This is an associated final project for CENG3430 - Rapid Prototyping of Digital Systems @ CUHK CSE. And this project is still ongoing and under maintenance after the course finished as we are hoping to create a more polished product and make it a reasearch.

• @Jellyfish227: PL, DPU and PS-PL integration
• @sennettcheng: PS

• CPU: >= 2GHz, >= 8 cores
• GPU: NVIDIA CUDA-enabled GPU only. Having one is strongly recommended, not sure what grade is required, but I just used my old GTX 1660Ti and it seems fine. Model quantization and optimization of Vitis AI could be GPU-accelerated. They also mentioned that you should use GPU-acceleration as long as you have one.
• RAM: >= 8GB (I used 16GB, and I feel like 16GB is bare minimum to run faster)
• Storage: >= 800GB (Actually I don't know how many GB have I used just to download all the softwares and libraries, but they are just fuxxing huge. You know, It's Xilinx practice, HAHA. So for safety, I recommend >= 1TB)

• Ubuntu 20.04 or later (I suggest using the latest LTS version, most of the time it can promise to be supported)
• Windows is less recommended as I found that most of the tools work better or even only work on Linux.
• Please try to make sure that you can physically access the host machine as it is so troublesome to setup remote hardware connection for writing bitstream to the board.
• Note from @Jellyfish227: I used Windows 11, hosting VM with Ubuntu 24.04 LTS. I do so just because my remote desktop client does not support hosting on Linux, so I host the remote RD on Windows so I can RD to the PC with my Mac, but it's not a good idea as the performance is lowkey shxt. If you don't have strong reason to use Windows, I suggest you just use Linux host system.

• Vivado 2024.2 or later (newer version is recommended for supporting the evaluation kit used in the project - ZCU102 / ZCU104)
• Vitis 2024.2 or later
• Vitis AI 3.0 or later
  • Docker
  • If using GPU-acceleration, you will need:
    • CUDA Toolkit
    • NVIDIA Driver
    • NVIDIA Container Toolkit
• Petalinux 2024.2 or later (Petalinux 2024.2 requires)
  • It requires /bin/sh to be linked to /bin/bash, so you need to do the following:

    sudo ln -fs /bin/bash /bin/sh

• ZCU102 / ZCU104
• Or any other Zynq UltraScale+ MPSoC FPGA evaluation kit as long as you can declare your custom platform into an .xsa to let Vitis to be recognized
• ZCU102 / ZCU104 is recommended as the they are native boards from AMD Xilinx, so the supports and guidlines are more complete.

1. Install Ubuntu OS on the host machine : Done on 2025-05-07
2. Install Vivado and Vitis : Done on 2025-05-07
3. Install Petalinux : Done on 2025-05-08 01:32:38
4. Install Docker, CUDA Toolkit, NVIDIA Driver, NVIDIA Container Toolkit : Done on 2025-05-08 01:32:38
5. Install Vitis AI : Done on 2025-05-08 01:18:48
6. Build Docker Container from Xilinx Recipe : Done on 2025-05-08 01:44:05
7. Vitis AI GPU-enabled test : Done on 2025-05-08 01:50:20
8. Vitis AI up and running : Done on 2025-05-08 01:50:20
   • Note: to run Vitis AI, execute the following command:

     docker images # check the vitis ai image repo and tag
     # In my case, it is xilinx/vitis-ai-pytorch-gpu:3.5.0.001-1eed93cde
     cd <Vitis AI dir>
     ./docker_run.sh vitis-ai-<image-repo>:<image-tag>

9. Setup reference design in Vivado workflow system : Done on 2025-05-09 15:57:13
10. TRD Synthesis : Done on 2025-05-09 16:22:01
11. TRD Implementation : Done on 2025-05-09 17:52:09
12. Generate TRD bitstream : Done on 2025-05-09 17:57:31
13. Create PetaLinux project with Vivado workflow
14. (Optional) To adapt with newer version of toolchain, convert .xsa hw description into new System Device Tree(sdt) system with XSCT.
15. Configure PetaLinux project with hw platform configurations (i.e. .xsa(deprecating approach) or dst(our approach)), together with corresponding bsp
16. Configure system image build, including rootfs, kernals, drivers, library and enabled modules and features
17. Include and set bitBake build receipes, like vai-recepies
18. Build PetaLinux project into .elf bit files
19. Package generated images into MCS images(BOOT.BIN, boot.scr, *.gz.u-boot)
20. Package all MCS images into wic image
21. Flash wic image to SD card for boot media