The AI-fastest supercomputer is Aurora.

 

 

Intel claims that Aurora has gone through the exascale barrier and is the fastest AI supercomputer at ISC 2024. It also discusses the significance of an open ecosystem in HPC and AI.

What’s New: At ISC High Performance 2024, Intel, in partnership with Argonne National Laboratory and Hewlett Packard Enterprise (HPE), revealed that the Aurora supercomputer has achieved 10.6 AI exaflops, making it the fastest AI system in the world for open science.

It has also broken the exascale barrier at 1.012 exaflops. Additionally, Intel will discuss how open ecosystems are essential to the advancement of AI-accelerated high performance computing (HPC).

“By surpassing exascale, the Aurora supercomputer will enable it to pave the way for discoveries of the future.”

Supercomputers act as a compass, pointing the way toward solutions to incredibly challenging scientific problems that could advance civilization, from deciphering cosmic riddles to comprehending patterns in the climate.–Ogi Brkic, general manager of Data Center AI Solutions at Intel and vice president

Why This Is Important: From the beginning, Aurora was intended to be an AI-centric system that would enable scientists to use generative AI models to hasten scientific discoveries. Early AI-driven research at Argonne has advanced significantly. Among the many achievements are the mapping of the 80 billion neurons in the human brain, the improvement of high-energy particle physics by deep learning, and the acceleration of drug discovery and design using machine learning.

Details of the Aurora Supercomputer: The Aurora supercomputer is one of the largest GPU clusters in the world, with 166 racks, 10,624 compute blades, 21,248 Intel® Xeon® CPU Max Series processors, and 63,744 Intel® Data Center GPU Max Series units.

84,992 HPE slingshot fabric endpoints make up Aurora’s largest open, Ethernet-based supercomputing connection on a single system. The Aurora supercomputer crossed the exascale barrier at 1.012 exaflops using 9,234 nodes, or just 87% of the system, yet it came in second on the high-performance LINPACK (HPL) benchmark. With 39% of the machine, the Aurora supercomputer also achieved third place on the high-performance conjugate gradient (HPCG) benchmark, processing 5,612 teraflops per second (TF/s).

The goal of this benchmark is to evaluate more realistic situations that offer insights into memory access and communication patterns—two crucial components of real-world HPC systems.

It provides a full perspective of a system’s capabilities, complementing benchmarks such as LINPACK.

How to Optimize AI: The Intel Data Center GPU Max Series is the brains behind the Aurora supercomputer. The core of the Max Series is the Intel Xe GPU architecture, which includes specialized hardware including matrix and vector computation blocks that are ideal for AI and HPC applications. Because of the unmatched computational performance provided by the Intel Xe architecture, the Aurora supercomputer won the high-performance LINPACK-mixed precision (HPL-MxP) benchmark, which best illustrates the significance of AI workloads in HPC.

The parallel processing power of the Xe architecture excels at handling the complex matrix-vector operations that are a necessary part of neural network AI computing.

Deep learning models rely heavily on matrix operations, which these compute cores are essential for speeding up. Together with a wide range of performance libraries, Intel’s suite of software tools, which includes the Intel® oneAPI DPC++/C++ Compiler, and optimized AI frameworks and tools, the Xe architecture supports an open ecosystem for developers that is flexible and scalable across multiple devices and form factors.

Advancing Accelerated Computing with Open Software and Compute Capacity: In his special session at ISC 2024, on Tuesday, May 14 at 6:45 p.m., (GMT+2) Hall 4, Congress Center Hamburg, Germany, CEO Andrew Richards of Codeplay, an Intel company, will address the growing demand for accelerated computing and software in HPC and AI.

He will highlight the importance of oneAPI, offering a unified programming model across diverse architectures. Built on open standards, oneAPI empowers developers to craft code that seamlessly runs on different hardware platforms without extensive modifications or vendor lock-in. This is also the goal of the Linux Foundation’s Unified Acceleration Foundation (UXL), in which Arm, Google, Intel, Qualcomm and others are developing an open ecosystem for all accelerators and unified heterogeneous compute on open standards to break proprietary lock-in. The UXL Foundation is adding more members to its growing coalition.

Concurrently, Intel® TiberTM Developer Cloud is augmenting its computational capacity through novel cutting-edge hardware platforms and novel service functionalities, thereby enabling developers and enterprises to assess the most recent Intel architecture, promptly innovate and optimize AI models and workloads, and subsequently implement AI models at large scale.

Large-scale Intel® Gaudi® 2-based and Intel® Data Center GPU Max Series-based clusters, as well as previews of Intel® Xeon® 6 E-core and P-core systems for specific clients, are among the new hardware offerings. Intel® Kubernetes Service for multiuser accounts and cloud-native AI training and inference workloads is one of the new features.

Next Up: Intel’s objective to enhance HPC and AI is demonstrated by the new supercomputers that are being implemented with Intel Xeon CPU Max Series and Intel Data Center GPU Max Series technologies. The Italian National Agency for New Technologies, Energy and Sustainable Economic Development (ENEA) CRESCO 8 system will help advance fusion energy; the Texas Advanced Computing Center (TACC) is fully operational and will enable data analysis in biology to supersonic turbulence flows and atomistic simulations on a wide range of materials; and the United Kingdom Atomic Energy Authority (UKAEA) will solve memory-bound problems that underpin the design of future fusion powerplants.

These systems include the Euro-Mediterranean Centre on Climate Change (CMCC) Cassandra climate change modeling system.

The result from the mixed-precision AI benchmark will be foundational for Intel’s next-generation GPU for AI and HPC, code-named Falcon Shores. Falcon Shores will leverage the next-generation Intel Xe architecture with the best of Intel® Gaudi®.

This integration enables a unified programming interface.
Early performance results on Intel® Xeon® 6 with P-cores and Multiplexer Combined Ranks (MCR) memory at 8800 megatransfers per second (MT/s) deliver up to 2.3x performance improvement for real-world HPC applications, like Nucleus for European Modeling of the Ocean (NEMO), when compared to the previous generation,1 setting a strong foundation as the preferred host CPU choice for HPC solutions.