AMD Instinct MI430X targets high-performance computing and AI with 432 GB HBM4

AMD has unveiled the Instinct MI430X, a data center accelerator positioned at the intersection of high-performance computing and Artificial Intelligence. Deployments are already underway at two large research facilities: the discovery supercomputer at Oak Ridge National Laboratory and the Alice Recoque exascale system in Europe. Both installations will pair the MI430X with AMD’s upcoming EPYC ‘Venice’ processors. By launching directly into production systems rather than a staged preview, AMD presents the MI430X as the immediate successor to its earlier Instinct lineup and as a core component for organizations seeking self-reliant, open Artificial Intelligence infrastructure.

The MI430X addresses memory-constrained workloads with a substantial 432 gigabytes of HBM4 memory and roughly 19.6 terabytes per second of bandwidth. Those specifications push the accelerator toward a near-memory-compute role for very large models, including trillion-parameter workloads. The chip is built on AMD’s next-generation CDNA architecture and retains hardware-level double-precision FP64 support to serve traditional scientific computing needs. At the same time, it adds native FP8 and FP4 precision modes to improve efficiency for Artificial Intelligence training and inference.

Design priorities extend beyond raw throughput to sustained efficiency. AMD states the MI430X emphasizes performance-per-watt improvements to suit exascale environments where power efficiency is critical for long-running simulations and large-scale model training. With production deployments already in place at Oak Ridge and within a European exascale system, the MI430X is positioned as a practical option for institutions balancing scientific computing precision and the scaling demands of modern Artificial Intelligence workloads.