ROCm
also: Radeon Open Compute, HIP
ROCm (Radeon Open Compute) is AMD's open-source platform for GPU computing that enables developers to use AMD graphics cards for general-purpose computing tasks like machine learning and scientific simulations.
ROCm provides a complete software stack including compilers, libraries, and runtime support for programming AMD GPUs. It allows developers to write heterogeneous applications that offload compute-intensive workloads to AMD Radeon GPUs, similar to how NVIDIA's CUDA platform works with GeForce and Tesla cards.
ROCm supports popular programming models including HIP (Heterogeneous-compute Interface for Portability), OpenCL, and standard languages like C++ and Python. The HIP API is particularly designed for portability, allowing CUDA code to be adapted for AMD GPUs with minimal changes.
Example use case: A machine learning researcher can use ROCm with frameworks like PyTorch or TensorFlow to train neural networks on AMD GPUs, accessing them through libraries like rocBLAS for linear algebra or MIOpen for deep learning primitives.