Webinar 2021 ROCm: AMD's platform for GPU computing
From SHARCNETHelp
Jump to navigationJump to search
ROCm (Radeon Open Compute) platform developed by AMD aims to provide a fully open-source alternative to Nvidia's CUDA platform. It incorporates HIP (Heterogeneous-Computing Interface for Portability), a C++ dialect designed to ease conversion of CUDA applications to portable C++ code. It comes with a tool to port CUDA source code automatically to HIP. HIP code can be compiled to run both on NVIDIA and AMD GPUs. A rich set of mathematical libraries is included in ROCm, as well as support for major ML frameworks like TensorFlow and PyTorch. The seminar will introduce ROCm, show some examples of HIP code, and look at performance obtained on NVIDIA and AMD GPUs.