Nvidia will build 700-petaflop supercomputer for University of Florida

Nvidia and the University of Florida (UF) today announced plans to build the fastest AI supercomputer in academia. By enhancing the capabilities of UF’s existing HiPerGator supercomputer with the DGX SuperPod architecture, Nvidia claims the system — which it expects will be up and running by early 2021 — will deliver 700 petaflops (one quadrillion floating point operations per second) of performance.

Some researchers within the AI community believe that capable computers, in conjunction with reinforcement learning and other techniques, can achieve paradigm-shifting AI advances. A paper recently published by researchers at the Massachusetts Institute of Technology, MIT-IBM Watson AI Lab, Underwood International College, and the University of Brasilia found that deep learning improvements have been “strongly reliant” on increases in compute. And in 2018, OpenAI researchers released an analysis showing that from 2012 to 2018, the amount of compute used in the largest AI training runs grew more than 300,000 times with a 3.5-month doubling time, far exceeding the pace of Moore’s law.

UF and Nvidia say the revamped HiPerGator will give faculty and students the tools to apply AI across focuses including rising seas, aging populations, data security, personalized medicine, urban transportation, and food insecurity. Already, AI models at UF Health are being deployed to help collect, organize, and monitor patient conditions in real time through a system known as DeepSOFA.

HiPerGator — which will be roughly 18 times as powerful as the University of Texas at Austin’s Frontera — will be among the first to receive Nvidia’s DGX A100 systems. DGX A100 packs eight 7-nanometer Ampere-based A100 Tensor Core GPUs, providing 320 gigabytes of memory and the latest high-speed Mellanox HDR 200Gbps interconnects. A single A100 GPU’s 54 billion transistors can execute 5 petaflops of performance, according to Nvidia.

As a part of the boost, HiPerGator will gain 140 DGX A100 systems powered by 1,120 NVIDIA A100 Tensor Core GPUs, coupled with 4 petabytes of storage from DDN and 15 kilometers of optical cable. It will also benefit from Nvidia’s suite of AI application frameworks covering data analytics, AI training and inference acceleration, and recommendation systems.

Deep collaboration

Beyond HiPerGator, Nvidia said its partnership with UF will extend to ongoing support and collaboration in three principal AI areas. The Nvidia Deep Learning Institute will work with UF to develop curriculum, coursework, and programming tailored to address the needs and encourage the interest of young adults and teens in science, technology, engineering, and mathematics (STEM) and AI. The company will also found an Nvidia AI Technology Center at UF where graduate fellows and Nvidia employees will partner to advance the field of AI. And Nvidia solution architects and product engineers will team up with UF on the installation, operation, and optimization of the supercomputing resources on campus, including HiPerGator.

UF says this will lay the groundwork for the integration of AI with all of its disciplines, making it a “ubiquitous part” of the institution’s academic enterprise. It will offer certificates and degree programs in AI and data science with technical and industry-focused curriculum modules, and it’s committing to hiring 100 more faculty members focused on AI.

UF aims to train 30,000 graduates with AI skills by 2030. Steps to achieve that goal include collaborating with historically Black colleges, Hispanic-serving institutions, and K-12 programs and establishing the Equitable AI program, which will seek to bring faculty members across the university to create standards and certifications for developing tools and solutions cognizant of bias, unethical practice, and legal and moral issues. In addition, UF intends to partner with industry and other academic groups such as the Inclusive Engineering Consortium, led by UF Herbert Wertheim College of Engineering faculty member Dr. Damon Woodard, whose students will work with members to conduct research and recruitment to UF graduate programs and provide training in AI.

The initiatives and HiPerGator enhancements are anchored by a $50 million gift — $25 million from UF alumnus and Nvidia fellow Chris Malachowsky and $25 million in hardware, software, training, and services from Nvidia. UF will also make investments around its new machine “well beyond” $20 million, targeted at upgrading its on-campus datacenter.

from Hacker News https://ift.tt/32FVa2N