NERSC’s Perlmutter supercomputer offers significant discounts to aid AI research during September.
In a bid to support the growing demand for computing power in AI research, the National Energy Research Scientific Computing Center (NERSC) has announced a substantial discount on Nvidia’s A100 GPUs for the month of September. This move comes at a crucial time when access to high-performance computing resources for AI workloads is scarce globally.
Half-Price Offer for NERSC Members
Between September 7 and October 1, NERSC members will enjoy a significant reduction in the cost of running AI jobs on the Perlmutter supercomputer. Typically, a three-hour job utilizing seven nodes incurs a charge of 21 GPU node hours. However, during this special promotion, the same job will cost only 10.5 GPU node hours.
Supporting the AI Community
Rebecca Hartman-Baker, the leader of the user engagement group at NERSC, emphasized the broader benefits of this offer, stating, “We are discounting all jobs executed on the Perlmutter GPU nodes by 50% starting tomorrow and through the end of September to encourage usage today, which benefits the broader NERSC community and spreads demand more fairly throughout the year.”
In addition to the cost reduction, NERSC is also providing assistance to users who may need help optimizing their scripts or require guidance on getting started with code development.
The Perlmutter Supercomputer
Launched in 2021, the Perlmutter supercomputer is an HPE Cray EX system that combines AMD Zen 3 Epyc CPUs with Nvidia A100 Tesla Core GPUs. The machine’s initial phase included 1,536 GPU-accelerated AMD CPU nodes, each equipped with four A100 GPUs, along with 35PB of all-flash Lustre-based storage. In its second phase, the supercomputer was expanded with 3,072 CPU-only nodes, featuring two AMD Epyc processors and 512GB of memory.
This supercomputer primarily serves critical scientific research purposes, such as nuclear fusion simulations, climate projections, and material and biological studies. Notably, it has contributed to projects like uncovering the intricacies of atomic interactions, which have the potential to lead to advancements in battery technology and biofuels.
While this discount is a welcome opportunity for AI researchers, it is exclusive to NERSC members. The offer was first brought to attention by Glenn Lockwood, a specialist in high-performance computing at Microsoft, who suggested that NERSC could potentially leverage idle capacity during the summer months for commercial workloads.
It’s worth noting that alternative options for renting GPUs, including Akash’s decentralized Supercloud for AI network, are available for those outside of NERSC. However, this initiative stands as a unique opportunity for NERSC members to advance their AI research at a significantly reduced cost.
In a landscape where access to GPU capacity for AI work is at a premium, NERSC’s discounted offer on Nvidia A100 GPUs is a notable development, providing crucial support for scientific computing and AI advancements.