AMD shared new HPC-focused performance data for the Instinct MI430X GPU. The new accelerator comes with a performance class that exceeds 200 TFLOPS in native FP64 compute. This value stands out as one of AMD’s strongest moves in the double-precision processing side, which is critical for scientific simulations, modelling, high-precision computing tasks and artificial intelligence-supported research infrastructures. AMD MI430X brings HPC and artificial intelligence in the same package in supercomputers.
AMD Instinct MI430X is positioned especially for high-precision HPC workloads in the company’s new generation Instinct family. The new GPU is built on the next generation AMD CDNA architecture and offers 19.6 TB/s memory bandwidth with 432 GB HBM4 memory. This structure stands out as one of the key elements that reduce the memory bottleneck in both large-scale scientific simulations and artificial intelligence training processes.
According to the shared data, MI430X offers over 200 TFLOPS performance in local FP64 vector calculation. In NVIDIA’s Vera Rubin NVL72 technical table, the native FP64 value for a single Rubin GPU is listed as 33 TFLOPS, and the Tensor Core-based FP64 DGEMM emulation value is listed as 200 TFLOPS. That’s why the 6-fold difference that AMD highlights is based on the native FP64 vector computing benchmark, especially on the classic HPC side.
The distinction is important here. On the artificial intelligence side, FP4, FP6, FP8, BF16 and similar low-precision formats stand out in large model training and inference works. However, FP64 still holds a critical place in climate modelling, fluid dynamics, nuclear engineering, materials science and physics-based simulations that require high accuracy. Therefore, MI430X is positioned not only as an artificial intelligence accelerator, but also as an HPC-focused GPU that combines artificial intelligence and classical supercomputer workloads on the same platform.
One of the first large systems where MI430X will be used will be the Discovery supercomputer to be installed at Oak Ridge National Laboratory. Discovery will use AMD Instinct MI430X GPUs and new generation AMD EPYC processors together. ORNL’s schedule includes delivery of the system in 2028 and user operations to begin in 2029. Discovery; It will be used for scientific research in areas such as energy, biology, advanced materials, national security and production innovation.
On the European side, the Alice Recoque system will be the second major supercomputer project, where MI430X stands out. The France-based system will be put into operation in cooperation with GENCI and CEA operation. Alice Recoque will offer exascale class performance with next generation AMD EPYC CPUs and AMD Instinct MI430X GPUs. It was announced that the system will exceed 1 exaflop on the HPL side.
Comments
You can write your views about this story. Comments may be moderated according to site settings.