Aurora Supercomputer — Argonne National Laboratory

Aurora is the United States’ second exascale supercomputer, located at Argonne National Laboratory’s Argonne Leadership Computing Facility (ALCF) in Lemont, Illinois. It achieved exascale performance — exceeding 1 exaflop (10^18 operations per second) — in 2024, becoming a major node in US government AI computing infrastructure alongside Oak Ridge’s Frontier.

Hardware: Aurora is built on Intel’s Ponte Vecchio (Intel Data Center GPU Max 9470) architecture — Intel’s first purpose-built data center GPU for HPC and AI. The system pairs Intel Xeon Sapphire Rapids CPUs with 63,744 Intel GPU Max accelerators across 10,624 nodes. Aurora has a peak theoretical performance of 2 exaflops in mixed precision, making it optimized for AI workloads including transformer-based model training.

Scale and cost: The system cost approximately $500 million. It has roughly 10 petabytes of high-bandwidth memory and a Slingshot 11 interconnect running at 200 Gbps per port. The storage subsystem provides 230 petabytes of capacity, enabling handling of the massive datasets required for foundation model training.

Power: At full load, Aurora draws approximately 60 megawatts — making it one of the most power-hungry computing facilities in the US. The facility is powered by the Commonwealth Edison (ComEd) grid in northern Illinois.

AI workloads: Aurora’s design explicitly targets AI and machine learning alongside traditional scientific computing. Priority workloads include large-scale LLM training, drug discovery (via generative molecular models), cosmological simulation with AI-assisted analysis, and climate modeling. The DOE’s INCITE program grants access to national laboratory teams, universities, and AI research groups.

Intel’s Ponte Vecchio gamble: Aurora’s deployment was repeatedly delayed (originally planned for 2021) due to Intel’s struggles with Ponte Vecchio yield and performance. The delays allowed AMD’s MI250X-powered Frontier to claim the first exascale title. Despite the difficult launch, Aurora represents a significant commercial validation of Intel’s Xe GPU architecture for HPC and AI inference and training.

Operator: ANL is managed by UChicago Argonne LLC — a partnership between the University of Chicago and other organizations — for the US Department of Energy’s Office of Science.