AMD CEO confirms ROCm will support consumer graphics cards
In the year 2016, AMD unfurled ROCm (Radeon Open Compute Ecosystem), an open software platform designed as an open-source alternative to NVIDIA’s CUDA platform. This platform was crafted with the pursuit of superior flexibility and performance in mind, targeting accelerated computation, and is not confined to any programming language. It allows participants of the machine learning and high-performance computing communities to expedite their code development through a variety of open-source computing languages, compilers, libraries, and redesigned tools. It is ideally suited for large-scale computation and multi-GPU computing, supporting both CDNA and RDNA architecture GPUs.
Reports surfaced in April of this year that the ROCm SDK was poised to make its debut on the Windows operating system, and was concurrently broadening its support for consumer-grade Radeon graphics cards. Recently, Dr. Lisa Su, CEO of AMD, confirmed in response to George Holtz of Tiny Corp that future efforts would be geared towards increasing GPU support in ROCm, incorporating more consumer-grade Radeon graphics cards and collaborating with the community to provide superior support for their users.
Currently, the fully supported hardware on the ROCm compatibility list includes the Instinct MI250X, MI250, MI210, MI100, and MI50, and among the Radeon Pro series, the Radeon Pro W6800 and Radeon Pro V620 are supported. AMD has extended the list to encompass the Radeon RX 6900 XT, Radeon RX 6600, and Radeon R9 Fury. However, only the Radeon R9 Fury boasts complete software-level support from the ROCm platform, with the other two RDNA 2 architecture products receiving partial support.
It’s understood that AMD is tirelessly striving to incorporate RDNA 3 architecture GPUs into ROCm, which would translate to support for the Radeon RX 7000 series of graphics cards. Additionally, AMD intends to unveil a new batch of Instinct computational cards, based on the CDNA 3 architecture, later in the year.