AMD to enable ROCm on Windows, add support for some gaming Radeon GPUs
AMD’s open software platform for GPU computing is coming to Windows.
The company released the first version of ROCm in 2016. It was initially a research project called Boltzmann Initiative, which was later turned into a programming interface. It was a necessary step with the growing popularity of the NVIDIA CUDA library.
In essence, ROCm (Radeon Open Compute platforM) is a software stack for GPU programming. It works in various domains, serving as a backbone for HPC, AI or OpenCL computing. The stack provides access to several programming models and tools that can more easily accelerate the workload. AMD also provides access to HIP, a C++ runtime that can easily run CUDA code on AMD GPUs.
Today ROCm is successfully deployed in the world’s most powerful supercomputers, such as El Captain or Frontier. Both systems are running Linux operating systems, but such a large deployment is not the same experience as general computing on a single system.
However, AMD is now working on bringing ROCm support to Windows OS. The announcement came from documentation for ROCm 5.6.0 Alpha. This documentation is now hidden behind a login page, so the news about Windows support was likely not meant to go public just yet.
The hardware support list for ROCm Windows is conservative, as developers still work on verifying other GPUs. Interestingly, GPUs such as Radeon RX 6900 XT or RX 6600 are on the list. In fact, even the R9 Fury with HBM memory supports the programming library, but only Linux support is currently listed.
Source: AMD (now behind login)