Data center FEB RAS Superkompyyuter

The supercomputer placed in the Khabarovsk Computer center of the FEB RAS is created based on the hybrid architecture developed by the staff of data center with a peak capacity of 55 trillion operations per second. The computer system is focused on solving of tasks in the field of artificial intelligence and machine learning, including algorithm elaboration of computer vision and image processing.

Technical characteristics (on one node)

• 2 10-core processor IBM Power 8 with a frequency of 2.86 GHz (only 160 flows);
• Memory of ECC, 256 GB;
• 2 x 1 TB 2.5" 7K RPM SATA HDD;
• 2 x Nvidia Tesla P100 GPU, NVLink.
• Data networks: EDR InfiniBand.
• Managing network: Gigabit Ethernet.
• Performance of a cluster (Tflops): 55.84 (peak), 40.39 (Linpack)

Performance is provided with graphic Nvidia Tesla P100 accelerators on architecture of Pascal, and data exchange between the central processor IBM Power 8 and the graphic processor is performed via the high-speed NVLink interface with capacity to 80 Gbytes per second.

Peak capacity of the supercomputer reaches 55 trillion operations per second

2017: Start

On July 6, 2017 it became known of start in the Khabarovsk Computer center of the FEB RAS (Data center FEB RAS) with assistance of the Federal Agency for Scientific Organizations (FASO Russia) of the computer system with a performance of 55 trillion operations per second.^[1]

According to Alexey Sorokin, the head of TsKP "Data Center of the FEB RAS", in 2016 FASO Russia selected 300 million rubles for upgrade of six supercomputer centers of the subordinated scientific organizations. Nodes with the specialized switch cost 20 million rubles. It became possible thanks to a target subsidy from the Federal Agency for Scientific Organizations.

In the spring of 2017 delivery and basic check of the equipment were executed, and at the beginning of summer the main part of works on installation, debugging system (operating systems, the scheduler, libraries, etc.) and the application software is carried out.

The computer system was tested by Alexey Sorokin's team together with the staff of Institute of water problems of RAS (Moscow), Institute of sea geology and geophysics of the FEB RAS (Yuzhno-Sakhalinsk) and Institute of materials science of HNTs FEB RAS (Khabarovsk). On real tasks the efficiency of architecture was estimated, optimal operation modes of a new cluster were fulfilled.

Using the supercomputer it was also succeeded to carry out the first research calculations. So, scientists from IVP RAS keep the projects connected with the supershort-term forecast of the movement of fields of rainfall in which the library of machine learning tensorflow developed in Google is used. On Sakhalin using author's algorithms calculations of distribution of waves of a tsunami from seismic sources on a global scale are conducted. Scientists were satisfied with the speed of calculations.

Besides, for the purpose of increase in number of the directions of the researches conducted based on a computing cluster, the data center FEB RAS signed agreements on cooperation with IBM and Nvidia according to standards of consortium OpenPOWER. Together with giants of the industry the center is going "minimize barriers to scientists when using hybrid computing systems and such technologies as machine learning, artificial intelligence and many other things".

The data center FEB RAS already began to accept requests for providing access to system resources. Preferences are given to those projects which will receive a real gain in performance in comparison with clusters based on X86 of architecture without graphic accelerators. Those who submit applications for carrying out a research by means of supercomputer resources will be able to use both ready mathematical libraries, and the author's.

You See Also

Supercomputer

Supercomputer: projects

Supercomputers

Supercomputers: market of the CIS

Supercomputers: operating systems

Supercomputers: world market

Notes