AMD has introduced a guide for running OpenClaw locally on Windows using two hardware setups: “RyzenClaw” and “RadeonClaw.” These configurations are designed to handle AI workloads without relying on cloud services. This reflects AMD’s new vision—an “Agent Computer” approach. They believe people and businesses should have complete control over their data and AI abilities.
The RyzenClaw setup uses a Ryzen AI Max+ system with 128 GB of unified memory. AMD suggests setting aside 96 GB for variable graphics memory, which helps with performance. When testing with the Qwen 3.5 35B A3B model, this setup processes about 45 tokens per second and can handle 10,000 input tokens in around 19.5 seconds. It also supports six AI agents at once.
On the other hand, RadeonClaw features the Radeon AI PRO R9700. It’s quicker, reaching about 120 tokens per second, and processing the same 10,000 input tokens in just 4.4 seconds. However, it has a smaller context window of 190K tokens and can only manage two agents at a time.
But, getting into local AI solutions isn’t budget-friendly. The RyzenClaw setup can cost around $2,700, while the RadeonClaw configuration starts at $1,299 for the GPU alone. This makes these options more suited for early adopters and developers rather than everyday users.
Interestingly, as AI technology advances, a growing number of individuals and companies are prioritizing local processing for security and cost concerns. According to a recent survey by Gartner, 65% of IT leaders are exploring on-premises AI solutions due to data ownership and security worries.
As the conversation around AI shifts toward personal control and security, these AMD setups represent a significant step. They are timely responses to a broader trend of prioritizing privacy and efficiency in AI.
For those looking to dive deeper into the advancements in AI hardware, you might find insights from companies like NVIDIA or Intel helpful, as they also develop components for AI applications (source: Gartner Survey).
