Trusence Every claim has a source
Last updated 17 November 2026 Search Türkçe
Infrastructure

Lux, an AMD-built AI supercomputer, opens to scientists at Oak Ridge

Half of its annual capacity is reserved for the Genesis Mission, and a second machine in the same DOE and AMD arrangement is due in 2028.

A brass clockwork automaton stands beside a tall stack of gold coins, with an empty second pedestal waiting beside it.
AI-generated illustration

AMD and the U.S. Department of Energy have opened Lux, a supercomputer at Oak Ridge National Laboratory built by HPE from AMD EPYC CPUs, Instinct MI355X GPUs, Pensando networking and ROCm software, and it debuted at No. 14 on the latest TOP500 list. The system is intended as a dedicated AI resource for the Genesis Mission, the DOE-led effort to apply AI to science, and its allocated projects cover fusion plant design, materials and manufacturing, and river and water forecasting in the Tennessee Valley. Tom's Hardware reports about 4,000 MI355X GPUs across more than 500 nodes, with roughly 317 PetaFLOPS of FP64 and 20 ExaFLOPS of FP8 for AI training, and half of its 3.5 million annual node-hours reserved for Genesis work. The same TOP500 edition shows AMD hardware in 202 systems, about 40 percent of the list, up from 177 a year earlier. Lux is one of two machines in a $1 billion DOE and AMD arrangement, and the second, Discovery, is slated for 2028.

Why it matters

Lux gives the Department of Energy a dedicated AI system for science rather than a general-purpose machine, with half of its annual node-hours set aside for Genesis Mission research. Its debut at No. 14 on the TOP500 list, alongside AMD hardware appearing in 202 systems, shows AMD's share of the world's top supercomputers growing. Scientists working on fusion plant design, materials and manufacturing, and river and water forecasting are the groups whose access to this capacity changes first. The companion system, Discovery, is slated for 2028, so the arrangement is a multi-year public investment rather than a single machine.

Sources · 2