|Image Source : AI Imagine|
The National Energy Research Scientific Computing Center (NERSC) is granting approved researchers a rare opportunity: 50% discounted access to Nvidia A100 GPUs on the Perlmutter supercomputer through September. This valuable high performance computing capacity would normally cost commercial AI firms a premium.
NERSC facilitates scientific breakthroughs by providing supercomputing resources to qualified researchers. Located in California, NERSC is sponsored by the U.S. Department of Energy Office of Science. Its new Perlmutter system ranks among the world’s fastest supercomputers.
Featuring state-of-the-art Nvidia A100 Tensor Core GPUs, Perlmutter delivers exceptional AI modeling and simulation performance. Access is highly competitive, allocated only to projects advancing fundamental science.
But with many academics on summer break, Perlmutter has unused capacity. NERSC aims to fill this vacant compute time by offering accredited researchers a 50% discount on A100 node usage through September 30.
“Using your time now benefits the entire NERSC community and spreads demand more evenly throughout the year, so to encourage usage now, we are discounting all jobs run on the Perlmutter GPU nodes by 50% starting tomorrow and through the end of September,” wrote NERSC’s Rebecca Hartman-Baker in an email obtained by [Source].
This represents huge savings. A 3-hour job utilizing 7 nodes would normally incur a charge of 21 GPU node-hours — instead users would pay only 10.5 GPU node-hours.
Commercial AI Demand Outstrips Supply
The discounted access comes as demand for AI modeling compute power far exceeds supply industry-wide. The explosion of large language models like ChatGPT has spurred shortages of advanced GPUs like Nvidia’s A100.
Cloud services firms and data centers are stretched to their limits provisioning resources to commercial AI developers. Waitlists for Nvidia GPUs now extend into 2023 as supply chain shortfalls continue constraining production.
Against this backdrop, NERSC’s offer is exceedingly generous. The center could easily rent access commercially at a sizable premium given insatiable demand. But its charter focuses on enabling scientific research, rather than profit.
Only Select Researchers Eligible
The key caveat is that this discounted access applies solely to authorized NERSC users — not commercial entities. All supercomputing time must be allocated via a peer-reviewed proposal process proving scientific merit.
NERSC resources are reserved for projects benefiting fundamental research and national priorities. National security sensitivities and intellectual property considerations also limit eligibility.
So companies aiming to leverage A100s to train the next ChatGPT or Stable Diffusion won’t be able to jump on this NERSC deal. Strict limits maintain the center’s research orientation.
But for those scientists who are already NERSC users, it represents an uncommon opportunity to significantly reduce GPU computing costs. With manypeer-reviewed projects accepted annually in diverse scientific domains, the potential impact is substantial.
U.S. Leads in Supercomputing Investments
The latest Top500 list of global supercomputers demonstrates America’s continued leadership in high performance computing. The U.S. claims 7 of the top 10 systems, including Perlmutter at #3.
But rival nations are ramping up investments in this strategic technology. China aims to surpass the U.S. in supercomputing speed by 2025 under its National Scientific Innovation Plan.
To maintain its competitive edge, the U.S. passed the CHIPS Act last year, authorizing $52 billion to bolster domestic semiconductor research and manufacturing. Supercomputing leadership is vital for both economic competitiveness and national security.
NERSC facilities like Perlmutter push the boundaries of HPC capabilities through close public-private partnerships with tech firms like Nvidia. Keeping these cutting-edge systems fully utilized ensures taxpayers see maximum returns on investments.
Striking a Balance on Usage
But operational realities require balancing open science access, industry collaboration, and national security. Purely commercial uses of government supercomputing take a backseat to research advancing knowledge.
While leasing idle capacity to AI developers could generate extra funding, it would detract from NERSC’s fundamental mission. Strategic priorities necessitate keeping these HPC assets focused on areas like energy, advanced materials, and cosmic mysteries.
Yet NERSC does seek to enable technologies like artificial intelligence to benefit the broader economy when possible. Collaboration between government-funded scientists and private industry ultimately speeds ideal innovations like Perlmutter to global markets.
The discounted offer through September illustrates how agencies like NERSC can pragmatically optimize supercomputer utilization while furthering scientific discovery. America’s computing dominance relies on these world-leading facilities enabling researchers across countless disciplines to push boundaries.
For more on the transformative science taking place at NERSC and partner supercomputing centers, see our related coverage here. The quest for breakthroughs depends on access to the most powerful systems.