IN5050 - NVIDIA GeForce GPU Resources & FAQ

 

A page for resources and frequently asked questions for the Jetson AGX Xavier machines. If you have any other questions, please send an email to in5050@ifi.uio.no

Remember that you might have to SSH into login.ifi.uio.no to access the login machine for IN5050 (129.240.66.57). Following this guide, it is possible to set up SSH and SSHFS from external machines.

A username and password have been provided to all groups. The following table gives an overview of the status of the ARM machines with GPU at IFI:

GPU Status
Computer GPU GPU Core Memory Multiprocessors (CUDA Cores) Compute Capability Status
tegra-1 NVIDIA Tegra Xavier GV11B 16 GB 8 (512) 7.2 Up
tegra-2 NVIDIA Tegra Xavier GV11B 16 GB 8 (512) 7.2 Up
tegra-3 NVIDIA Tegra Xavier GV11B 32 GB 8 (512) 7.2 Up
tegra-4 NVIDIA Tegra Xavier GV11B 32 GB 8 (512) 7.2 Up

 

GPU Programming Resources

Paper on optimizing the Motion JPEG encoder for Cell and GPU (access from UiO)

NVIDIA CUDA Toolkit 11.4 Documentation

Application Note - CUDA for Tegra

 

Frequently Asked Questions

 

Q: Can I use my own GPU?

A: Yes, you can. However, we do not recommend this. The program has to compile and run on the Jetson AGX Xavier Development Kits. Your CPU code should be optimized for 64-bit ARMv8.2, and the GPU code should be optimized for the Volta/Turing architecture (Compute 7.x).

 

Q: Do we have any video source files to test with? 

A: The video source files are stored in /mnt/sdcard.

 

Q: My video is broken, are there some tools to analyze the video?

A: Yes, there are! Try out YUView, which is an open-source and cross-platform YUV player and analysis tool.

 

Q: What software do I need if I want to run on my own GPU?

A: Here at IFI, we are running Ubuntu 20.04 LTS (ARM 64-bit) with CUDA 11.4 from NVIDIA. You have to download both a CUDA-certified driver and CUDA 11.4.4 toolkit from NVIDIA. The CUDA SDK is optional, but it contains several useful functions.

 

Q: Are there any differences between the GPU on a Tegra, and a GPU connected with PCIe (dGPU)?

A: Yes! This is something you should be aware of, especially when writing code to transfer data between the host (CPU) and device (GPU) on the Tegra, the integrated GPU (iGPU) share the memory with the CPU. This application note explains some crucial differences.