More details of the world’s most powerful supercomputer have emerged in its official opening in the US.
Frontier is based at Oak Ridge National Laboratory (ORNL) and is the first to reach exascale (1000 PetaFLOPS) performance for machine learning applications. It is also the most power efficient supercomputer, as shown in the Hot500 and Green500 listings.
Each node contains one optimized third-generation AMD EPYC processor and four AMD Instinct MI250X AI accelerators for a system-wide total of 9,472 CPUs and 37,888 GPUs in 74 liquid cooled cabinets.
- US fights back with first exascale supercomputer
- Europe aims for open source exascale supercomputers
- SiPearl in massive expansion for exascale chip design
The nodes are connected by the Slingshot high-performance Ethernet fabric developed by HP Enterprise (HPE) for HPC and AI solutions. By connecting the CPUs, GPUs and high-performance storage, Slingshot enables larger data-intensive workloads that would otherwise be bandwidth limited and provides higher speed and congestion control between the HPE Cray EX cabinets. The system uses 90 miles of interconnect cables.
“As the world’s most powerful AI machine, Frontier’s novel architecture is also ideally suited for delivering unprecedented machine learning and data science insights and automations that could vastly improve our understanding of critical processes, from drug delivery to nuclear fusion to the global climate,” said Doug Kothe, associate laboratory director of ORNL’s Computing and Computational Sciences Directorate and director of the Exascale Computing Project.
An I/O subsystem from HPE is being brought online this year to support Frontier and the OLCF. The I/O subsystem features an in-system storage layer and Orion, a centre-wide file system based on Lustre, a Linux cluster system that is used by 60 of the top 100 supercomputers.
The in-system storage layer will employ compute-node local storage devices connected via PCIe Gen4 links to provide peak read speeds of more than 75Tbyte/s, peak write speeds of more than 35 Tbyte/s and more than 15 billion random-read input/output operations per second.
The Orion file system will provide around 700 Pbytes of storage capacity and peak write speeds of 5 Tbyte/s.
“Research that might once have taken weeks to complete, Frontier will tear through in hours, even seconds,” said Deputy Secretary of Energy David Turk at the official opening of the system. “Oak Ridge has positioned the United States to lead the world in solving massive scientific challenges across the board.”
“Exascale computing is a powerful tool that will allow us to advance the core missions of the Office of Science — to deliver scientific discoveries and major scientific tools that will transform our understanding of nature and advance the energy, economic, and national security of the U.S.,” said DOE Office of Science Director Asmeret Asefaw Berhe. “Frontier makes exascale computing a reality and opens many doors for the future of scientific research to solve big problems.”
“This is a very important milestone for the nation and the world,” said Gina Tourassi, director of ORNL’s National Center for Computational Sciences, which oversees the OLCF. “The computational models we can build with this computer will help us fill in missing pieces of the puzzle for a range of scientific inquiries, from matter and energy to life itself, and will give the next generation of scientists the tools and the springboard they need to make even greater leaps of understanding.”
ORNL’s scientific partners, such as General Electric Aviation and GE Power, plan to use the supercomputer to develop sustainable hydrogen propulsion and hybrid electric technologies and to maximize the potential of clean-energy technologies such as wind power.
“GE Aerospace and Research will be using exascale computing, including time on the Frontier supercomputer, to revolutionize the future of flight with sustainable hydrogen propulsion and hybrid electric technologies,” said David Kepczynski, chief information officer at GE Research. “In pursuit of a net-zero carbon future, exascale supercomputing systems will be indispensable tools for GE researchers and engineers working at the cutting edge to ‘Build a World that Works.’”
“Frontier is a landmark in computing that will usher in a new era of insights and innovation,” said Antonio Neri, president and CEO of HPE. “We are proud of this massive achievement that will help make significant contributions to science, push the envelope for artificial intelligence, and strengthen U.S. industrial competitiveness. Frontier was made possible through powerful engineering and design, and most importantly, through a strong partnership between Oak Ridge National Laboratory, HPE and AMD.”
Next steps include additional testing and validation of the system, which remains on track for final acceptance and early science access later in 2022. Full access for science applications is expected at the beginning of 2023.
- RISC-V power controller for supercomputer chip
- Researcher builds supercomputer with Raspberry Pi boards
- Getting closer to Microsoft’s AI supercomputer power figures
- Finnish supercomputer contender will be climate neutral