Top 10 Energy-Efficient Supercomputers

In June, Top500 released its biannual Green500 ranking, featuring the world’s most energy-efficient supercomputers.

The Green500 list has evolved since the early 2000s, chronicling a positive trend of companies achieving higher efficiency and sustainability benchmarks over the years. The latest data reveals that high-performance computing (HPC) systems are inching closer to a median efficiency of 30 gigaflops per watt, or 30 billion floating-point operations for every watt of energy consumed.

The high turnover between the Green500 November 2022 and June 2023 rankings suggests an intense competition to match impressive computing power with ambitious sustainability goals.

The latest Green500 ranking was unveiled at ISC High Performance 2023 in Hamburg, Germany. Here’s a closer look at each of the energy-efficient supercomputers featured on the prestigious list.

#10. Atos THX.A.B.

French IT giant Atos developed the THX.A.B. system using its BullSequana XH2000 HPC platform. Atos THX.A.B. achieves an energy efficiency of 41.411 Gflops per watt. The BullSequana XH2000 is an evolution of the XH1000 model and incorporates a patented Direct Liquid Cooling solution, which cools all critical components of the system.

Atos THX.A.B. debuted at the 5^th position on the Green500 list in June 2022. However, its ranking gradually declined to 8th in November of the same year and further to 10th in the most recent June rankings.

The system runs on a Xeon Platinum 8358 32C 2.6GHz processor with 25,056 CPU cores. Additionally, Atos THX.A.B. is equipped with a Quad-rail NVIDIA HDR100 Infiniband interconnection. It achieved a peak efficiency performance of 3.5 PFlops/s, which translates to 3.5 quadrillion floating-point operations per second.

#9. Goethe-NHR

Goethe University Frankfurt’s system is a recent addition to Germany’s National High Performance Computing Alliance. The Goethe-NHR system secured the No. 9 spot on Green500’s June list due to its energy efficiency of 46.543 Gflops per watt. The system was jointly manufactured by MEGWARE and Supermicro.

Goethe NHR_Goethe University

The Goethe-NHR system is designed to support computationally intensive research in the sciences. The upgraded system runs on an AMD EPYC 7452 32C 2.35GHz processor and includes high-performance IBM Power9 CPUs. For efficient communication between components, it uses the Mellanox InfiniBand EDR interconnect.

#8. amplitUDE (GPU partition)

Researchers in engineering and science at the University of Duisburg-Essen in Germany will have access to the future amplitUDE mainframe, which ranks as the 8th most energy-efficient supercomputer globally. The supercomputer, manufactured by MEGWARE, stands out for its innovative fanless compute nodes and direct water-cooling system.

During performance tests in Chemnitz, Germany, the amplitUDE GPU partition showed an energy efficiency of 51.343 GFlops per watt. The university website notes that it is the most energy-efficient system in Germany.

The amplitUDE operates on a Xeon Platinum 8480+ 56C 2GHz processor and is built with 6,768 CPU cores. Moreover, it is equipped with an Infiniband NDR interconnection. 

#7. LUMI

The LUMI HPE CRAY EX235A system is ranked as the 3^rd most powerful and the 7^th  most energy-efficient supercomputer. The HPC is manufactured by Hewlett Packard Enterprise (HPE) and is located in Kajaani, Finland.

Notably, the waste heat from the system is used to heat nearby homes. The naturally cool climate of the region also contributes to the system’s efficiency as it requires less energy for cooling.

LUMI’s system runs entirely on carbon-free hydroelectric energy. It boasts an efficiency ratio of 51.382 Gflops per watt.

The LUMI is powered by an AMD Optimized 3^rd Generation EPYC 64C 2GHz processor and an AMD Instinct MI250X accelerator. It contains a total of 2.2 million cores and a Slingshot-11 interconnect. 

#6. Frontier

The Frontier supercomputer, located at the Oak Ridge National Laboratory (ORNL) in Tennessee, has the distinction of being the world’s first exascale computer. Its development is the result of over a decade of collaboration between scientists, academia, and private industry. Notably, the Department of Energy’s Exascale Computing Project played a key role in this collaborative effort.

This year, the Frontier supercomputer achieved an energy efficiency of 52.592 Gflops per watt, thanks in part to its innovative cooling system, high-performance AMD Optimized 3rd Generation EPYC 64C 2GHz processor, and AMD Instinct MI250X accelerator.

The installation and testing of the 74 HPE Cray EX supercomputer cabinets took place during COVID-19 lockdowns. Since its debut on the Top500 in June 2022, Frontier has consistently remained one of the most powerful and energy-efficient systems worldwide.

#5. Dardel GPU

The GPU partition of the Dardel HPC system at the KTH Royal Institute of Technology in Sweden has been available to researchers since the spring of this year. It serves as a massively powerful resource for various fields such as computational fluid dynamics, climate modeling, AI, and more.

dardel_royal

The Dardel GPU system, manufactured by HPE, consists of 56 GPU nodes and achieves an energy efficiency of 56.491 Gflops per watt. It has 52,864 cores and runs on an AMD Optimized 3rd Generation EPYC 64C 2GHz. Additionally, the system uses a Slingshot-11 interconnect with Dragonfly topology.

#4. Setonix – GPU

The Setonix GPU system is located at the Pawsey Supercomputing Centre in Kensington, Western Australia. It is modeled on the architecture of other cutting-edge exascale computer projects: an HPE CRAY EX235A system, AMD Optimized 3rd Generation EPYC 64C 2GHz processor, AMD Instinct MI250X accelerator, and a Slingshot-11 interconnect.

setonix-pawsey-abc

In July 2022, the Setonix GPU system launched its first phase and scored an energy efficiency rating of 56.983 Gflops per watt.  

 #3. Adastra

Like several other top energy-efficient supercomputers in the Green50 list, GENCI-CINES’ Adastra is an HPE Cray EX235a system with an AMD processor and accelerator and Slingshot-11 interconnect.

Since May 2022, French researchers have had access to Adastra for numerical simulation, high-performance data analytics, and AI tasks.

Adastra, which takes its name from the Latin phrase “to the stars,” achieved an efficiency ratio of 58.021 Gflops per watt.  

#2. Frontier TDS 

The Frontier Test & Development System (TDS) is just one rack of the full Frontier system at Oak Ridge. The TDS achieved an efficiency of 62.684 Gflops per watt. That score is more than twice as efficient as the top-ranking supercomputer from just a few years ago, highlighting the HPC market’s rapid sustainability advancements. 

#1. Henri 

For the second time, the Henri system in New York City’s Flatiron Institute has claimed the top spot on the Green500 list, solidifying its position as the most energy-efficient HPC system to date. The Henri system has a power efficiency ratio of 65.396 Gflops per watt.

Flatiron

Developed by Lenovo and NVIDIA, Henri is based on the ​​Lenovo ThinkSystem SR670 with Intel Xeon Platinum and Nvidia H100 components.

Top500 also notes that Henri showed vast improvements in performance between its first and second appearances on the Top500 list, leaping from the 405^th spot to the 244^th position. As a part of the elite group of ultra-powerful and increasingly energy-efficient supercomputers, Henri will undoubtedly serve as a critical tool for researchers pushing the boundaries of machine learning, quantum physics, and climate modeling.