ROCm

Overview

ROCm (Radeon Open Compute) is AMD's open-source software stack for GPU computing, encompassing libraries, compilers, tools, and runtimes for the Instinct accelerator line. ROCm is upstreamed to all major AI frameworks (PyTorch) and compilers (Triton), enabling workload portability from legacy infrastructure to AMD hardware without proprietary framework locks. As of 2025, ROCm provides functional and performance support for over one million models on Hugging Face, with day-zero support for frontier models upon their release. AMD has accelerated the ROCm release cadence as part of its four-pillar AI strategy emphasizing openness and ease of migration.