AMD is targeting Nvidia in the high-performance computing (HPC) sector with its next-generation accelerator, the Instinct MI430X. According to technical data shared by the company, the MI430X will claim the title of the most powerful FP64 GPU ever developed, reaching a native FP64 processing power of 200 TFLOPs.
AMD also announced that the new accelerator offers up to 6 times higher performance in classic HPC workloads compared to Nvidia’s next-generation Rubin architecture.
200 TFLOPs Native FP64 Performance
While low-precision data formats like FP4, FP6, and FP8 have recently come to the forefront in the artificial intelligence sector, FP64 still holds critical importance in scientific simulations, engineering calculations, and research-oriented supercomputer systems.
According to data shared by AMD, the Instinct MI430X reaches 200 TFLOPs in raw FP64 performance. This value stands out as the highest FP64 computing capacity announced to date among existing HPC GPUs. The company's technical data indicates that NVIDIA Rubin GPUs offer approximately 33 TFLOPs of performance in native FP64 vector computation.
AMD's advanced CDNA architecture, new manufacturing processes, and advanced packaging technologies contribute to the new accelerator achieving this level. The GPU is also stated to come with a large amount of HBM4 memory offering high bandwidth. AMD positions the MI430X as a hybrid solution optimized not only for scientific calculations but also for AI workloads.
The MI430X will be part of the company's next-generation MI400 series. The MI450X, also in the same series, will stand out as AMD's primary AI accelerator.
To be Used in Supercomputers
Furthermore, AMD announced two major supercomputer projects where the MI430X will be utilized. The first of these will be the Discovery supercomputer, to be established at the Oak Ridge National Laboratory in the USA. Scheduled to be operational in 2028, the system aims to strengthen the USA's leadership in artificial intelligence and HPC. The Discovery system will also be one of the flagship platforms of the U.S. Department of Energy (DOE).
The system will be built using numerous AMD Instinct MI430X GPUs and next-generation AMD EPYC processors.
AMD's new accelerators will not be limited to the USA. The company announced that MI430X accelerators will also be used in the next-generation exascale-class supercomputer system called Alice Recoque, being developed in Europe.
0 Comments: