AMD at a press conference announced the ROCm technology that facilitates the development of applications for high-performance computing.
AMD's latest offering is AMD ROCm , a new initiative for developers creating applications for the high-performance computing market. It will be primarily based on the open-source OpenGL , but will also support CUDA using a software translator and the Heterogeneous-Compute Interface for Portability compiler. Exporting will also be supported in ISO C++ and ANACONDA , and it will offer maximum utilization for OpenUCX.
Radeon Open Compute Platform (ROCm) is based on AMD's Boltzman platform, but with some modifications. The initiative arose from a number of needs identified by Boltzman developers, the most important being that it be completely open source. This allows for maximizing the platform's potential by implementing advanced compatibility between the leading coding programs used in this field.
It features support for Native 16 Float Instructions , along with highly automated, high-performance conversion. AMD's goal is for its system to become a very accessible option for developers. Thanks to this tool, all the necessary tasks can be performed on platforms that use open-source code. Additionally, it will be compatible with platforms that use proprietary software for high-level computing functions, such as deep learning machines.
The hardware compatible with this initiative includes Radeon WX4100, 5100, and 7100 graphics cards , as well as desktop cards such as the Radeon RX460, RX470, and RX480. It will also be compatible with previous models, such as the R9 390 Nano and FirePro. It will be integrated into the AMD64 Support ecosystem , ARM AArch64, IMP OpenPower, GenZ, CCIX, and OpenCapi.
Source: TechPowerUp
