AMD Ryzen™ AI Software 1.4: Features for Next Gen AI PCs

Apr 18, 2025

Introduction

We are excited to announce the AMD Ryzen AI 1.4 software release for AMD AI PCs.  Ryzen AI 1.4 software highlights new capabilities including additional hardware, model support, and new ease of use developer tools. Ryzen AI 1.4 provides support for state-of-the-art Large Language Models (LLMs), Natural Language Processing (NLP) models, and Convolutional Neural Networks (CNNs). For developers and end users, it enables seamless compilation and deployment of models in INT8 or BF16 configurations, providing flexibility to build applications without altering their environment. Additionally, Ryzen AI 1.4 introduces new developer tools like Digest AI, Lemonade SDK, GAIA, and TurnkeyML, making it easier to work with SOTA models and accelerate AI application deployment. This update enhances model performance, enables new experiences, and offers a simplified approach to deploying AI on AMD PCs, ensuring developers have the tools needed to innovate in the rapidly evolving field of AI.

Unified Support for LLMs, NLP and CNN Models in a Single Release Package

Ryzen AI 1.4 offers support for LLM (Deepseek, Gemma, QWEN, etc.), NLP (BERT, Embedding Models families) and CNN models. It offers a unified installer allowing end users to seamlessly compile their CNN and NLP models in either INT8 or BF16 configurations, as well as deploying ready to run LLMs in hybrid and NPU only execution modes. This streamlined approach simplifies the development process, offering ease of use and improving developer efficiency.

Quantization and Compilation for BF16 CNN and NLP Models on Windows

Ryzen AI provides the tools and recipes for quantization and compilation with BF16 data type on Windows delivering improved performance without accuracy loss. BF16 quantization uses 16 bits with a floating-point format that preserves much of the dynamic range of full 32-bit floating-point (FP32) computations, making it more suitable for models that rely on small or very large numbers, such as deep neural networks that require high dynamic range for weights and activations.

Public Release of LLM Hybrid ONNX Runtime Gen AI (OGA) Flow 

The 1.4 release enables developers to run LLMs in hybrid or NPU-only execution mode. In hybrid execution mode, developers can deploy LLMs on NPU as well as iGPU. Hybrid execution mode optimally partitions the model such that different operations are scheduled on NPU and  iGPU. This minimizes time-to-first token (TTFT) in the prefill-phase and maximizes token generation (tokens per second, TPS) in the decode phase.

Ryzen AI 1.4 also provides support for an OGA-based flow where the iGPU is not solicited and where the compute-intensive operations are exclusively offloaded to the NPU. Both methods offer a comprehensive and simplified approach to building a Ryzen AI model from a trained foundational or finetuned model, giving end users a powerful and accessible way to quickly deploy their model for real-world AI use cases on AI PCs. List of models which can be deployed using NPU1 and Hybrid Mode2.

New Ease of Use Open-Source Developer Tools: Digest AI, Lemonade, GAIA and TurnkeyML

AMD is excited to introduce various open-source tools to help developers run LLMs on AMD AI PCs. 

Digest AI is a powerful and user-friendly tool that can significantly enhance your machine learning development. By providing detailed model analysis, multi-model support, and integration with the Hugging Face hub, Digest allows users to analyze individual models to extract parameters, FLOPs, meta parameters, FLOPs, meta data, IO tensors and more. It also allows you to compare high level statistics of multiple model's side by side and lets you save all extracted statistics into a comprehensive report for easy sharing and collaboration

Lemonade SDK provides a set of high-level APIs and tools that make it easy to experiment with LLMs using the hybrid execution mode, which uses the integrated GPU and the NPU as well as in NPU-only mode. This allows you to try out different models, prompt them, and benchmark them to measure inference speed, time to first token, and tokens per second. You can also measure task performance with accuracy tests like MMLU. It is a multi-vendor, multi-backend SDK that supports Runtime Gen AI, Llama.cpp and Hugging Face Transformers, allowing you to compare across different backends. Additionally, it provides support for a CPU backend that will work on other AMD laptops

GAIA is an open-source solution designed for the quick setup and execution of generative AI applications on local PC hardware. It supports basic chat, RAG enhanced applications and specialized agents as well as used the AMD NPU and iGPU for hybrid acceleration. GAIA also provides both a command-line interface and a graphical user interface option for easy interaction with models and agents or allows you to easily build and integrate your own agents and use-cases.

AMD, in collaboration with the ONNX community, is excited to introduce TurnkeyML, an open-source toolchain designed to improve the way we handle AI models for inferencing. TurnkeyML is a comprehensive framework that streamlines the process of ingesting any open-source PyTorch model, optimizing them, and executing them across a diverse set of hardware targets. 

Summary

We are excited to announce continuous improvements to Ryzen AI software and encourage developers to provide feedback on how these updates enhance their experience. For a comprehensive overview of the new features and improvements in the 1.4 software release, check out the official release notes.

Stay at the forefront of AI innovation by signing up for the latest Ryzen AI software updates to ensure you have access to the tools and resources that will help you push the boundaries of what’s possible on AI PCs. 

Additional resources

Download Ryzen AI Software
Share:

Article By

Profile Image of Anshul Gupta

Anshul Gupta