the supercomputer Border has already been officially recognized as the world’s fastest exascale supercomputer, and that translates to nothing less than achieving a performance of 1.1 ExaFlops when tested on the Linpack FP64 benchmark. The said test was in sustained performance, while its peak performance reached 1.69 ExaFlops. In the HPL-AI benchmark, it achieved 6.88 ExaFlopsand that translates to processing 68 million instructions per second for each neuron in the brain.
This performance is obtained by adding 602,112 AMD Zen3 cores coming from its 3rd generation AMD EPYC processors with 64 cores each at 2.00 GHz, which gives us the result it uses 9,408 AMD EPYC processorsto which is added support for AMD Instinct MI250X graphics, generating a total consumption of 21.3 MW.
In this way, Frontier takes the position of the Japanese supercomputer Fugakuequipped with 7,200,000 cores via 150,000 Fujitsu A64FX processors based on ARMv8.2-A scalable vector extension architecture.
The Frontier supercomputer was built by Hewlett Packard Enterprise (HPE) and is housed at the Department of Energy’s Oak Ridge National Laboratory (ORNL) in the southern United States, specifically in the state of Tennessee.
The system has 9,408 compute nodes, each with a 64-core AMD “Trento” processor paired with 512GB of DDR4 memory and four AMD Radeon Instinct MI250X GPUs, giving us a total of 37,632 GPUs which grant 8,138,240 cores, also offering 4.6 petabytes of HBM memory (128 GB per GPU). The processors and GPUs are connected to each other using the HPE Cray Slingshot network fabric.
information is complete 4,500 NVMe SSDs with a total capacity of 11.5 PB, all being water-cooled. It is expected that this title will last very little for Frontier, because next year it will arrive the captainwhich will rob him of the throne, since he will use the new AMD architectures.