AMD Ryzen™ AI Software

The AMD Ryzen processors that include an NPU are compatible with Ryzen AI Software. The list of supported processors

AMD Ryzen AI Software enables AI model deployment on the Ryzen NPU, on the integrated GPU (iGPU), or on a hybrid of NPU and iGPU for balanced performance and efficiency. The AMD ROCm software stack enables AI model training, inference, and optimization on AMD GPUs, including the Ryzen iGPU.

For large models, models that require high throughput or memory bandwidth, or workloads that include heavy graphics or high computing, advanced developers should use ROCm to optimize models and configure and manage the iGPU.

Deploying models on the iGPU requires the AMD ROCm software stack, DirectML, or Lemonade.

• ROCm is a development platform that offers flexibility for model development, training, and optimization along with configuring and managing AMD GPUs.
• DirectML offers a straightforward workflow for converting models to ONNX and optimizing them to deploy on the Ryzen processor’s iGPU.
• Using Lemonade, the open-source multimodal API from AMD, will automatically configure and allocate the CPU, NPU, and iGPU to run many AI applications.

The best applications to develop on a neural processing unit (NPU) are those that require low-power, efficient AI inference. That may include tasks such as image recognition, language processing, and real-time video and audio enhancements. NPUs can run small-to-medium AI models, including LLMs, computer vision, and image generation.

NPU applications are a great way to bring AI capabilities to laptops, desktop PCs, and mobile workstations, especially when prioritizing battery life. NPUs are also ideal for local AI, where users need to keep data private, require low latency, or can’t rely on an internet connection.

Given the overwhelming popularity of PyTorch for AI development, many developers want to know if they can run models developed in PyTorch on Windows PCs. Pre-trained AI models in PyTorch or TensorFlow can be converted to the ONNX format and quantized for compatibility with AMD Ryzen AI Software. This enables AI application development on both Windows and Linux for Ryzen processors with an NPU. 

The best software tools for NPU application development are optimized for specific NPU hardware and can support open-source frameworks and models. This gives developers a combination of performance and flexibility.

AMD Ryzen AI Software includes developer tools to build applications for AMD NPUs. It also provides tools to convert models trained on popular open-source frameworks, including PyTorch and TensorFlow, into optimized ONNX formats for NPU execution.

The best tools to build local AI agents are optimized for NPU and integrated GPU hardware. This helps balance performance and efficiency, so that AI tasks don’t drain PC batteries. They also work with a variety of open-source, pre-trained models so that developers can choose what best fits their use case. Finally, software tools should keep all data local and should not require cloud services.

GAIA is an open-source AMD tool for building local AI agents. It runs 100% locally on AMD Ryzen AI hardware, eliminates cloud costs, and has zero API fees, usage limits, or subscriptions.

The AMD Developer Community forum and the AMD Developer Discord server allow you to connect directly with other developers and AMD experts to discuss questions and help troubleshoot issues. You can also submit a bug on GitHub.