For several years, some members of the HPC community have suspected that China was hiding its true supercomputer capabilities from the world. These suspicions were confirmed by the publication last week of a research paper in which Chinese university researchers indicate that 40 million heterogeneous nuclei inside China’s Sunway supercomputer were assigned to a simulation based on deep learning.
The reaction of Steve Conway, senior advisor, HPC Market Dynamics, to Hyperion Research, reflects a sentiment shared by many observers in the supercomputing industry. “This development adds credibility to the idea that China may have deliberately failed to report Linpack’s exascale results for the November 2021 Top500 list to avoid attracting further restrictions from the U.S. government,” says Conway.
The Sunway TaihuLight system was developed by the National Research Center of Parallel Computer Engineering & Technology (NRCPC) in China and installed at the National Supercomputing Center in Wuxi, Jiangsu Province, China. According to the latest TOP500 list of the world’s most powerful supercomputers, Sunway TaihuLight ranks fourth with a stated performance of 93 pFLOPS. However, China has reported this level of performance since at least 2016, when it was ranked as number one. 1 in the world.
The performance required for Sunway’s deep learning quantum simulation has not been reported. “We report that a simulation protocol based on deep learning can achieve the solution with state-of-the-art accuracy in Hilbert space as large as 21296 for the spin system and 3144 for the fermion system, using a hybrid framework HPC-AI on the new Sunway supercomputer, “said researchers at Chinese universities.
With a high scalability of up to 40 million heterogeneous cores (sw26010pro), our applications measured 94% weak scaling efficiency and 72% strong scaling efficiency. work paves the way for the simulation of spin and fermion models in latticework of unprecedented size, with extreme precision. ”
While many assume that China does not report the performance of its supercomputers, another theory is that China is lagging behind the United States in the exascala race and has not shared the recent performance of its systems for not feeling embarrassed. Depending on the interpretation of the quantum simulations of many Sunway bodies, both theories could be supported.
HPC in Europe
In a press release issued on December 21, LightOn, a French startup that develops photonic computers for artificial intelligence, announced the integration of one of its photonic coprocessors into the supercomputer Jean Zay. In addition to being one of the most powerful computers in the world, currently ranked 105th in the Top500, the French supercomputer Jean Zay is now the first high-performance computer (HPC) to have a photon coprocessor (which can run on light instead of electric current).
The LightOn photocopier will be available to selected users in the Jean Zay research community for the next few months. LightOn’s OPU (Optical Processing Unit) uses photonics to accelerate random algorithms on a very large scale, while allowing researchers to access quality data.
LightOn’s optical processing unit is the first commercially available AI accelerator photon chip for non-von Neumann scale calculations, reaching 1500 teraOPS. It is based on a combination of free space optics with commercial components, as well as a software API that allows seamless integration with Python-based processing channels. LightOn works in a variety of use cases and hybrid network architectures, where the OPU is used in combination with the CPU / GPU.
- CPU: for Central processing unit or central processing unit, it is considered the brain of a computer, because it controls and executes all operations. CPUs perform all kinds of data processing operations; without them, a computer cannot function properly. In the early days of Bitcoin, CPUs were widely used. As mining became more competitive, this was no longer possible. However, there are still other cryptocurrencies today that can be extracted with a CPU, such as Loki and Nimiq;
- GPU: for the graphics processing unit or graphics processing unit, they are more powerful than CPUs when performing certain functions. GPUs are supposed to be efficient for performing large amounts of computation. This feature makes them very suitable for mining certain cryptocurrencies, such as Ethereum. Although it is technically possible to extract Ethereum with a CPU, this habit would be strongly discouraged. GPUs are much more efficient at exploiting Ethereum and other cryptocurrencies such as Ravencoin and Beam;
- OPU: they are types of processors based not on electronics but on optics. They are very promising in terms of performance, as the information is transmitted by light. We could say, in a reductive way, that the OPU is to the traditional processors what is the optical fiber to the electrical cables.
The European Union published on 17 December a call for expressions of interest for the selection of a host entity for a high-end supercomputer. The EU says the purchase “will support the development and adoption of innovative and competitive, demand-driven, user-driven supercomputing systems based on a supply chain that will ensure components, technologies and knowledge that limit the risk of disruption and the development of a wide range of applications optimized for these systems ”.
The European Joint Undertaking for High Performance Computing (EC EuroHPC) was established by an EU Council Regulation of 13 July 2021 and entered into force on 8 August 2021. According to Article 3 of the Regulation, the mission of the EC EuroHPC is to develop, deploy, extend and maintain in the Union a federated, secure and hyperconnected ecosystem of supercomputers, quantum computing. Its mission is also to support the development and adoption of innovative and competitive supercomputing systems for European science and technology.
“The Perlmutter system will play a key role in advancing scientific research in the United States and is at the forefront of a number of critical technologies, such as advanced computing, artificial intelligence and data science,” he said. a spokesman for Lawrence Berkeley National Laboratory. The system will also be widely used in climate and environmental studies, clean energy technologies, semiconductors and microelectronics, and quantum information science. “
Sources: Tsinghua University, research article
What is your opinion on the subject?
Atos announces the new exascale BullSequana XH3000 supercomputer, with six times the power of the previous generation, available in the fourth quarter of 2022
Europe now wants its own high-powered supercomputer, with 50% coverage of acquisition costs and 50% coverage of operating costs.
French supercomputer Jean Zay is now the first HPC that can run on light, instead of electricity