AMD targets its software gap with ROCm 10, putting its own skills into Claude and Cursor
The release adds a command-line tool, an AMD Skills pack that plugs into coding agents, and the open-source Hyperloom. AMD reports a 3.3x inference speedup over ROCm 7.0.
AMD has announced version 10 of ROCm, its GPU software stack. The weak point critics have long pointed to at AMD was software rather than silicon, and this release aims straight at it.
Three pieces are new. ROCm CLI pulls installation and environment management into a single command-line tool. AMD Skills packages the company's own ROCm knowledge as a skill set that plugs into coding agents such as Claude Code, Cursor and Codex, so a developer does not have to already know AMD-specific tuning to get it out of the agent. Hyperloom is a distributed execution layer released as open source.
On performance, AMD reports that ROCm 10 is 3.3x faster at inference and 2.4x faster at training than ROCm 7.0. These are the company's own measurements and there is no independent verification yet. It is also worth noting that the comparison is against AMD's own earlier release, not against competing hardware.
AMD has also gathered the whole stack under the name ROCm.AI, moving documentation and downloads to a single address.
Details are in the ROCm documentation.
