Georgia Tech × U. of Illinois Urbana-Champaign · National Science Foundation
Nexus is a supercomputer built for AI: NVIDIA DGX B200 systems, an all-flash data platform, and an 800 Gb/s fabric, with a portal that takes you from your first notebook to a full multi-node DGX run without a support ticket in between.
Hardware
Tightly coupled DGX systems for the largest training jobs, flexible high-memory GPU nodes for everything else, and VAST all-flash storage feeding both over a dedicated fabric. Full specifications are in the hardware guide.
16 NVIDIA DGX B200 systems for large-scale model training and tightly coupled multi-GPU science.
42 Blackwell RTX nodes for fine-tuning, inference, simulation, and data-intensive workloads.
VAST all-flash storage built for high-bandwidth AI workloads and large shared datasets.
System totals also count 12 management and interactive nodes (1,152 CPU cores, 6 TB memory) that run system services and user access.
Platform
Nexus is an HPC system with AI and interactive computing designed in from the start, not bolted on. Work the way your research works.
Jupyter and VS Code sessions on GPU nodes. Develop, debug, and visualize where the data lives.
Slurm scheduling from a single GPU to full multi-node DGX jobs, with GPU-aware partitions and fair-share allocation.
Container-first software environments. Run the exact image you built, with the system's optimized libraries when you want them.
Long training runs survive maintenance windows and preemption: checkpoint, requeue, resume.
High-throughput transfer in and out of the all-flash platform, sized so staging data never becomes the experiment.
Work extends outward to national systems such as Delta and DeltaAI when a project outgrows a single machine.
Projects, allocations, usage, and access in one place. Request time, add collaborators, and watch consumption against your award without leaving the browser.
Access
Nexus serves researchers and students nationwide, in partnership with the U.S. National Science Foundation, the ACCESS allocations ecosystem, and the National AI Research Resource (NAIRR) Pilot. Compute time goes to approved projects at no cost to the researcher.
Allocations with national scope for AI, ML, and data-intensive science in every domain, from genomics to astrophysics.
Applications opening soonCourse allocations and training programs at multiple levels, so the next generation learns on the same class of machine the field runs on.
Programs in developmentCollaboration opportunities for institutions and vendors building the software and systems that large-scale AI science depends on.
Get in touch
NAIRR Pilot
APACHE
APACHE AIRAVATAGet started
Applications open soon. Join the list, and you will be running from the quickstart within an hour of approval.
