Intel-Powered Aurora Becomes The Fastest Supercomputer For AI, Breaks the Exascale Barrier

The Aurora supercomputer, a collaborative effort between Intel, Argonne National Laboratory, and HPE, has achieved a significant milestone by breaking the exascale barrier. Equipped with Intel’s Ponte Vecchio hardware, this powerhouse now holds the title of the fastest AI system in the world dedicated to open science.

Key Specifications

• Compute Performance: The Aurora supercomputer achieves an impressive 1.01 exaflops in general compute performance and 10.6 AI exaflops.
• Hardware Components:
  • 166 racks
  • 10,624 blades
  • 21,248 Intel Xeon CPU Max chips (4th Gen Sapphire Rapids)
  • 63,744 Intel Data Center GPU Max series units (Ponte Vecchio)
  • Utilizes the HPE Slingshot fabric for interconnect
  • 84,992 endpoints

Performance Metrics

• HPL LINPACK Benchmark: Ranked second with 1.012 exaflops using 87% of total node capacity (9234 nodes out of 10,624).
• HPCG Test: Ranked third with 5612 TFLOPs/second using 39% of the system.
• AI Performance: Tops the charts with a total rated performance of 10.6 AI exaflops, leveraging the Xe core architecture and AI hardware blocks.

Despite initial delays, the Aurora supercomputer has now fulfilled its promise, providing a powerful platform for scientific research, high-performance computing, and AI applications. Intel emphasizes the crucial role of open ecosystems in driving AI-accelerated HPC.