When AMD initially launched the Boltzmann Initiative late last year, it had these three goals in mind:
- New Linux Driver and Runtime Stack optimized for HPC & Ultra-scale class computing,
- Heterogeneous C and C++ compiler which best address the whole system not just a single device
- HIP acknowledging the need for platform choice when utilizing GPU computing API
Since its inception, the Boltzmann Initiative has morphed into the Radeon Open Compute Platform (ROCm). ROCm provides an outlet for open source building blocks for hyper-scale and HPC GPU computing. Inside ROCm you’ll find open source math libraries, deep learning frameworks, highly optimized parallel programming frameworks, a rich compiler foundation and a Graphics Core Next (GCN) headless 64-bit Linux driver that leverages the power of AMD’s Heterogeneous System Architecture (HSA).
Version 1.0 arrived in April, bringing with it support for AMD’s Fiji family of GPUs. ROCm 1.1 launched in June with new capabilities for multi-GPU enabled graphics cards. ROCm 1.2 landed in August with support for Hawaii GPUs and brought with it the beta version the Low Level Virtual Machine (LLVM) 4.0 native GCN compiler.
Today sees the emergence of ROCm 1.3, which brings the official release of the LLVM native compiler and support for AMD’s current Polaris family of 14nm GPUs (Radeon RX 480, RX 470, RX 460). Although AMD didn’t specifically call attention to it, the Polaris-based Radeon Pro WX Series GPUs are also supported by ROCm 1.3, which means that the WX 4100, WX 5100, and WX 7100 all come along for the ride.
Also hopping aboard is support for OpenCL 1.2+. More specifically, ROCm 1.3 support the OpenCL 1.2 runtime along with the OpenCL 2.0 kernel language. According to AMD, a developer preview will launch exactly one month from today on December 14th. AMD will build on ROCm in the future, and we’ll continue to see support for even more powerful graphics hardware in the near future — AMD is scheduled to launch new Vega series GPUs at the start of 2017.