Skip to content

The Riallto Open Source Project from AMD

License

Notifications You must be signed in to change notification settings

jgmelber/Riallto

 
 

Repository files navigation

Riallto

Riallto is an open source project that aims to bridge the gap between newcomers and performance ninjas in AMD's latest AI Engine technology. If you have a machine with an AMD processor with Ryzen AI, you have access to an NPU (Neural Processing Unit), a dedicated neural network accelerator.

The educational examples and software in this repository will enable you to interact with the NPU and learn how to write custom applications with this novel accelerator.

Note that NPUs can also be referred to as Inference Processing Units, or IPUs. You may see the NPU referred to as an "IPU" in some Ryzen AI documentation. Once the Ryzen AI driver is installed on your computer, you will see a device named "IPU" in the Windows Device Manager."

Quick Start

The Riallto installer is separated into 'Lite' and 'Full' installation versions. The 'Lite' version has all the essential packages required for the Riallto and ONNX runtimes on the NPU to get started with prebuilt vision applications. The 'Full' version enables developers to write their own custom applications, which requires additional tools.

The steps below will allow you to get quickly started using the 'Lite' version of Riallto. For details on requirements and installation for the 'Full' version, follow the instructions in the Install Riallto guide.

  • Download and unzip the the v10.1109.8.100 IPU driver. You will need to create an AMD account if you do not have one.
  • Open a Windows Command Prompt, or Powershell, as administrator.
  • Change directory to the unzipped IPU driver directory, then run the amd_install_kipudrv.bat file. Press Enter if prompted.
  • Confirm the IPU driver has installed correctly by running the following command in a Powershell terminal and checking the output matches that of the figure below:
(Get-WmiObject -Class Win32_PnPSignedDriver | Where-Object { $_.DeviceName -eq "AMD IPU Device" }).DriverVersion

IPU driver dialog

  • Download and unzip the Riallto installer.
  • Navigate to the installer.exe using the File Explorer and run as administrator, making sure the 'Lite' install option is selected when prompted, as shown in the figure below.

Riallto installer options

Once the installation process has finished a Jupyter Lab session will open in your web browser, allowing you to start running the Riallto notebooks immediately. The notebooks can be opened by selecting them from the file browser pane on the left hand side of the Jupyter Lab session.

The installer places a shortcut on the desktop, which allows you to launch Riallto anytime after the installation process.

Note that the notebooks from sections 1, 2, 3, and 5 can be run using the 'Lite' install option. A 'Full' install is required to run the notebooks from section 4. See instructions in the Install Riallto guide for more details.

Contribute

Contributions to this repository are welcome. Please refer to CONTRIBUTING.md for how to improve Riallto.

Support

For support, feel free to ask questions on the GitHub discussions page.

Licenses

MIT License

Open Source Components

License and Copyrights Info TAR/GZIP

Open Components Source Code TAR/GZIP

About

The Riallto Open Source Project from AMD

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Jupyter Notebook 50.0%
  • HTML 46.5%
  • Python 2.0%
  • MLIR 1.0%
  • C++ 0.3%
  • PowerShell 0.2%