T-Platforms: A-Class

A-Class is the supercomputer solution of T-Platforms company.

On June 24, 2014 the company announced T-Platforms creation of the integrated supercomputer solution A-Slass focused on calculations of multipetaFLOPS range.

A-Class system fragment

Class A - the system of level of a rack with the highest computing density, scalability, energy efficiency and reliability. The chassis of a system integrates in a uniform computing resource of 2 head nodes, 256 computing nodes and 60 InfiniBand and Ethernet switches.

A-Class system element

Performance

Initial peak capacity of A-Class more than 420 TFLOPS on one system is also scaled up to 54 petaFLOPS by means of consolidation of 128 chassis. The first released computing module contains four specialized motherboards equipped with the Intel Xeon processor E5-2600 v2 and the NVIDIA Tesla K40 accelerator.

The 1+1 configuration allows to implement optimum NVIDIA GPU Direct technology and to provide the greatest efficiency of the solution of computing problems of a different class. Capacity of processors and accelerators in the A-Class system is balanced with available capacity of interfaces of InfiniBand at the level of 3.3 GB / with to Tflops.

Network

A system has the balanced and productive network infrastructure consisting of two Ethernet networks and two networks InfiniBand. One of independent networks FDR InfiniBand is used for MPI traffic, the second serves for data exchange between computing nodes. A-Slass allows to implement different topology of switching of computing nodes, including modern 3D - and 4D-torus, Flattened butterfly and Hypercube for achievement of effective interaction of tens of thousands of computing nodes of the supercomputer.

Cooling

The managing, computing and switching modules A-Class are cooled with hot water with inlet temperature up to 45 degrees Celsius and more than 50 degrees at the exit from modules. Electronic payments fasten directly on the special radiator skintight to components of payments for heat extraction.

The technology of direct cooling with hot water allows to reach high energy efficiency, it is essential to reduce operational noise in the computing hall and to implement the year-round mode of "free cooling" of a system, without purchase costs of compressors, refrigerators and their operation. In winter time there is a possibility of reuse of already heated water for heating of premises.

Fault tolerance

In A-Class equipment rooms and perspective software tools of increase in fault tolerance of the supercomputer are implemented. Two independent modules of management with the selected network factories Ethernet support hot swap, providing functions of failsafe management and monitoring of a system.

Independent servers of management A-Class allow to trace a status of components of a system and to manage loading and configurations of the set software. Independent input lines of power supply and "excess" power supplies hot measurements maintain continuity and reliability of operation of the supercomputer of multipetaFLOPS range.

2014

Achievement in rating the Top 500

On November 18, 2014 it became known - the A-Class system took the 22nd place in 44 editions of rating of computing systems of the planet the Top 500.

The rating is formed two times a year experts of the American State research computer center of the Ministry of Energy, the universities of Mannheim and Tennessee. Ranging of supercomputers happens according to their performance level, shown on the standard Linpack test.

Listing fragment

The upgraded system of A-Class company which project of installation for November 18, 2014 is kept in the Moscow State University rose by the 22nd line of rating with a peak capacity of 2.576 of Pflops and performance on Linpack 1.849 of Pflops.