T-Platforms’ completed development of a prototype supercomputer multipetaflopsnogo

"T-Platforms’ multipetaflopsnogo completed the development of a prototype supercomputer commissioned by the Ministry of Education

The company "T-Platforms", international supercomputer developer and supplier of a complete range of solutions and services for high-performance computing, announced the completion of the project to create a prototype of a computer system to create a new generation of supercomputers.

Design and manufacture of high-tech solutions were carried out with the financial support of the Ministry of Education and Science of the Russian Federation within the framework of a state contract "Development of hardware and software multiprocessor computing system (ERM) for the construction of high-performance computing systems multipetaflopsnogo range."

The main advantages of this solution are the high efficiency and packing density of compute nodes, greatly exceeding these characteristics horizontally scalable servers.

The system is cooled with hot water and the air, allowing to achieve optimum balance of energy efficiency, reliability and cost solutions. Key components such as compute nodes and the switching system, cooled by the temperature difference between the inlet and outlet streams of water. As a result, specialized computer rack can take 150-200 kW heat which is 5-6 times higher than in standard rack systems with air-cooled.

Peak performance of 400 teraflops provided by the configuration based on single-processor motherboards with the accelerator. Dual-processor motherboards allow achieve the target of more than 100 teraflops. In a rack 512 mounted to the computing nodes combined in modules with a common radiator. Each radiator provides cooling of the boards 2 to 4 electronic components. Radiators are connected to the general system of circulation of hot water in the rack.

ERM has in its traditional nomenclature such thing as a server chassis. The system components are mounted directly to the rack. Backplanes associate compute nodes, switches, and all levels of management modules into the system without using a cable. This increases the reliability of the inter-node communications, driving the "hot swapping" of components and makes it possible to speed up the service complex.

The system is designed to create supercomputers petaflopcnogo multipetaflopsnogo and capacity ranges. In the framework of the existing interconnect maximum computing cluster based on the IAC can be up to 128 computing racks.

Vsevolod Opanasenko, CEO of the "T-Platforms": "The solution, created with funding from the Ministry of Education, has significant potential for the introduction of both the Russian and the international market. On the basis of the prototype will be developed and manufacturing of large-scale HPC systems. "

Like this post? Please share to your friends: