Reconfigurable FPGA-based Clusters: Next step in Supercomputing

100 %
0 %
Information about Reconfigurable FPGA-based Clusters: Next step in Supercomputing

Published on March 19, 2009

Author: vivekv80



Presented at Graduate Student Symposium 2007.

Reconfigurable FPGA-based Clusters: Next step in Supercomputing Vivek Venugopal, Kevin Shinpaugh HPReC systems FPGA Interconnect Introduction • Partially Reconfigurable System processor/co-processor network nodes Bus/Switch 00 01 02 03 • Current High Performance Computing (HPC) Inside each node FPGA • Cluster of FPGAs equivalent to a huge processor with embedded 10 11 12 13 applications include genome sequencing (BLAST), reconfigurability, replication and parallelism molecular dynamics simulation (AMBER, NAMD), RAM 20 21 22 23 • Issues prevalent with systems: astrophysics simulation, weather prediction, etc. 30 31 32 33 • Scalability of the system with respect to type of application GPP Motivation Interconnection Network • Availability of a fast interconnection network for I/O bound • Completely Reconfigurable applications (Bandwidth access) • HPC systems cater to two types of applications: Bus/Switch System • More processors or more floating point cores for compute Inside each HPC applications Infiniband interconnect node I/F FPGA bound applications Memory RAM User Logic >100 million gates Compute I/O bound n bound FPGA Power PC application application Interconnection Network • HPC systems need to be built according to the Test platforms (HPReC) CPU AMD application for maximum efficiency. Opteron Current Future Cray XD1 SGI RASC 3.2 GB/s platform platform HPC scenario HPReC scenario Processing AMD Opteron + Intel Itanium + hardware Xilinx Virtex-4 FPGAs Xilinx Virtex-II FPGAs Cache Memory FPGA RapidArray Interconnect 16 MB Xilinx Application Interconnect Application Interface chip Rapid Array Interconnect Numalink Interconnect QDR SDRAM Virtex 4 3.2 GB/s 12.8 GB/s network Bandwidth 4 GB/s 12.8 GB/s access 2 x 2 GB/s • Most of the hardware mapping from the software is automated and is based on the availability of specified libraries or processors for implementation, eg. GPP with fixed flexible hardware for RapidArray Interconnect Bus logic Mitrionics, Handel-C, etc. computation and blocks communication Performance and speedup Better performance and HPReC applications scales with more processors speedup with FPGA Reconfigurable systems Select map CPU Loader prog. interface Intel FPGA • A combination of I/O bound and compute Itanium2 • Reconfigurable computing is based on the concept bound applications PCI that the application defines the processor. 66 MHz • Bioinformatics 2 x 3.2 GB/s •FPGAs are inherently parallel with lower power •Smith Waterman algorithm Cache Memory Algorithm dissipation and are available with a huge library of 16 MB TIO FPGA •BLAST QDR SDRAM Xilinx Virtex II 9.6 GB/s application cores. • Physics • Reconfiguration can result in (i) efficient hardware 4 x 3.2 GB/s • Coliter data utilization for repeated operations in a specific • Molecular Simulation Dynamics application and (ii) better data passing on the Numalink 4 Interconnect Bus •AMBER interconnection network between the processors. References [1] “Cray XD1 datasheet,” Cray Inc., Technical report, 2005.. Available: [2] “Cray XD1 supercomputer for reconfigurable computing,” Cray Inc., Technical report, 2005. Available: [3] “ SGI Reconfigurable Application Specific Computing: Accelerating Production Workflows,” Silicon Graphics Inc., Technical report, December 2006. Available: [4] “ Extraordinary Acceleration of Workflows with Reconfigurable Application-specific Computing from SGI,” Silicon Graphics Inc., Technical report, November 2004. Available:

Add a comment

Related presentations

Presentación que realice en el Evento Nacional de Gobierno Abierto, realizado los ...

In this presentation we will describe our experience developing with a highly dyna...

Presentation to the LITA Forum 7th November 2014 Albuquerque, NM

Un recorrido por los cambios que nos generará el wearabletech en el futuro

Um paralelo entre as novidades & mercado em Wearable Computing e Tecnologias Assis...

Microsoft finally joins the smartwatch and fitness tracker game by introducing the...

Related pages

Reconfigurable supercomputing - Is high-performance ...

Clusters are the current ... reconfigurable computing the next supercomputing ... suggest that FPGA-based acceleration is a promising ...
Read more

A Scalable FPGA-based Multiprocessor -

A Scalable FPGA-based ... The next step in the design flow was to ... OpenFPGA - Defining Reconfigurable Supercomputing. http ...
Read more

FPGA Supercomputing Platfroms, Architecture, And ...

EURASIP Journal on Embedded Systems FPGA Supercomputing Platforms, Architectures, and Techniques for Accelerating Computationally Complex Algorithms Guest ...
Read more

EURASIP Journal on Embedded Systems -

EURASIP Journal on Embedded Systems -
Read more

Content-based image retrieval algorithm acceleration in a ...

Content-based image retrieval algorithm acceleration in ... ory with a cluster of FPGA-based reconfigurable ... of FPGA-based clusters ...
Read more

Booklet Proof Reading | DATE 2014

... SYSTEM DESIGN CHALLENGES FOR NEXT ... We want to demonstrate an FPGA-based probability-aware fault ... of 19.4 fJ/conversion-step without ...
Read more

Publications - Cognitronics and Sensor Systems

FPGA-based circular hough transform with graph ... Server Architecture for Next-Generation Compute Clusters. In: ... with Supercomputing ...
Read more

FPGA Based Project 1 -

FPGA Based Project 1 - Free ... David Andrews, Richard S. Miller, and Daniel Stanzione, Jr. Reconfigurable Computing Systems Lab† University of ...
Read more