Ilya I. Levin

Reconfigurable Computer Systems Based on Virtex-6 and Virtex-7 FPGAs

Abstract In the paper we consider computational modules Saiph and Rigel designed on the base of Xilinx Virtex-6 FPGAs and a computational module 24V7-750 based on Xilinx Virtex-7 FPGAs which is now in the design phase. The computational modules are intended for creation of reconfigurable computer systems (RCS) of various configurations. We compare technical characteristics of the considered CMs and describe a software suit for RCS programming, based on a high-level programming language. Use of this language provides simplicity of programming and helps to reduce the time of development of applications for tasks from various problem domains.

High-Performance Reconfigurable Computer Systems Based on Virtex FPGAs

The paper covers architectures and comparison characteristics of reconfigurable computer systems RCS based on field programmable gate arrays FPGAs of the Xilinx Virtex-7 family and technologies of task solving by means of software development tools. In the paper we also consider architecture and assembly of next-generation RCS with a liquid cooling system and give results of calculations and prototyping of principal technical solutions which provide the performance of 1 PFlops for a standard computational 47U rack with the power of 150 kWatt. This is promising approach because of RCS with a liquid cooling system have a considerable advantage for lot of engineering and economical parameters such as real and specific performance, power efficiency, mass and dimension characteristics, etc., in comparison with similar systems.

High Performance Reconfigurable Computer Systems on the base of FPGA technology

Abstract The paper covers the problems of design and creation of high performance reconfigurable computer systems based on FPGA technology. Authors compare performance values of up-to-date cluster systems and original reconfigurable computer systems, suggest principles of design of modular scalable reconfigurable computer systems with high real performance and give examples of various implementations of such systems. In addition, principles of applications development are suggested. Also the paper covers the problems of reconfigurable system programming. There is viewed an original complex of system software for reconfigurable system programming.

High-Performance Reconfigurable Computer Systems with Immersion Cooling

In the paper, we review the design principles and architecture of reconfigurable computer systems with immersion cooling. We prove that systems with immersion cooling are the most promising for the design of high-performance computer complexes. We give selection criteria and design results for the principal components of the immersion cooling system. We demonstrate the design of our computational module prototype, based on advanced Xilinx UltraScale FPGAs and give testing results for the principal technical solutions. We prove that the designed immersion cooling system has a high power efficiency and power reserve for designing advanced reconfigurable computer systems on the basis of new UltraScale+ FPGAs and other next-generation FPGAs. We suggest new design solutions for the case of our computational module, as well as for the layout of the main computational board and other components of the computational module for use of Xilinx UltraScale+ FPGAs.

Design of Advanced Reconfigurable Computer Systems with Liquid Cooling

The paper covers problems of design of reconfigurable computer systems with a liquid cooling system. Design principles of liquid cooling systems of open and closed types are specified, and their comparative analysis is presented. It is shown that open type liquid cooling systems are the most acceptable for high-performance computer systems design. Architecture features of various immersion liquid cooling systems are analyzed; selection criteria of the main cooling system components are given. The paper presents the results of modelling, prototyping and experimental verification of the main technical solutions for our energy-efficient computational module with liquid cooling. The design of the computational module, designed on the base of the modern FPGAs of Xilinx UltraScale series, is presented. It is shown that the developed solutions have power reserve for design of advanced computer systems, based on the new UltraScale+ FPGA family. For the UltraScale+ FPGAs it is necessary to perform some modifications, concerning both the layout of the main computational circuit board and the design of the computer unit and its cooling system. The upgraded design of our advanced computer unit with liquid cooling is presented.

Realization of problem of liquid filtering in porous medium on reconfigurable computer system

The paper covers solution of the problem of liquid filtering in porous medium on a reconfigurable computer system (RCS). To solve this problem on the RCS we transformed it into a structural-procedural form, and optimized the task structure to organize parallel calculations. The suggested realization of the filtering problem allows effective scaling of the solution if available hardware resource of the RCS is growing. Owing to use of the RCS it is possible to reduce the time and to increase precision of simulation of various processes of the Earth interior. This is an urgent problem of oil industry for geophysical exploration and well survey.

Solution of the problem of biological rehabilitation of shallow waters on multiprocessor computer system

The paper covers problems of solving tasks of water ecology on multiprocessor computer system (MCS). We proposed a new model of biological rehabilitation of shallow waters in view the factors that have a significant influence on the water quality. Its discretization was performed with using the balance method and the implicit scheme with central differences. The proposed numerical method for the solution of the model problem is most common and is suitable to the study of hydrobiological processes occurring in shallow waters. Since it allows to correctly design computational algorithms on the boundary between the integration domain and environments. One of the objectives of the work is reducing the calculation time and saving the accuracy of the results of solving problem of biological rehabilitation of shallow waters by using a multiprocessor computer system. Two algorithms have been developed in the implementation of the parallel algorithm for solving problem on the MCS for the distribution of data between the processors. There is the algorithm on the basis of the k-means method, based on the minimization of the functional of the total sample variance of scatter elements about the center of gravity of the subdomains, which allows increasing the efficiency of the parallel algorithm of the hydrobiology problem of shallow water. Using the MCS can significantly reduce the calculation time while saving the accuracy of the solution. The latter fact provides the fast and qualitative interpretation of hydrobiological data.

High-performance reconfigurable computer systems

The FPGA-based reconfigurable computer systems have high real performance and provide practically linear growth of performance when hardware system resource is growing. The paper deals with design features, technical characteristics and values of real performance of computational modules of reconfigurable computer systems, designed on the base of Virtex-6 FPGAs. In addition, a software suit, intended for development of parallel applications for the RCS, is considered.