Exaflops barrier will be overcome in 2021
We are talking about the fastest supercomputer in the United States and other participants of the “exaflops race”.
Photo - OLCF at ORNL - SS BY / The most powerful supercomputer today - Summit
The development of the Aurora exaflop computer is handled by the US Department of Energy (DOE). It is being built at the Argonne National Laboratory near Chicago. Aurora will be five times more powerful than the current world leader - IBM's Summit computing system installed at Oak Ridge National Laboratory , which also belongs to the US Department of Energy.
')
A team of specialists from Cray (which HPE bought a month ago) and Intel are working on the project. According to experts, the contract value is approaching $ 500 million. It is this amount that the US government allocated to the DOE research department for the development of exaflops systems.
In general, Aurora was planning to launch in 2018 , but with much less computing power. However, the Ministry of Energy decided to increase the system performance five times and to postpone the release. The launch of a new supercomputer is expected in 2021.
According to the representative of the Argonne National Laboratory, the Aurora system (unlike other DOE supercomputers) will not solve the problems of the defense industry. Its power will be directed to the exploration of space and the effects of global warming, the search for drugs to treat cancer and the development of new materials for solar panels.
As the engineers say, one of the key features of Aurora will be intelligent algorithms. They will optimize the simulations. For example, a supercomputer will not simulate the effects of all possible drug variations. AI systems will pre-determine the most promising combinations, and the machine will deal with them first. This approach will speed up the development of medicines.
The Aurora supercomputer consists of 200 racks with hardware, which together take up almost 600 square meters. Each rack is a cluster system Cray Shasta. It is based on Intel Xeon Scalable processors. They have an integrated Ethernet controller, a hardware and software solution for creating RAID, and Intel VMD technology to integrate NVMe devices into logical storage.
Shasta also uses the Intel Xe graphics architecture. It helps accelerate scalar, matrix, vector and tensor operations in data centers. All supercomputer racks are connected to each other with the Cray Slingshot interface, which provides five times the bandwidth than traditional interconnects.
Engineers use liquid cooling to remove heat from iron. But while the developers do not disclose details about the power consumption of the new supercomputer.
According to analyst firm Hyperion Research, in the next three years, companies from different countries will spend $ 10 billion on exaflops systems. In addition to America, China, Japan and France are participating in the “exaflops race”.
The development of a supercomputer with a performance of one exaflops of the PRC began in 2017. Their machine is called Tianhe-3 and is built on the basis of Chinese Phytium processors. They have 64 cores and have a capacity of 512 gigaflops. Tianhe-3 will be used for sequence analysis of genes and protein structures and the development of new drugs. Its prototype is already being tested at the National Supercomputer Center in Tianjin. The supercomputer will be put into operation in 2020. However, a number of experts believe that its launch may be delayed by six months – year for technical reasons.
Photo - O01326 - SS BY-SA / Tianhe-2 - the previous version of the new supercomputer
In 2016, the Japanese Fujitsu announced its plans to create an ex-flop supercomputer by 2021. The Post-K computing system will be assembled on the basis of A64FX chips (ARM architecture). They have 52 cores - 48 of them perform computational operations, and the remaining four - management functions. According to the editors of the portal for the Top500 supercomputers, Japanese developers use more than 370 thousand of such processors to achieve performance in one exaflops.
In Europe, too, they are engaged in the development of "new generation supercomputers". In particular, the French manufacturer of computing equipment Atos will supply an exaflops computer to the CEA research institute. He is engaged in the design of nuclear reactors and research in the field of seismology. Sequana supercomputer built on the basis of Intel Xeon processors and NVIDIA graphics cards.
The emergence of exaflops machines awaits a large number of scientific projects. According to the participants of the Human Brain Project (studying the human brain), the use of new systems will be immediately. Therefore, it can be expected that soon more countries will begin to develop super-power supercomputers.
