AMD Unveils RyzenClaw and RadeonClaw for Local AI Agent Execution on PCs

AMD Charts the Future of Personal Computer AI Capabilities

In a significant move underscoring its commitment to enhancing personal computing power in the artificial intelligence domain, AMD has announced the launch of innovative technologies under the names RyzenClaw and RadeonClaw. These solutions aim to enable users, particularly developers and researchers, to run AI agents locally on their devices, opening new avenues for data processing without the need for constant cloud connectivity.

Specialized Pathways for Exceptional Performance

AMD offers two distinct pathways to cater to different needs. The RyzenClaw pathway is powered by the Ryzen AI Max+ platform, featuring a substantial 128GB of unified memory, with 96GB allocated for variable graphics processing. This robust setup allows for highly efficient processing of complex AI models. It can run the Qwen 3.5 35B model at a speed of up to 45 tokens per second and accommodate expansive context windows reaching 260,000 tokens. Furthermore, this pathway can process 10,000 tokens in approximately 20 seconds and support the simultaneous operation of up to six AI agents, making it ideal for tasks requiring parallel processing and substantial memory capabilities.

The RadeonClaw pathway, on the other hand, focuses on enhanced graphical performance, utilizing the Radeon AI PRO R9700 card with 32GB of graphics memory. This pathway boasts exceptional token processing speeds, reaching up to 120 tokens per second, and can process 10,000 tokens in just about 4.4 seconds. While it supports a context window of up to 190,000 tokens and allows for two concurrent agents, this pathway delivers outstanding processing speed for tasks that do not require massive unified memory.

Empowering Developers Amidst Market Challenges

Through these technologies, AMD aims to provide powerful tools for developers and companies seeking to explore the potential of local AI, moving away from complete reliance on data centers. The operational environment leverages technologies like WSL2 and LM Studio, facilitating local model execution via the llama.cpp library and supporting local memory indexing and storage without any external connection. This integration ensures data privacy and rapid response times, which are critical factors in many applications.

However, these advanced systems come at a premium price, positioning them within the professional hardware category. The RyzenClaw platform starts at $2,700 USD, while the RadeonClaw card alone costs $1,299. These prices reflect the high technical capabilities and specialized nature of these solutions, indicating they are primarily targeted towards advanced experimentation and professional use cases. The market for local AI for the average consumer is still in its nascent stages.

This move by AMD comes at a time when the importance of personal AI agents is growing, with major companies like OpenAI actively enhancing the capabilities of these agents. Providing powerful, locally accessible platforms may accelerate the pace of innovation and pave the way for new and unexpected applications in the field of artificial intelligence.