The 10 Most Powerful Supercomputers in the World Today

Sunway TaihuLight is a Chinese supercomputer. As of November 2018, it ranked third in the TOP 500 list with a LINPACK performance score of 93 petaflops. This speed is nearly three times faster than the previous Tianhe-2 supercomputer. Designed by the National Research Center of Parallel Computer Engineering and Technology (NRCPC), it is located at the National Supercomputing Center in Wuxi, Jiangsu Province, China.

Once the world's fastest supercomputer from June 2016 to June 2018, according to the TOP 500 list, it was surpassed by IBM's Summit in June 2018. Sunway TaihuLight is powered by 40,960 SW26010 64-bit RISC processors based on the Sunway architecture, each containing 256 processing cores and four auxiliary cores for system management, totaling 10,649,600 processing cores. The system runs on its own operating system, Sunway RaiseOS 2.0.5, based on Linux, and uses custom OpenACC 2.0 to support parallel programming. TaihuLight was developed solely to advance human technological ambitions. Fifteen years ago, China was a small player in global technology, but now it boasts some of the most prominent names across all fields, even earning recognition in the United States. The Sunway TaihuLight is a clear demonstration of China's technological prowess, far exceeding American capabilities by five times. More importantly, this achievement is a result of China's own processor designs.

Nvidia recently unveiled the assembly process for the Selene supercomputer, which was ranked the 7th fastest in the world as of June. The entire supercomputer was assembled during the pandemic in just weeks, with a team of only six workers adhering to social distancing guidelines and aided by a robot assistant named Trip.

Selene is a unique supercomputer in its design.

It uses Nvidia's SuperPOD DGX GPU-accelerated architecture, unlike most supercomputers in the Top500 list that rely on customized CPU designs. Selene ranks second on the Green500 list of the most energy-efficient supercomputers. Its configuration includes 560 AMD Epyc 7742 CPUs and 2,240 Nvidia A100 GPUs. Theoretical peak performance is 35,000 teraflops, and its efficiency was a key factor in enabling such a rapid assembly.

Nvidia's main task was to position each DGX pod, connect the cables, and ensure everything worked correctly. While wiring a supercomputer is no easy feat, Nvidia used Mellanox's InfinitiBand circuits to minimize cable requirements while maximizing system bandwidth.

Tianhe-2 is a 33.86 petaflop supercomputer located at the National Supercomputing Center in Guangzhou, China. It was developed by a team of 1,300 scientists and engineers. Tianhe-2 was ranked the world's fastest supercomputer in the TOP500 list from June 2013 through November 2015. However, this record was surpassed by Sunway TaihuLight in June 2016. In 2015, the planned collaboration between Sun Yat-sen University and the Guangzhou government to double its computing power was halted after the U.S. government denied Intel's export license for the required CPUs and co-processor boards.

In response to U.S. sanctions, China introduced the Sunway TaihuLight in 2016, a far more powerful machine, which now holds the fourth spot on the TOP500 list. The system also uses entirely domestic technology, including Sunway's many-core processors. Tianhe-2A was developed by the National University of Defense Technology (NUDT) in China. However, China is believed to already have at least two exascale systems, Sunway Oceanlite and Tianhe-3, in development.

Researchers have criticized Tianhe-2 for its usability. Chi Xuebin, deputy director of the Computer Network and Information Center, stated that while it leads the world in computational power, its functionality still lags behind that of supercomputers in the U.S. and Japan. Some users may need years—or even a decade—to write the necessary code. Additionally, Tianhe-2's location in southern China, with its warmer climate, increases power consumption by around 10% compared to locations in northern China.

JUWELS Booster Module is a Blue Gene/Q system supercomputer developed by IBM. Funded by the Helmholtz Association and the Gauss Supercomputing Center (GCS), with equal contributions from federal and state funds from North Rhine-Westphalia, it was launched in 2012 at Forschungszentrum Jülich as the successor to the JUGENE supercomputer.

