Supercomputers are computer systems capable of performing extremely fast and large-scale computations, often consisting of thousands or millions of processor cores capable of handling complex scientific and engineering problems. Supercomputers have important applications in various fields, such as weather forecasting, climate simulation, biomedicine, materials science, energy research, artificial intelligence, etc.
Currently, the world's most powerful mainframe computer is the Frontier system at Oak Ridge National Laboratory (ORNL), which tops the list with a speed of 1.102 EFlop/s, making it the only truly exascale computer on the list. It was unveiled at the High Performance Computing Professional Conference (SC22) held in Dallas, USA in November 2022. Built by AMD and supercomputer maker HPE Cray for Oak Ridge National Laboratory, it cost $600 million to make advanced calculations in areas such as nuclear and climate research.
China has two supercomputing systems – Sunway Taihu Light and Tianhe 2A ranked 7th and 10th respectively. Sunway Taihulight is the world's first supercomputer that uses all processor chips with China's independent intellectual property rights, with more than 10 million processor cores, dual-precision floating point peak of up to 125PFlops (1.25 billion billion operations per second), and stable performance of 93PFlops. It first topped the world supercomputer rankings in 2016 and has won four consecutive championships.
It is mainly used for scientific research in climate simulation, life sciences, material design, and other fields. Tianhe 2A is the world's first supercomputer with peak performance exceeding 100 PFlops, with nearly 5 million processor cores, dual crystal floating point peak up to 100.68PFlops, and stable performance of 61.44PFlops. It first topped the world supercomputer rankings in 2013 and has won the championship six times in a row. It uses the Kirin operating system, currently uses Intel processors, and plans to replace them with domestic processors in the future. It is mainly used in lunar exploration engineering, manned spaceflight, oil exploration, automobile and aircraft design and manufacturing, gene sequencing and other government scientific research projects and civilian fields.
In addition to the United States and China, other countries and regions are also developing their own supercomputers. For example, Fugaku of the RIKEN Computing Science Center (R-CCS) in Japan is in second place, it is the world's first supercomputer with an ARM architecture, with 7.63 million processor cores, a dual-precision floating-point peak of up to 537.21PFlops, and a stable performance of 442.01PFlops. It debuted at the top of the world supercomputer rankings in 2020 and topped the charts for two consecutive years until it was squeezed out by Frontier.
It is mainly used to solve social problems in the fields of new coronavirus research, disaster prevention, and energy development. Finland's CSC-hosted LUMI is in 3rd place, it is the fastest supercomputer in Europe, with 2.22 million processor cores, dual crystal floating point peak up to 428.7PFlops, and stable performance of 309.1PFlops. It started operations in November 2021 and is mainly used in areas such as climate research. Leonardo of EuroHPC/CINECA, Bologna, Italy, ranked 4th, it is the world's first supercomputer with Nvidia Grace Hopper Superchip, with 1.46 million processor cores, dual floating point peak up to 255.75PFlops, and stable performance of 174.7PFlops. It began construction in November 2021 and is mainly used in fields such as energy, materials, pharmaceuticals, etc.
What is the trend of supercomputers? According to the analysis and predictions of some experts and observers, future supercomputers may appear changes and innovations in the following aspects:
Simulate the world: Supercomputers will be able to simulate increasingly real and complex virtual worlds, including phenomena and laws at all levels of physics, chemistry, biology, and society. These simulations will help humans better understand and explore the natural world and human society, and will also provide more entertainment and educational possibilities for humans.
Weather systems and complex models: Supercomputers will be able to predict global problems such as weather changes and warming more accurately and for the longer term, and will also be able to process more data and variables to build more complex models such as earthquakes, tsunamis, volcanoes, ecosystems, etc.
Artificial brain: Supercomputers will be able to simulate the structure and function of the human brain, even beyond the capabilities of the human brain, to achieve higher levels of cognition and intelligence. These artificial brains will be able to learn and improve themselves, as well as be able to communicate and cooperate with humans.
Green supercomputers: Supercomputers will be able to operate more energy-efficient and environmentally friendly, requiring less power and cooling systems, and also reducing environmental impact. These green supercomputers will use new materials and technologies, such as photonic chips, quantum chips, biochips, and more, to improve efficiency and performance.
Exascale computers and above: Supercomputers will be able to reach or exceed exascale computing power, which is to perform exascale operations per second. This will open up a new era of science and technology for mankind, allowing mankind to solve some difficult problems that cannot be solved at present, such as dark matter, dark energy, protein folding, human brain, etc. At present, the United States, China, Japan, the European Union, etc. are actively developing and building exascale supercomputers, which are expected to be put into use before 2023. Above the exascale, there are higher goals, such as the 100 billion scale (Zettascale) and the trillion (Yottascale), which may require more advanced technologies and theories to achieve, and may also bring more challenges and opportunities.
NVIDIA GH200 super-AI computer
The Nvidia GH200 is a new AI supercomputer that combines 256 Nvidia Grace Hopper Superchips into a single GPU, providing 144TB of shared memory space. It is mainly used to handle huge AI models in the fields of large-scale recommendation systems, generative AI and graph analysis.
The Nvidia Grace Hopper Superchip is a new type of chip that integrates Nvidia Grace CPUs and Nvidia Hopper GPUs in the same package, eliminating traditional PCIe CPU-to-GPU connections, improving bandwidth and reducing power consumption. The Nvidia Grace CPU is a high-performance processor based on the ARM architecture, designed for big data and AI applications. Nvidia Hopper GPUs are next-generation graphics processors based on Multi-Chip Module (MCM) technology that deliver higher performance and scalability.
The Nvidia GH200 is developed and produced by Nvidia, but is not controlled by any one country or organization. It is a neutral and independent existence that can communicate and cooperate with other supercomputers or humans, and can also learn and innovate autonomously. It is currently one of the most advanced AI supercomputers in the world and an important direction for the development of future supercomputers.