Dear PYNQ users,
DPU-PYNQ now officially supports overlays on 21 boards – which is the most platforms we have seen a design supported on! Additionally, this iteration of DPU-PYNQ supports Vitis AI 2.5.
This release is compatible with:
- Vitis AI 2.5
- Vitis / Vivado 2022.1
- PYNQ v3.0 image
You will notice that some of the overlays have been unverified yet, but since we have the capability of producing a great deal of platforms we decided to share them anyway.
If you are in possession of one of the boards in the unverified overlay list, that has a PYNQ image running on it, you can verify DPU-PYNQ by simply typing in python3 -m pytest --pyargs pynq_dpu after installing DPU-PYNQ on your device. If the tests pass on your board, feel free to let us know that DPU-PYNQ is verified on that board by amending the board support table via a pull request.
| Platform | DPU Architecture | Number of cores | Verified |
|---|---|---|---|
| KR260 SOM | B4096 | 1 | Yes |
| KV260 SOM | B4096 | 1 | Yes |
| Pynq-ZU | B4096 | 1 | Yes |
| RFSoC2x2 | B4096 | 2 | Yes |
| RFSoC4x2 | B4096 | 2 | Yes |
| Ultra96v2 | B1600 | 1 | Yes |
| ZCU104 | B4096 | 2 | Yes |
| ZCU111 | B4096 | 2 | Yes |
| ZCU208 | B4096 | 2 | Yes |
| Genesys ZU-5EV | B4096 | 1 | |
| T1 Telco RFSoC | B4096 | 2 | |
| T1 Telco MPSoc | B4096 | 2 | |
| TySOM-3A-ZU19EG | B4096 | 2 | |
| TySOM-3-ZU7EV | B4096 | 2 | |
| Ultra96v1 | B1600 | 1 | |
| UltraZed-EG | B4096 | 1 | |
| ZCU102 | B4096 | 2 | |
| ZCU106 | B4096 | 2 | |
| ZCU1285 | B4096 | 2 | |
| ZCU216 | B4096 | 2 | |
| ZUBoard-1CG | B800 | 1 |
Some additional notes:
- Most of the supported overlays ship with a B4096 DPU architecture, compatible with the Vitis AI model zoo. This means you can download the .xmodel files from the official Vitis AI repository and use them with DPU-PYNQ overlays.
- For other DPU configurations (e.g. Ultra96v1, Ultra96v2 and ZUBoard-1CG), we have provided custom arch.json files and user-friendly scripts for compiling models from the Vitis AI model zoo.
- The compile.sh script has been replaced with a python-based alternative called compile.py. It serves the same function, but is slightly more user friendly and expandable.
- Thanks to a pull request by one of our community members @haipnh, we now ship with an additional notebook showcasing the yolo_v3 model for object detection.
We always appreciate your engagement and continued contributions. Please use the PYNQ discussion forum to discuss and report any issue.
Thanks,
PYNQ team