AMD has unveiled the initial version of its new Instinct MI430X graphics accelerator, part of the Instinct MI400 series. As reported by IT Home, this GPU has already been announced for several supercomputers. This is reported by news source.
Unlike the MI455X model, designed for training and optimizing large-scale AI models, the MI430X is tailored for both AI tasks and traditional high-performance computing (HPC). This distinction is clearly reflected in the MI430X's support for the high-precision FP64 data format.
According to AMD, the FP64 computing power of the MI430X is expected to exceed 200 TFLOPs. This figure is six times higher than Nvidia's Rubin architecture, making it the most powerful GPU in history for FP64 operations.
Company representatives emphasize that computational precision is crucial for fields such as climatology, materials science, nuclear physics, and fluid mechanics. High-precision data formats like FP64 serve as a solid foundation for AI-driven scientific research.
Read “Zamin” on Telegram!
