Cumulus-2 began to be thought up in 2020 by the Oak Ridge National Laboratory’s (ORNL) which also hosts Frontier, the world’s current fastest supercomputer. Two years later, it is already up and running.
This new system was launched by the Department of Energy (DOE) in April, with the purpose of supporting climate research, namely in the collection, dissemination and provision of data to the scientific community.
It is an HPC cluster, designed for the Atmospheric Radiation Measurement (ARM) user facility of the department of energy (DOE), whose mission is, since 1986, to create a series of tools that record atmospheric data in order to better study the planet’s climate.
With 16,384 processing cores, Cumulus-2 has become more powerful than any other ORNL system previously built for this purpose. It surpasses the two previous systems built with 1,080 and 4,032 cores, respectively. In practice, it will have four times the power of its predecessor Cumulus-1.
“This new cluster will greatly accelerate processing speeds for simulations and boost capabilities to interpret ARM’s storehouse of data” – Giri Prakash, director of the data center for DOE’s Atmospheric Radiation Measurement (ARM) user facility,
Smaller and with a quite different architecture, the system consists in 32 chassis, 28 of which are individually equipped with four nodes. Each node has 7713 CPUs and 256 GB of memory. The remaining chassis have the same number of nodes but are equipped with 512 GB of memory.
This is an important step in ensuring the DOE the required HPC infrastructure to drive the work that is done. Giri Prakash, director of the ARM data center, handles more than 450 instruments that collect atmospheric phenomena, and ensures that every day thousands of data arrive at the facility for his team to collect, process and archive – that normally is available to the user after 24 hours. “This new cluster will greatly accelerate processing speeds for simulations and boost capabilities to interpret ARM’s storehouse of data”, he says.
The arrival of Cumulus-2 translates to better atmospheric model simulations, petascale data storage, analysis of massive amounts of data, and machine-learning for climate and atmospheric science research.
Its four-petabyte parallel file system for data storage complements the system to ensure that this increased computational power continues to serve the purpose of arriving at “a robust model prediction of Earth’s climate and environmental systems” as well as achieving sustainable solutions to key climate and environmental challenges.