AMD demonstrated a method for running AI agents directly on a PC equipped with Ryzen or Radeon processors using just 128 GB of RAM.

AMD Unveils OpenClaw, an Open AI Agent for Windows

AMD has released a guide for installing and using its own artificial intelligence agent—OpenClaw—which runs on Windows via WSL2 and LM Studio without cloud access. The instructions outline two possible hardware configurations:

What to Expect from Each Configuration
- RyzenClaw
- Uses the Qwen 3.5 35B A3B model.
- Generation speed: ~45 tokens per second.
- Context window: up to 260,000 tokens.
- Supports simultaneous operation of up to six agents.

- RadeonClaw
- Also uses Qwen 3.5 35B A3B but at a higher frequency.
- Generation speed: ~120 tokens per second.
- Processing 10,000 input tokens takes about 4.4 seconds.
- Allows fewer agents to run concurrently compared with RyzenClaw.

Cost
- A mini‑PC based on the Strix Halo (Ryzen AI 300 Max) with 128 GB of memory in the U.S. costs around $2,399.
- The Radeon AI Pro R9700 GPU for workstations is priced at approximately $1,299.99.

Where to Get the Guide
The full step‑by‑step OpenClaw installation plan is available on AMD’s official website. It covers setting up WSL2, configuring LM Studio, and optimizing memory settings for both hardware platforms.