In the tight and competitive race of AI development, developers are searching for new ways to give them the edge.
With Nvidia being a current AI leader due to its hardware offerings, some organisations are seeing high-performance computing, supercomputers, as a way to gain technological advancement.
In fact, AI developer Elon Musk is doing just that as a way to make his Gen AI system Grok better than its competitors.
Although not all supercomputers are AI, many possess dual capabilities or can stand to be added with such solutions.
With this in mind, we examine the top 10 supercomputers that exemplify the pinnacle of computational prowess.
10. Summit (IBM)
Summit, the IBM-built supercomputer housed at Oak Ridge National Laboratory in Tennessee, stands as a testament to the pinnacle of computing power. Launched in 2018, this technological marvel boasts an impressive 200 petaflops of peak performance, allowing it to process a staggering 200 quadrillion calculations per second. At its core, Summit comprises 4,608 nodes, each equipped with two IBM Power9 CPUs and six NVIDIA V100 GPUs, interconnected with NVIDIA's high-speed NVLink technology.
The supercomputer's prowess extends beyond raw computational power. Summit claimed the coveted top spot on the prestigious TOP500 list of the world's fastest supercomputers from November 2018 to June 2020. Its groundbreaking capabilities have been harnessed for a wide array of scientific endeavours, from simulating extreme weather patterns and earthquakes to advancing genomics research. Perhaps most notably, Summit played a crucial role in the fight against COVID-19, simulating 8,000 compounds in mere days to identify 77 small-molecule compounds that could potentially weaken the virus. This remarkable feat underscores Summit's immense value in accelerating scientific discovery and addressing global challenges.
9. Eos (NVIDIA)
NVIDIA's Eos supercomputer stands as a testament to the company's prowess in AI and high-performance computing. This behemoth of computational power boasts 576 DGX H100 systems, each equipped with eight H100 GPUs, totalling an impressive 4,608 GPUs. The system's architecture is designed to deliver a staggering 18.4 exaflops of AI computing performance, utilising FP8 precision. This remarkable feat positions Eos as a formidable contender in the realm of AI supercomputers, outpacing its predecessors by a significant margin.
At the heart of Eos lies the NVIDIA Quantum-2 400Gb/s InfiniBand network, which facilitates seamless communication between its components. This advanced interconnect enables Eos to achieve an Rmax of 121.4 PFlops/s on double precision LINPACK benchmarks. The supercomputer's design incorporates liquid cooling technology, evident in the rack manifolds visible behind the systems, ensuring optimal thermal management for sustained high-performance operation. Eos not only serves as a showcase of NVIDIA's technological capabilities but also as a practical tool for advancing research in climate science, digital biology and the future of AI.
8. Selene (NVIDIA)
Nvidia's Selene supercomputer stands as a testament to the company's prowess in high-performance computing. This behemoth, comprising 280 DGX A100 systems and a staggering 2,240 Tensor Core GPUs, secured the seventh spot on the Top500 list of the world's fastest computers in June 2020. Selene's impressive 63.4 petaflops performance on the HPL benchmark underscores its formidable capabilities, whilst its energy efficiency is equally noteworthy, achieving 20.52 gigaflops per watt and clinching second place on the Green500 list.
Beyond its raw computational power, Selene's versatility truly sets it apart. The system tackles a diverse array of tasks, from AI data analytics to traditional machine learning and HPC applications. Nvidia leverages Selene's capabilities for internal research, including chip design, model development, robotics, and self-driving car technology. Moreover, the supercomputer has been instrumental in advancing COVID-19 research, with Argonne National Laboratory utilising Selene's architecture to explore potential solutions to combat the virus. This multifaceted approach demonstrates Selene's significance not only as a technological marvel but also as a vital tool in addressing real-world challenges.
7. Perlmutter (HPE)
The Perlmutter supercomputer, housed at the National Energy Research Scientific Computing Center (NERSC), stands as a testament to the cutting-edge advancements in high-performance computing. Named after Nobel laureate Saul Perlmutter, this HPE Cray EX system boasts an impressive array of hardware, including 1,536 GPU-accelerated nodes and 3,072 CPU-only nodes. At its core, Perlmutter utilises AMD EPYC 7763 processors and NVIDIA A100 GPUs, delivering a staggering theoretical peak performance of 69.5 PFLOPS for GPU operations and 15.4 PFLOPS for CPU computations.
What truly sets Perlmutter apart is its revolutionary storage system. The supercomputer features a 35 PB all-flash Lustre file system, capable of delivering over 5 TB/sec of aggregate bandwidth and an astounding 4 million IOPS. This remarkable storage infrastructure combines the speed of a burst buffer with the stability of a traditional disk-based parallel file system, enabling researchers to process vast amounts of data at unprecedented speeds. Perlmutter's innovative design extends to its networking capabilities, employing the HPE Cray Slingshot interconnect, which provides high-speed connectivity between compute nodes and facilitates seamless access to external networks.
6. Sierra (IBM)
Sierra, the supercomputer unveiled by the National Nuclear Security Administration and Lawrence Livermore National Laboratory in 2018, stands as a testament to the cutting-edge advancements in high-performance computing. This remarkable machine, ranked as the world's third-fastest supercomputer upon its debut, boasts a peak performance of 125 petaFLOPS, which translates to an astounding 125 quadrillion floating-point operations per second.
