Perlmutter supercomputer built with more than 6,000 NVIDIA A100 GPUs
Perlmutter supercomputer built with more than 6,000 NVIDIA A100 GPUs
NVIDIA and the National Energy Research Scientific Computing Center (NERSC) in the United States have brought online what they are calling “the world’s fastest supercomputer”.
Called Perlmutter, it will deliver nearly four exaflops of AI application research performance for more than 7,000 scientists.
Built by HPE Cray, the system is being delivered in two phases and will include 6,144 A100 Tensor Core GPUs and will be tasked with tasks like stitching together the largest 3D map of the universe.
Phase 1 features 1,536 GPU-accelerated nodes, each containing four NVIDIA NVLink-3-connected A100 Tensor Core GPU processors and one AMD Milan CPU processor. Phase 1 also includes a 35 PB all-flash Lustre file system that will provide very high-bandwidth storage to the compute nodes.
Phase 2, set to arrive later this year, will add 3,072 CPU-only nodes, each with two AMD Milan processors and 512 GB of memory per node.
The A100, NVIDIA’s first chip based on its NVIDIA Ampere architecture, is a 7-nanometer GPU processor with more than 54 billion transistors.
Perlmutter is said to be three to four times faster than NERSC’s current supercomputer Cori.
No costs were revealed during the launch, but as a point of comparison, Singapore’s new supercomputer also built by HPE Cray and using AMD processors came as part of a S$200 million government grant. And the speed for that is only in the petaflops, not exaflops.