A chip manufacturer just created a consumer product category for AI agents.
AMD published an official guide for running OpenClaw locally on AMD hardware. Two branded configurations, dedicated product pages, the whole playbook.
RyzenClaw vs RadeonClaw
RyzenClaw (Ryzen AI Max+ APU): 128GB unified memory, ~45 tokens/sec on Qwen 3.5 35B, 260K token context, up to 6 concurrent agents.
RadeonClaw (Radeon AI PRO R9700 GPU): Dedicated VRAM, ~120 tokens/sec, processes 10K input tokens in 4.4 seconds.
RyzenClaw is the more interesting config. 128GB unified memory means larger models without a separate GPU. Six concurrent agents on a single chip makes desktop agent swarms possible.
RadeonClaw is the speed play. 120 tokens/sec on a 35B model is production-grade from a single GPU.
AMD vs NVIDIA
This is AMD answering NVIDIA's DGX Spark. Both companies now have branded OpenClaw hardware with dedicated marketing. AMD calls them "Agent Computers," which honestly sounds better than anything NVIDIA named.
Two of the world's biggest chip companies competing over local AI agent hardware. That's not theoretical anymore.
AMD also announced a Developer Cloud with free vLLM-powered OpenClaw inference on AMD silicon. No hardware purchase needed. Classic developer funnel.
Full post: AMD Enters the OpenClaw Hardware Race
Top comments (0)