JUWELS is the fastest supercomputer in Europe and ranks fifth on the TOP500 list of the most powerful supercomputers. It was also one of the world’s most energy-efficient systems at the time, securing the fifth spot on the Green500 list. With 458,752 processing cores, it offers a peak computational power of 5.9 petaflops. JUWELS has been used for several research projects, including the Human Brain Project. It ceased operation in May 2018 after six years of service, replaced by the next-generation JUWELS system. The JUWELS Booster Module, built by Atos, is the latest addition to the system. The newly installed BullSequana machine at Forschungszentrum Jülich (FZJ) in Germany is the most powerful in Europe, with a performance of 44.1 petaflops HPL.

JUWELS is based on a modular architecture and is powered by AMD processors and NVIDIA GPUs, similar to the Selene system. The system’s design features strong operational capabilities, equipped with AMD processors and NVIDIA GPUs to deliver high performance.

HPC5 is a supercomputer built by Dell and installed by Eni, capable of reaching 51.721 petaflops and ranking 9th in the TOP500 list as of November 2021. It is located at the Green Data Center in Ferrera Erbognone, northern Italy. In June 2020, HPC5 ranked 6th in the Green500 list.

HPC5 is an upgraded version of the HPC4 system, developed by Hewlett Packard Enterprise and used by Eni, also known as HPC4+. The system spans 1,820 Dell EMC PowerEdge C4140 servers, each equipped with two 24-core Intel Gold 6252 processors and four Nvidia V100 GPU accelerators. In total, it includes 7,280 Nvidia V100 GPUs. The primary task of HPC5 is to process extremely complex internal algorithms to analyze subsurface data. HPC5 has a peak processing power of 70 petaflops, meaning it can perform 70 quadrillion operations per second.

HPC5 is equipped with four CPU sockets, over 3,400 processors, and 10,000 graphics cards, enabling the creation of 3D models of the Earth's crust at depths of 10 to 15 km, covering areas of hundreds of square kilometers with a resolution of several dozen meters. The system achieves speeds up to 35,450 petaflops, with peak performance reaching 51.721 petaflops.

If you're interested in the world of supercomputing, you might recall that in August of last year, Dell EMC and Intel announced a collaboration that captured the attention of experts and tech enthusiasts worldwide. These two industry giants came together to design and develop the world’s most powerful academic supercomputer, named Frontera. The project received a substantial $60 million grant from the U.S. National Science Foundation and was chosen to replace Stampede2, an aging supercomputer that had served for many years at the Texas Advanced Computing Center in Austin.

Intel experts stated that the Frontera supercomputer could achieve a peak performance of 38.7 petaflops, or 38.7 quadrillion floating-point operations per second, making it one of the fastest machines on the planet. It was specifically designed for heavy academic workloads like modeling and simulation, big data, and machine learning. To put this into perspective, the Stampede2 supercomputer delivered 18 petaflops, meaning Frontera is roughly twice as fast. Additionally, Frontera incorporates Intel Optane DC, a breakthrough memory technology developed by Intel and Micron Technology, which works in tandem with DDR4 memory and large cache pools to improve overall system performance.

Moreover, the combination of Intel Optane DC with the latest Xeon Scalable processors allows Frontera to achieve an astonishing processing speed of 287,000 operations per second, compared to the 3,116 operations per second of typical DRAM-based systems. With this level of capability, Frontera’s reboot time is an impressive 17 seconds.

Today, Aramco and stc have unveiled the Dammam 7 supercomputer, a new addition to the top 10 most powerful supercomputers in the world. This advanced system opens up new possibilities in exploration and development, while enhancing Aramco’s decision-making capabilities for exploration and investment. It marks a key milestone in Aramco’s digital transformation, complementing a suite of cutting-edge technologies that are reshaping core operations, driving efficiency, and reinforcing its leadership in geological science.

Developed in the Dhahran Technology Valley through a collaboration between Solutions (a subsidiary of stc Group) and CRAY (a subsidiary of Hewlett Packard Enterprise), the Dammam 7 boasts a peak computational power of 55.4 petaflops, enabling it to process and visualize the world’s largest geophysical datasets. Named after the first oil well discovered in Saudi Arabia, Dammam 7 transcends traditional exploration and production boundaries through advanced technology. Sophisticated imaging algorithms and deep learning capabilities will allow it to run highly detailed 3D Earth models, enhancing the company’s ability to detect and recover oil and gas, while minimizing exploration and development risks.

