In the National Computational Center for Energy Studies of the United States (NERSC) of the National Laboratory. Lawrence in Berkeley took place a solemn ceremony dedicated to the official launch of the Perlmutter supercomputer, also known as the NERSC-9 created by HPE in partnership with NVIDIA and AMD.
This is the most powerful AI-supercomputer in the world, based on 6159 NVIDIA A100 accelerators and approximately 1500 AMD EPYC MILAN processors. Its peak performance in mixed accuracy calculations is 3.8 EFFLS or almost 60 PFShpls in FP64-calculations.
Perlmutter is based on the HPE CRAY EX platform with direct liquid cooling and the interconnect Slingshot. The system includes both GPU nodes and nodes with processors. For data storage, the LUSTER file system is used by the volume of 35 PBB data exchange rate of more than 5 TB / s, which is deployed on All-Flash HPE Clusterstor E1000.
Installing Perlmutter is broken into two stages. Today’s presentation announced the completion of the first (phase 1) stage, which began in November last year. Within its framework, 1.5 thousand computing nodes were installed, each of which has four NVIDIA A100 accelerator, one AMD EPYC MILAN processor and 256 GB of memory. At the second stage, 3 thousand CPU nodes with two AMD EPYC Milan and 512 GB of memory will be added at the end of 2021., As well as another 20 access nodes and four nodes with a large memory volume.
Also at the first stage, service nodes were deployed, including 20 user access nodes, where you can prepare containers with applications for subsequent launch on a supercomputer and use KUBERNETES for orchec. The development environment will include NVDIA HPC SDK in addition to CCE compiler sets, GCC and LLVM to support various parallel programming tools, such as MPI, OpenMP, CUDA and OpenACC for C, C ++ and Fortran.
It is reported that perlmutter is preparing more than two dozen applications for calculating in the field of astrophysics, predicting climate change and in other fields. One of the tasks for the new system will be the creation of a three-dimensional map of the visible universe based on DARK ENERGY Spectroscopic Instrument. Another direction for which the supercomputer use is devoted to the material science, the study of atomic interactions, which can indicate the path to creating more efficient batteries and biofuels.