$1 billion: AMD and partners aim to build new supercomputer in just 6 months

The US Department of Energy (DOE), together with AMD, Hewlett Packard Enterprise (HPE), Oracle and Oak Ridge National Laboratory (ORNL), has announced the construction of two new supercomputers. The total volume of the order amounts to approximately $1 billion. AMD is supplying the processors, AI accelerators and network chips. HPE is contributing the server architecture, while Oracle Cloud Infrastructure and ORNL are handling the software and research integration. The project is being financed as a public-private partnership and the computing power will be shared.

"Lux" in 6 months, "Discovery" by 2029

The first computer, called "Lux", is scheduled to be completed in just six months, which, according to AMD CEO Lisa Su, would be "the fastest implementation of a system of this magnitude we've ever seen". AMD MI355X accelerators and specially developed CPUs will power the system, which is expected to achieve three times the AI performance of current DOE supercomputers.

The second supercomputer, "Discovery", is planned as a follow-up project and will utilize AMD's new MI430 AI chips from the upcoming MI400 series. It will combine classic high-performance computing (HPC) with advanced AI capabilities. Discovery is scheduled to be delivered in 2028 and go into operation in 2029.

Big goals

According to the Department of Energy, the supercomputers are intended to "supercharge" research in the fields of nuclear and fusion energy, drug development and national security. The long-term goal is to be able to realistically simulate fusion reactions within two to three years, which would massively accelerate the development of functional fusion reactors and thus the path to virtually unlimited, clean energy. Furthermore, most currently fatal cancers are expected to become treatable within five to eight years.