At the heart of Sierra's impressive capabilities lies its heterogeneous architecture, combining 17,280 NVIDIA Tesla V100 graphics processing units with 8,640 IBM Power9 central processing units. This unique configuration allows Sierra to perform calculations six to ten times more efficiently than its predecessor, Sequoia, whilst consuming only marginally more power. The supercomputer's sophisticated design includes a 100-gigabit-per-second network for swift data transfer between processors and a substantial 1.38 petabytes of memory capacity, enabling it to tackle complex simulations and data-intensive calculations with unprecedented speed and accuracy.
5. LUMI (HPE Cray)
LUMI, the Large Unified Modern Infrastructure supercomputer, stands as a testament to Europe's commitment to advancing high-performance computing. Housed in Kajaani, Finland, this HPE Cray EX system boasts an impressive 362,496 cores and a theoretical peak performance exceeding 550 petaflops, securing its position as the fastest supercomputer in Europe as of January 2023.
At the heart of LUMI's architecture lies a powerful combination of 64-core AMD EPYC CPUs and AMD Radeon Instinct GPUs, chosen primarily for their cost-effectiveness and performance capabilities. The system's 1.75 petabytes of RAM and 117 petabytes of storage, including a 7-petabyte flash partition, underscore its formidable data processing potential. Notably, LUMI's environmental credentials are equally impressive, with its operations powered entirely by hydroelectric energy and its generated heat repurposed for local building heating, making it one of the world's most eco-friendly supercomputers.
4. Leonardo (Atos)
Leonardo, the Italian aerospace and defence company, has partnered with Atos, a global leader in digital transformation, to develop cutting-edge supercomputing technology. The collaboration has yielded impressive results, with the creation of one of Europe's most powerful supercomputers, installed at Leonardo's Genoa facility as part of the Leonardo Labs initiative.
This supercomputer, developed in conjunction with Atos, boasts a remarkable computing power of over 5 petaflops, equivalent to 5 million billion floating point operations per second. It is one of the first in Europe to utilise the latest generation NVIDIA A100 accelerators, featuring more than 100 supercomputing units. The system's storage capacity is equally impressive, at approximately 20 petabytes. This technological marvel serves as the nerve centre for Leonardo's research and development efforts, supporting innovation across traditional sectors and fostering the development of new technologies.
3. Fugaku (Fujitsu)
Fujitsu's Fugaku, the supercomputer named after Mount Fuji, has been making waves in the world of high-performance computing since its debut in 2020. Developed jointly by RIKEN and Fujitsu, this petascale powerhouse boasts an impressive 7,630,848 cores and a peak performance of 537.21 petaflops. Fugaku's heart lies in its A64FX processors, which are based on the ARM v8.2-A architecture and incorporate Scalable Vector Extensions (SVE) for enhanced computational capabilities.
What sets Fugaku apart is not just its raw power, but its versatility. The system has consistently topped global rankings, including the TOP500, HPCG, HPL-AI, and Graph500, demonstrating its prowess across various computational tasks. Its ability to deliver 2.0 exaflops in the HPL-AI benchmark showcases its potential for AI applications. Moreover, Fugaku's energy efficiency is noteworthy, with a power consumption of 29,899.23 kW, reflecting the developers' commitment to balancing performance with sustainability. This technological marvel has already been put to practical use, contributing to COVID-19 research and even powering the development of Japanese language AI models.
2. Aurora (Intel and HPE)
Intel and HPE have unveiled Aurora, a groundbreaking exascale supercomputer that stands as a testament to their collaborative prowess in high-performance computing. Installed at the U.S. Department of Energy's Argonne National Laboratory, Aurora boasts an impressive array of hardware, including 10,624 compute blades equipped with 21,248 Intel Xeon CPU Max Series processors and a staggering 63,744 Intel Data Center GPU Max units. This colossal system, occupying a space equivalent to two professional basketball courts, is poised to deliver unprecedented computational power, with expectations of achieving more than 2 exaflops of theoretical peak performance.
Aurora's capabilities extend far beyond raw processing power. The system incorporates over 1,024 storage nodes utilising Intel's distributed asynchronous object storage (DAOS), providing a whopping 220 PB of capacity at 32 TBps of total bandwidth. Its unified memory architecture and HPE Slingshot 11 interconnect ensure seamless data flow across the vast network of processors. Notably, Aurora has already demonstrated its prowess by reaching 1.012 exaflops on 87% of the system, securing its position as the second-fastest supercomputer globally, just behind Frontier. This remarkable achievement underscores Aurora's potential to revolutionise scientific research across various fields, from brain mapping and particle physics to fusion reactor simulations.
1. Frontier (HPE Cray)
Frontier, the HPE Cray supercomputer housed at Oak Ridge National Laboratory, stands as a testament to the pinnacle of high-performance computing. This exascale behemoth, capable of performing more than a quintillion calculations per second, has held the top spot on the Top500 list since June 2022. Frontier's impressive performance is underpinned by its architecture, which combines AMD EPYC processors with AMD Instinct accelerators, connected via HPE's Slingshot interconnect.
The system's prowess extends beyond raw computational power. Frontier has demonstrated remarkable energy efficiency, leading the Green500 list with a performance of 62.68 gigaflops per watt. This achievement underscores the importance of balancing performance with sustainability in the era of exascale computing. Frontier's infrastructure is equally impressive, boasting 74 HPE Cray EX cabinets, each weighing over 8,000 pounds, and utilising 90 miles of networking cables. The supercomputer's storage capabilities are equally formidable, with the Orion system providing 700 petabytes of capacity and peak write speeds exceeding 35 terabytes per second.
Explore the latest edition of AI Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.
Discover all our upcoming events and secure your tickets today.
AI Magazine is a BizClik brand
