Aurora is a state-of-the-art exascale supercomputer developed through a collaboration between Intel, Hewlett Packard Enterprise (HPE), and the US Department of Energy. Hosted at Argonne National Laboratory, Aurora is engineered to advance high performance computing (HPC) by supporting extremely large-scale simulations, data analytics, and artificial intelligence (AI) workloads. The system is built on a Cray EX platform and features 10,624 compute nodes, each equipped with two Intel Xeon Max Series CPUs and six Intel Data Center GPU Max Series accelerators. Aurora’s architecture includes a unified memory system with high-bandwidth memory (HBM) on both CPUs and GPUs, interconnected by the HPE Slingshot high-performance fabric. The storage subsystem leverages the Distributed Asynchronous Object Store (DAOS), providing 230 petabytes of capacity and 31 terabytes per second of bandwidth. Aurora is designed to achieve a theoretical peak performance exceeding 2 exaflops, making it one of the most powerful supercomputers in the world. The platform supports a wide range of programming models and frameworks, including MPI, OpenMP, SYCL/DPC++, Intel oneAPI, and others, enabling researchers to port and optimize applications for scientific discovery, machine learning, and data-driven research.

Key Features

Exascale Performance Delivers over 2 exaflops of theoretical peak performance for HPC and AI workloads.

• Enables unprecedented simulation and modeling scale
• Supports large-scale data analytics and deep learning

Advanced Compute Architecture Nodes with Intel Xeon Max CPUs and Intel Data Center GPU Max Series.

• 10,624 nodes, each with 2 CPUs and 6 GPUs
• Unified memory with high-bandwidth memory (HBM) on CPUs and GPUs

High-Performance Interconnect HPE Slingshot 11 fabric with Dragonfly topology.

• 25.6 TB/s per switch, adaptive routing
• Low-latency, high-bandwidth communication

Innovative Storage System Distributed Asynchronous Object Store (DAOS).

• 230 PB storage, 31 TB/s bandwidth
• Optimized for data-intensive workloads

Flexible Software Stack Supports multiple programming models and frameworks.

• Intel oneAPI, MPI, OpenMP, SYCL/DPC++, Kokkos, RAJA, and more
• Data and machine learning frameworks included

Scalability and Reliability Designed for large-scale, multi-user scientific and industrial workloads.

• 21,248 CPUs and 63,744 GPUs in total
• Enterprise-grade reliability and support

Benefits

Accelerated Scientific Discovery Enables researchers to solve complex problems at unprecedented scale.

• Supports advanced simulations in physics, chemistry, biology, and engineering
• Facilitates breakthroughs in AI and data science

Versatile Workload Support Handles traditional HPC, AI, and data analytics workloads.

• Flexible for a wide range of scientific and industrial applications
• Optimized for both simulation and machine learning

Future-Proof Architecture Built with the latest Intel and HPE technologies.

• Ready for emerging research needs and evolving software ecosystems
• Supports code portability and optimization

Collaboration and Training Backed by programs for code migration, optimization, and user training.

• Early Science Program for application readiness
• Technical support from Intel, HPE, and Argonne experts