Nvidia has partnered with the University of Florida to up grade the university’s HiPerGator supercomputer. Set to go stay in early 2021, the challenge will make HiPerGator the fastest AI supercomputer in academia with 700 petaflops of efficiency, according to Nvidia and UF.
HiPerGator three. will use a cluster comprised of one hundred forty Nvidia DGX A100 nodes, which are $two hundred,000 supercomputing AI programs that every comprise 8 A100 GPUs. The DGX A100, providing 320 GB of RAM for education substantial AI datasets, is able of five petaflops of efficiency.
The cluster of one hundred forty DGX A100 nodes, named a DGX SuperPOD, will therefore strike 700 petaflops, Nvidia and the university mentioned. HiPerGator three. will also involve 4 petabytes of significant-efficiency storage and tap Nvidia Mellanox HDR 200Gb/s InfiniBand networking.
“The projected 700 petaflops are absolutely up there,” mentioned Nick McQuire, senior vice president and head of AI and business research at CCS Perception. “It is part of an arms race we are looking at at a national stage and in academia to make certain universities continue to be competitive globally.”
UF is the 1st academic institute to use the DGX A100 engineering, which was declared in May perhaps.
The fastest supercomputer on the Super Personal computer Major five hundred list from June is a technique at RIKEN Heart for Computational Science in Japan named Supercomputer Fugaku, which hits 415 petaflops of efficiency. For particular tasks, Fugaku peaks at 1,000 petaflops, or 1 exaflop.
If Nvidia’s statements bear out, UF’s future technique will be equivalent to Fugaku in phrases of electricity.
Nick McQuireAnalyst, CCS Perception
Supercomputing opposition shifts
Up till now, most of the significant-profile opposition in AI supercomputing has centered on technique speed and been funded at the national govt stage, these as with Fugaku, McQuire mentioned.
The Major five hundred list does involve many supercomputers at universities, like the University of Texas, Nagoya University in Japan and Mississippi State University. Nvidia engineering can help electricity many of the supercomputers on the list, as do HPE and IBM. The a few tech giants dominate the supercomputing market alongside vendors these as Fujitsu, Dell, Lenovo, Atos and Penguin Computing.
The UF challenge is staying funded by way of a $twenty five million donation from UF alumnus and Nvidia co-founder Chris Malachowsky, alongside with $twenty five million in hardware, program, companies and education from Nvidia. UF also pledged $twenty million of its have funds.