server without the running software producing graphics.

Ok so If I want to check GPU data, I have to install mesa-utils, thanks !
But I don’t understand why pytorch don’'t work with correct CUDA version and why it doesn’t see the iGPU.
Is it possible that there is no driver graphic installed on my jetson? When I go to additional drivers in the setting app, there is no proprietary drivers listed.

Hi,

iGPU’s driver is integrated into the OS so it won’t show up like the desktop version.
Since we started to support nvidia-smi from JetPack 6, the function and information are also limited.

But if you can see the Orin listed in the nvidia-smi output, then the environment is good.

Thanks.

Ok thank you !

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.