BusyBeeATais
commented
8 months ago Hi, the gpu images are not public. But you can build the GPU image yourself in the vitis repo. The images for Vitis-AI 3.5 and 3.0 work. For the versions 2.0 and 2.5 you have to modify the base image. Here is a link to docker_build.sh https://github.com/Xilinx/Vitis-AI/tree/master/docker
To the GPU thematics, there you have to install the appropriate CUDA driver. Here is a link to the installation you need: https://xilinx.github.io/Vitis-AI/3.5/html/docs/install/install.html (Here is also something about the docker build script)
If you don't have an Ubuntu 20.04 system as host you can build one with wsl and gpu pass trough. Here is a tutorial: https://medium.com/htc-research-engineering-blog/nvidia-docker-on-wsl2-f891dfe34ab
Use Windows Terminal and ubuntu 20.04 to use the tools. Found a complete Tutorial on this issue: https://def-var.net/project/kira-kv260-part-ii-vitis-ai-workflow-on-windows-with-wsl-and-gpu-pass-through/
I am facing this error while launching the docker images Error response from daemon: pull access denied for xilinx/vitis-ai-tensorflow2-gpu, repository does not exist or may require 'docker login': denied: requested access to the resource is denied docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]].