Exploring AMD’s Ambitious ROCm Initiative

123

The open source ecosystem is a vast world of free tools and interacting components: drivers, APIs, compilers, and programming languages. The story of Linux and the open source movement has always been about building this ecosystem and completing the pieces of the puzzle to create a flexible, versatile, and all-free computing environment.

For many common scenarios, this constellation of free components is nearly complete; however, the crucial area of high-performance computing has had to contend with some non-free software components. In particular, the rise of the Graphics Processing Unit (GPU) has complicated a toolchain that had previously been focused on traditional CPU-based computing. Languages like CUDA evolved as a means for integrating the GPU into conventional C++ programming; however, CUDA was never really envisioned as a universal solution, and it is designed to support the GPU hardware of a single vendor. Other solutions, such as OpenCL, embrace the concept of open source but do not support a full range of programming alternatives.

As a leading vendor of both GPU and CPU technologies, AMD has taken up the challenge of bringing free, flexible, cross-platform, and language-independent computing to the GPU-accelerated HPC space. The result of this ambitious effort is the Radeon Open Compute Ecosystem (ROCm) platform, which AMD describes as “the first open-source HPC/Hyperscale-class platform for GPU computing that’s also programming-language independent.”

Read more at ADMIN