Aurora: The Fastest AI Supercomputer

Breaking New Ground

 

 

 

Intel recently announced that the Aurora supercomputer has shattered performance records at ISC High Performance 2024. This AI-centric system has surpassed the exascale barrier and become the fastest AI supercomputer in the world for open science. Here’s a detailed overview of Aurora’s breakthrough and what it means for high-performance computing (HPC) and AI.

Aurora Breaks the Exascale Barrier

At ISC 2024, Intel partnered with Argonne National Laboratory and Hewlett-Packard Enterprise (HPE) to unveil Aurora’s remarkable achievement. The system has reached 10.6 AI exaflops, pushing it beyond the exascale threshold of 1.012 exaflops.

This milestone marks a new era for HPC. Ogi Brkic, general manager at Intel, highlighted that Aurora will “pave the way for discoveries of the future.” Supercomputers like Aurora are essential for solving complex scientific challenges, from understanding cosmic phenomena to climate modelling.

Why Aurora Matters for AI and Science

From its inception, Aurora was designed as an AI-first supercomputer. It helps scientists leverage generative AI models to speed up research. Argonne’s early AI-driven projects show great promise. For example:

• Mapping 80 billion neurons in the human brain

• Enhancing high-energy particle physics with deep learning

• Accelerating drug discovery using machine learning

These breakthroughs demonstrate how Aurora’s power drives scientific progress.

Key Specifications of the Aurora Supercomputer

Aurora stands as one of the world’s largest GPU clusters. Its impressive hardware includes:

• 166 racks

• 10,624 compute blades

• 21,248 Intel® Xeon® CPU Max Series processors

• 63,744 Intel® Data Centre GPU Max Series units

• 84,992 HPE Slingshot fabric endpoints

Using just 87% of its 9,234 nodes, Aurora crossed the exascale barrier at 1.012 exaflops. It ranked second on the high-performance LINPACK (HPL) benchmark and third on the high-performance conjugate gradient (HPCG) benchmark. The HPCG benchmark assesses realistic memory and communication patterns, providing a fuller picture of system performance.

How Aurora Optimises AI Workloads

Aurora’s power comes from the Intel Data Centre GPU Max Series, based on the Intel Xe GPU architecture. This architecture includes specialised hardware blocks for matrix and vector computations, which are vital for AI and HPC tasks.

The Xe architecture excels at parallel processing, crucial for neural network calculations. It also powers Aurora’s top performance in the high-performance LINPACK mixed-precision benchmark, demonstrating AI’s growing role in HPC.

Additionally, Intel supports an open ecosystem for developers. Their suite includes the Intel® oneAPI DPC++/C++ Compiler and optimised AI frameworks, which allow flexible and scalable software development across various devices.

The Role of Open Ecosystems in HPC and AI

At ISC 2024, Intel emphasised the importance of open software and hardware ecosystems. Andrew Richards, CEO of Codeplay (an Intel company), highlighted the value of oneAPI, a unified programming model across diverse hardware.

Built on open standards, oneAPI helps developers write code that runs smoothly on different platforms without vendor lock-in. This aligns with the Linux Foundation’s Unified Acceleration Foundation (UXL), which includes major tech companies working to create an open ecosystem for heterogeneous computing.

Furthermore, Intel® Tiber™ Developer Cloud is expanding with new hardware and services. This allows developers to test, innovate, and scale AI workloads rapidly.

What’s Next for Intel and Aurora

Intel continues to advance HPC and AI with new supercomputers powered by Intel Xeon CPU Max Series and Data Centre GPU Max Series. Notable systems include:

• Italy’s ENEA CRESCO 8 for fusion energy research

• Texas Advanced Computing Centre for biology, turbulence, and materials science

• UK Atomic Energy Authority for future fusion power plant design

• Euro-Mediterranean Centre on Climate Change’s Cassandra system

Intel’s upcoming GPU, code-named Falcon Shores, will combine next-gen Xe architecture with Intel® Gaudi® technology. Early tests show Falcon Shores can deliver over twice the performance for real-world HPC applications compared to previous generations.

Aurora Leading the Future of AI Supercomputing

In summary, the Aurora fastest AI supercomputer, marks a pivotal moment in HPC and AI. Its record-breaking speed, powerful architecture, and open software ecosystem set a new standard for scientific computing. As Aurora drives forward, it will enable groundbreaking discoveries and accelerate innovation across many fields.

Aurora fastest AI supercomputer