AMD positions ROCm as an open software stack optimized for artificial intelligence workloads across AMD GPUs, describing support for open frameworks, models, and tools. The page highlights simplified model development, noting that 2 million Hugging Face models run out of the box, and cites upstream support for leading frameworks such as TensorFlow and PyTorch. For enterprise operations, AMD introduces the ROCm Operations Platform to help deploy at scale with Docker, Singularity, Kubernetes, and Slurm, while the AMD Infinity Hub offers containers and deployment guides for artificial intelligence applications.
The ecosystem section outlines solution areas where ROCm is applied, including computer programming, content generation, customer chatbots, data analysis, developer tools, employee copilots, knowledge management, process automation, recommendation systems, report generation, and summarization. An “AI Models” list references prominent models such as Bloom, Cohere Command R+, DBRX, Deepseek, Falcon LLM, Gemma, GPT-4, Llama, Mistral and Mistral MoE, Phi, Stable Diffusion, StarCoder, Qwen, QwQ, and Yi. AMD also shares partnerships and case studies with Lamini, Mosaic ML, Hugging Face, and the PyTorch Foundation, emphasizing integration of ROCm into open-source libraries and efforts to enable high-performance finetuning and training on AMD Instinct accelerators.
ROCm’s software components span accelerated artificial intelligence, math, and communication libraries, including AITER, Composable Kernel, hipBLAS, hipCUB, hipFFT, hipRAND, hipSOLVER, hipSPARSE, hipSPARSELt, hipTensor, MIGraphX, MIOpen, MIVisionX, RCCL, rocAL, rocDecode, ROCm Performance Primitives, rocThrust, and rocWMMA. Tools and compilers range from AMD SMI and HIPIFY to ROCm Bandwidth Test, ROCm Compute Profiler, ROCm Data Center Tool, ROCgdb, ROCm Info, ROCm Systems Profiler, ROCm Validation Suite, ROCProfiler-SDK, and TransferBench. Operating system and runtime references include CentOS, RHEL, SLES, and Ubuntu. The page points to AMD Instinct accelerators and notes that certain ROCm features support select AMD Radeon graphics cards, with supported GPU details available in ROCm documentation.
Developer resources include the ROCm Developer Hub, access to AMD Developer Cloud to build models on AMD Instinct accelerators, and guidance for machine learning development on AMD Radeon GPUs using ROCm starting with version 5.7 on Ubuntu Linux for the Radeon 7900 series. A “Latest News” section curates recent technical articles and blogs spanning benchmarking, medical imaging models, generative artificial intelligence tooling, retrieval augmented generation pipelines, and coding agents, underscoring active updates across the ROCm ecosystem.
