Без рубрики

AMD’s partnership with the U.S. Department of Energy’s Oak Ridge National Laboratory (ORNL) is expanding again. The lab already runs two major supercomputers, Frontier (built entirely on AMD technology) and Summit (from IBM/NVIDIA). Over the next few years, they will be joined by two new AMD-powered systems named Lux AI and Discovery. Together, these projects represent about one billion dollars in combined public and private investment.

Lux AI is being developed through a collaboration between ORNL, AMD, Oracle Cloud Infrastructure, and HPE. The system will use AMD EPYC processors, Instinct MI355X GPUs, and Pensando networking. It is expected to go online in early 2026 and will serve as the first U.S. “AI Factory” supercomputer that supports joint work between government and private industry. Lux is designed to train, improve, and deploy large-scale AI foundation models, especially for scientific and engineering research that relies on heavy data processing.

Discovery will share the same broad mission but will be built with the more powerful Instinct MI430X GPUs, part of AMD’s MI400 line. These GPUs are aimed at “sovereign AI” efforts, giving the U.S. the ability to develop high-end AI models on domestically built hardware to protect sensitive data and maintain global competitiveness in science.

Discovery is optimized for demanding scientific simulations and AI workloads. It will offer extremely high bandwidth across memory, nodes, and networking components to ensure strong performance at scale. Unlike Frontier, which runs on the HPE Cray EX4000 platform, Discovery will use HPE’s new Cray Supercomputing GX5000 system, designed specifically for environments that combine AI and traditional high-performance computing. Despite the platform change, the software ecosystems are compatible, allowing codes developed for Frontier to be carried over to Discovery without major modifications.