ITGLOBAL.COM provides private and hybrid cloud services, as well as solutions for analyzing the traffic of telecom operators. What we write about in the corporate blog:
Photo - OLCF at ORNL - SS BY / The most powerful supercomputer today - Summit
Aurora Project
The development of the Aurora exaflop computer is handled by the US Department of Energy (DOE). It is being built at the Argonne National Laboratory near Chicago. Aurora will be five times more powerful than the current world leader - IBM's Summit computing system installed at Oak Ridge National Laboratory , which also belongs to the US Department of Energy.
')
A team of specialists from Cray (which HPE bought a month ago) and Intel are working on the project. According to experts, the contract value is approaching $ 500 million. It is this amount that the US government allocated to the DOE research department for the development of exaflops systems.
In general, Aurora was planning to launch in 2018 , but with much less computing power. However, the Ministry of Energy decided to increase the system performance five times and to postpone the release. The launch of a new supercomputer is expected in 2021.
Machine assignment
According to the representative of the Argonne National Laboratory, the Aurora system (unlike other DOE supercomputers) will not solve the problems of the defense industry. Its power will be directed to the exploration of space and the effects of global warming, the search for drugs to treat cancer and the development of new materials for solar panels.
The Argonne Laboratory has already selected ten projects that will be the first to gain access to a high-performance system. Among them are programs to create a neural map of the brain and the study of dark matter.
As the engineers say, one of the key features of Aurora will be intelligent algorithms. They will optimize the simulations. For example, a supercomputer will not simulate the effects of all possible drug variations. AI systems will pre-determine the most promising combinations, and the machine will deal with them first. This approach will speed up the development of medicines.
How does it work
The Aurora supercomputer consists of 200 racks with hardware, which together take up almost 600 square meters. Each rack is a cluster system Cray Shasta. It is based on Intel Xeon Scalable processors. They have an integrated Ethernet controller, a hardware and software solution for creating RAID, and Intel VMD technology to integrate NVMe devices into logical storage.
Shasta also uses the Intel Xe graphics architecture. It helps accelerate scalar, matrix, vector and tensor operations in data centers. All supercomputer racks are connected to each other with the Cray Slingshot interface, which provides five times the bandwidth than traditional interconnects.
Engineers use liquid cooling to remove heat from iron. But while the developers do not disclose details about the power consumption of the new supercomputer.
Who else will overcome the exaflops barrier
According to analyst firm Hyperion Research, in the next three years, companies from different countries will spend $ 10 billion on exaflops systems. In addition to America, China, Japan and France are participating in the “exaflops race”.
The development of a supercomputer with a performance of one exaflops of the PRC began in 2017. Their machine is called Tianhe-3 and is built on the basis of Chinese Phytium processors. They have 64 cores and have a capacity of 512 gigaflops. Tianhe-3 will be used for sequence analysis of genes and protein structures and the development of new drugs. Its prototype is already being tested at the National Supercomputer Center in Tianjin. The supercomputer will be put into operation in 2020. However, a number of experts believe that its launch may be delayed by six months – year for technical reasons.
Photo - O01326 - SS BY-SA / Tianhe-2 - the previous version of the new supercomputer
In 2016, the Japanese Fujitsu announced its plans to create an ex-flop supercomputer by 2021. The Post-K computing system will be assembled on the basis of A64FX chips (ARM architecture). They have 52 cores - 48 of them perform computational operations, and the remaining four - management functions. According to the editors of the portal for the Top500 supercomputers, Japanese developers use more than 370 thousand of such processors to achieve performance in one exaflops.
In Europe, too, they are engaged in the development of "new generation supercomputers". In particular, the French manufacturer of computing equipment Atos will supply an exaflops computer to the CEA research institute. He is engaged in the design of nuclear reactors and research in the field of seismology. Sequana supercomputer built on the basis of Intel Xeon processors and NVIDIA graphics cards.
The emergence of exaflops machines awaits a large number of scientific projects. According to the participants of the Human Brain Project (studying the human brain), the use of new systems will be immediately. Therefore, it can be expected that soon more countries will begin to develop super-power supercomputers.
ITGLOBAL.COM provides private and hybrid cloud services, as well as solutions for analyzing the traffic of telecom operators. What we write about in the corporate blog:
Source: https://habr.com/ru/post/458176/
All Articles