Running PyTorch on ZYNQ Ultrascale+


PyTorch provides libraries which can be used in C++ applications. That is, whatever you do with PyTorch Python libraries, you can also do in your own custom C++ application and link your application with PyTorch libraries. This package provides you with all you need to build and run your C++ application that uses PyTorch on your ZYNQ Ultrascale+ device PS. Also it provides clear and easy-to-do steps for installing and running Pytorch Python libraries on ZYNQ Ultrascale+ devices.

 

Key facts and contents

  • Packaga provides all the ready-to-use PyTorch libraries and build scripts which allow you to use PyTorch routines within your C++ application and run your application on ZYNQ Ultrascale+ PS.

  • You can build your application on your local Linux machine using provided build scripts and then port it into your Zynq Ultrascale+ board.

  • Two Petalinux versions are supported: 2022.2 and 2024.2. Complete set of files, along with very easy to use scripts and instructions are provided.

  • Two PyTorch versions are supported: 1.13 (which can be used with Petalinux 2022.2) and 2.5 (which can be used with Petalinux 2024.2)

  • Package provides all steps necessary for building PetaLinux along with Python and install Pytorch Python libraries in Petalinux and using them on the board.

  • Package provides all steps necessary for building Pytorch for ZYNQ Ultrascale+ target from source code.

  • Ready-to-use SD card images for ZCU102, ZCU104 and ZCU106 boards with complete Pytorch packages installed, and all Pytorch libraries built from source code,

    along with scripts for building and linking your custom application with Pytorch support is included in the package.

  • Design examles showing how custom Pytorch applications can collaborate with FPGA logic (PL) to collect data and perform inference using trained model and shared FPGA fabric (PL) data are included. All design steps are described in detail.


 

Example Design(s)

The following Figure shows the first practical example application included in the package.


   

AD7616 ADC interface IP is included in the package. Design runs Petalinux on ZYNQ PS and shows how one can use AXI DMA to receive ADC sample data from ADC7616 IP and put the data in DRAM. A custom C++ application with support for Pytorch libraries then uses the data on DRAM along with a trained model to identify desired patterns in each of the incoming digitized 16 analog input signals.

Kernel level driver code is included in the package. Example code shows how we cast the data (copied by AXI DMA inside DRAM) into a tensor and how we use that data for inference using our model.


Target FPGA families

Any ZYNQ Ultrascale+ device. No other requirement is needed.


Support

The package comes with email based support. During your using of the package, whenever you have questions or doubts about the provided contents and examples, you simply write us an email and it will be answered as soon as possible.


 

Pricing

Index Package content Price (euros)
1 Pakcage containing all required files and scripts to run Pytorch on ZYNQ ultrascale+, build and run C++ applications with PyTorch libraries, example designs, and support. 990
 

If you are a student then there exists a 10% discount on the above prices for you.

 

Licensing

The license of the package asks the customer (individual, company or research lab) to pay attention that the entire provided content is for customer's own personal educational use only.

 

Questions?

Might you have any questions, please kindly look at the contacts menu and write me an email. It will be answered as soon as possible.

 

Online payment with credit card

If you would like to do the payment online using your credit card, you can do it here. You will receive official invoive along with your package content after payment.