Georgia Tech's dedicated AI supercomputer is a cluster of 20 Nvidia HGX H100s; the DOE's Venado is the first large-scale system with Nvidia Grace CPU superchips deployed in the U.S.
One Nvidia strategy that has paid off is getting its CUDA GPU programming language taught in universities. After less than 20 years, there are more than 700 universities worldwide teaching CUDA to computer science students, introducing them to the Nvidia hardware architecture.
Its latest move is very similar. The company is working with Georgia Tech’s College of Engineering to build a dedicated AI supercomputer called AI Makerspace that’s designed to provide undergraduate students with access to hardware that would normally only be available to graduate students and researchers.
AI Makerspace is a cluster of 20 Nvidia HGX H100 systems, totaling 160 GPUs and running Nvidia’s AI Enterprise software. Students will access the cluster online as part of their coursework, learning through hands-on experience.
“The launch of the AI Makerspace represents another milestone in Georgia Tech’s legacy of innovation and leadership in education,” said Raheem Beyah, Dean of the College of Engineering and Southern Company Chair, in a statement. “Thanks to Nvidia’s advanced technology and expertise, our students at all levels have a path to make significant contributions and lead in the rapidly evolving field of AI.”
Students and faculty will also receive support through Nvidia Deep Learning Institute resources, including faculty-run Nvidia workshops, certifications, a university ambassador program, curriculum-aided teaching kits, and a developer community network.
As part of the project, Georgia Tech is introducing 14 new courses on AI for undergraduates. It intends to make Makerspace a part of the curriculum of all eight engineering schools at Georgia Tech by this fall, and by next year, all engineering undergraduate and graduate students will have access.
Los Alamos National Lab launches Venado supercomputer
Nvidia-powered supercomputers are starting to make their presence felt in the research sector, starting with Venado at the Department of Energy’s Los Alamos National Laboratory. Venado is built by Hewlett Packard Enterprise’s Cray EX supercomputer division and has a total of 2,560 direct, liquid-cooled Grace Hopper Superchips. It also has 920 Grace CPU chips.
The Grace Hopper chips with a combination of an Arm CPU with 144 cores (dubbed Grace) and the H200 GPU. The Grace CPU chips are without a GPU added.
Venado is the first large-scale system deployed in the U.S. with Nvidia Grace CPU superchips. LANL claims that it can top out at 10 exaFLOPs of performance, well ahead of the current top supercomputer in the world, the AMD-powered Frontier. However, the Venado benchmark is at FP8 while Frontier is benchmarked at FP64, so it’s not a true apples to apples comparison.
Atos brand Eviden delivers supercomputer to French agency
Across the Atlantic, the French Alternative Energies and Atomic Energy Commission (CEA) and Atos’s HPC brand Eviden announced the delivery of a new EXA1 HE (high efficiency) supercomputer to CEA. The computer is based on Eviden’s BullSequana XH3000 technology with 477 Grace Hopper superchips.
Last month, Eviden was selected to upgrade the capacity of the Zay supercomputer at the National Center for Scientific Research by providing 1,456 Nvidia H100 GPUs to increase the peak computing power from 36.85 petaflops to 125.9 petaflops.