Furthermore, it will greatly improve decision-making for the exploration and development of conventional and unconventional hydrocarbon resources, as well as guide future investments in production and resource allocation.

Fugaku is a petascale supercomputer located at the RIKEN Center for Computational Science in Kobe, Japan. Its development began in 2014 as the successor to the K computer and was officially launched in 2020. Named after Mount Fuji, Fugaku quickly became the world’s fastest supercomputer in the June 2020 TOP500 list, marking the first ARM-based machine to achieve this distinction. At that time, it also achieved a remarkable performance of 1.42 exaFLOPS using the HPL-AI benchmark. Fugaku became operational in full capacity in 2021.

In May 2022, Fugaku was dethroned by Frontier, which claimed the title of the fastest supercomputer. However, there have been reports since May 2021 suggesting that China had developed and secretly operated a supercomputer with performance surpassing Fugaku by more than double, using its own proprietary technology. This Chinese system has reportedly been kept under wraps to avoid U.S. sanctions. A paper utilizing Fugaku won the prestigious Gordon Bell Prize for the best scientific paper. By May 2022, it was believed that China was operating at least two exascale supercomputers.

Built on the latest ARM architecture, Fugaku executes over 415 quadrillion calculations per second, making it 2.8 times faster than the previous leader, the Summit supercomputer built by IBM. Fugaku is now playing a key role in COVID-19 research, including diagnostics, treatment strategies, and simulations of the SARS-CoV-2 virus's spread.

Summit, also known as OLCF-4, is a supercomputer developed by IBM for the Oak Ridge National Laboratory. As of November 2019, it was the fastest supercomputer in the world, achieving 200 petaFLOPS. Its current LINPACK benchmark performance stands at 148.6 petaFLOPS. Additionally, it was ranked as the third most energy-efficient supercomputer globally, with an energy efficiency of 14.668 gigaFLOPS per watt.

IBM Summit was the first supercomputer to achieve exaFLOP (one quintillion, 10¹⁸ operations per second), reaching 1.88 exaFLOPS in genomics analysis and expected to reach 3.3 exaFLOPS with mixed-precision operations. Summit empowers scientists and researchers to tackle complex tasks in areas such as energy, artificial intelligence, healthcare, and more. Each of Summit’s 4,608 computing nodes is equipped with over 600 GB of shared memory accessible by all CPUs and GPUs, along with 800 GB of static RAM that can be used as a burst buffer or expanded memory.

The CPUs (Power) and GPUs (Volta) are interconnected via Nvidia's high-speed NVLink, allowing Summit to operate with a heterogeneous computing model. Summit is dedicated to civilian scientific research and is based at the Oak Ridge National Laboratory in Tennessee, where it is utilized for projects in fields like astronomy, medicine, and climate science.

Sierra is a supercomputer developed by IBM for the Lawrence Livermore National Laboratory (LLNL) in California. As of now, it holds the second position on the TOP500 list. Sierra is powered by 1,572,480 IBM Power9 processors, enhanced with 1,382,400 Nvidia Volta GV100 accelerators. Its performance has remained stable at 94.6 petaflops, consistent over the past six months.

Located at LLNL in California, Sierra features a high-performance LINPACK score of 94.6 petaflops. Each of its 4,320 nodes is equipped with 2 Power9 CPUs and 4 Nvidia Tesla V100 GPUs, mirroring Summit's architecture. Supercomputers like Sierra are massive systems, often occupying entire rooms or buildings, and consist of thousands of processors to handle vast streams of data. These machines are typically used in fields such as scientific research, astronomy, climate change forecasting, nuclear simulations, and medical studies, including cancer research.

Sierra also ranks 15th on the Green500 list of the world’s most energy-efficient supercomputers. It is utilized for applications such as nuclear weapons simulations, climate modeling, and cancer therapy research. Both Summit and Sierra are powered by IBM technology, using IBM Power9 processors and Nvidia Tesla V100 GPU accelerators.