AMD Instinct MI430X GPU Delivers Record 200 TFLOPs FP64 Performance to Power Global Exascale Supercomputing and AI for Science Workflows
AMD showcased the technical capabilities of its upcoming Instinct MI430X GPU during the HPC User Forum held in Austin Texas. According to official company disclosures this new accelerator is designed to deliver more than 200 TFLOPs of native FP64 performance. This metric establishes a new benchmark for scientific simulation and modeling. The MI430X is projected to provide more than 6 times the FP64 throughput compared to the next generation NVIDIA Rubin architecture making it the most powerful GPU ever built for high precision numerical tasks.
The engineering focus for the MI430X centers on the convergence of artificial intelligence and high performance computing. High fidelity simulations in fields such as nuclear engineering and fluid dynamics are becoming the primary source of training data for advanced AI models. As scientific teams move toward automated laboratories and closed loop discovery the quality of the data determines the accuracy of the resulting AI surrogate models. AMD designed the MI430X to ensure that the infrastructure produces correct answers at scale by maintaining strict numerical fidelity during massive computational workloads.
AMD confirmed that the MI430X will power several major global supercomputer projects in the coming years. One of the primary deployments is the Discovery system at the Oak Ridge National Laboratory in the USA. Planned for a 2028 launch under the Genesis Mission the Discovery system will function as a flagship AI factory for the Department of Energy. This supercomputer will utilize MI430X GPUs alongside next generation AMD EPYC CPUs to advance research in biology and advanced materials. Discovery is expected to be one of the 1st large scale AI factory installations in the United States using an open American AI stack.
In Europe the MI430X will be integrated into the Alice Recoque system. This exascale class supercomputer is being developed in cooperation with the Commissariat à l énergie atomique et aux énergies alternatives in France. The Alice Recoque system is designed to deliver more than 1 exaflop of HPL performance. It will focus on pressing societal and industrial challenges by combining data analysis with large scale physical simulations. The deployment emphasizes energy efficiency and scientific throughput across a sovereign European computing ecosystem.
The introduction of the MI430X serves as a foundation for the emerging AI for science ecosystem. By providing leadership in both FP64 precision and low precision AI capabilities in a single package AMD aims to streamline scientific workflows. High precision data is foundational for capturing the true structure of underlying science in climate research and materials engineering. The MI430X represents a step change in capability for centers where numerical fidelity is mission critical. This development ensures that future AI models are trained on stable and accurate data rather than low precision approximations that could inherit structural limitations.
