Nvidia, University of Florida building AI supercomputer

Nvidia and the University of Florida have teamed up to build what they claim will be the fastest AI supercomputer in academia, capable of hitting 700 petaflops of performance.

Nvidia has partnered with the University of Florida to upgrade the university's HiPerGator supercomputer. Set to go live in early 2021, the project will make HiPerGator the fastest AI supercomputer in academia with 700 petaflops of performance, according to Nvidia and UF.

HiPerGator 3.0 will use a cluster comprised of 140 Nvidia DGX A100 nodes, which are $200,000 supercomputing AI systems that each contain eight A100 GPUs. The DGX A100, providing 320 GB of RAM for training massive AI datasets, is capable of 5 petaflops of performance.

The cluster of 140 DGX A100 nodes, called a DGX SuperPOD, will thus hit 700 petaflops, Nvidia and the university said. HiPerGator 3.0 will also include four petabytes of high-performance storage and tap Nvidia Mellanox HDR 200Gb/s InfiniBand networking.

"The projected 700 petaFLOPS are certainly up there," said Nick McQuire, senior vice president and head of AI and enterprise research at CCS Insight. "It is part of an arms race we are seeing at a national level and within academia to ensure universities remain competitive globally."

UF is the first academic institute to use the DGX A100 technology, which was announced in May.

The fastest supercomputer on the Super Computer Top 500 list from June is a system at RIKEN Center for Computational Science in Japan called Supercomputer Fugaku, which hits 415 petaflops of performance. For specific tasks, Fugaku peaks at 1,000 petaflops, or 1 exaflop.

Nvidia, A100GPU
Nvidia's new A100 architecture will help power the University of 'Florida's HiPerGator 3.0 supercomputer.

If Nvidia's claims bear out, UF's upcoming system will be comparable to Fugaku in terms of power.

It is part of an arms race we are seeing at a national level and within academia to ensure universities remain competitive globally.
Nick McQuireAnalyst, CCS Insight

Supercomputing competition shifts

Up until now, most of the high-profile competition in AI supercomputing has centered on system speed and been funded at the national government level, such as with Fugaku, McQuire said.

The Top 500 list does include many supercomputers at universities, including the University of Texas, Nagoya University in Japan and Mississippi State University. Nvidia technology helps power many of the supercomputers on the list, as do HPE and IBM. The three tech giants dominate the supercomputing market alongside vendors such as Fujitsu, Dell, Lenovo, Atos and Penguin Computing.

The UF project is being funded through a $25 million donation from UF alumnus and Nvidia co-founder Chris Malachowsky, along with $25 million in hardware, software, services and training from Nvidia. UF also pledged $20 million of its own money.

Next Steps

Nvidia extends its AI reach with cloud-based development hub

Dig Deeper on AI infrastructure

Business Analytics
Data Management