AMD has introduced Instinct MI430X, a data center accelerator designed for convergence between high-performance computing and large-scale AI.
Since the launch of the Instinct MI300 series, AMD has been renewing its AI hardware catalogmaking advances in key areas such as chip architecture, industry-leading component integration and performance per watt. If last week it advanced the next generations for launch until 2027, in a new post on its blog, AMD has detailed the deployment of the Instinct MI430X, designed for large-scale AI environments.
Its implementation is already underway in two of the most important research centers in the world, in the Discovery supercomputer at the Oak Ridge National Laboratory and in the European exascale system, Alice Recoque.
Both systems use the Instinct MI430X along with AMD EPYC “Venice” processors. By releasing it directly to production systems rather than making a preliminary announcement, AMD positions the MI430X as the immediate successor to its previous Instinct line and a key element for countries and organizations looking to develop a open and self-sufficient AI infrastructure.
The MI430X solves the memory bottlenecks that hinder modern AI and simulation workloads with 432GB HBM4 memory, providing approximately 19.6 TB/s bandwidth. These specifications transform the accelerator into a near-in-memory computing device for models with trillions of parameters.
The chip incorporates AMD’s next-generation CDNA architecture. It maintains hardware-level double-precision FP64 support for traditional scientific computing, while introducing native FP8 and FP4 precision modes for efficient AI training and inference. The design also focuses on improving performance per watt, adapting to environments where power efficiency is as crucial as raw performance for sustained exascale operations.
The MI430X will be joined next year by the Instinct MI455X, designed for large-scale AI training and inference workloads. Already in 2027, the premiere of the new generation of AI accelerators, Instinct MI500, is expected.
