The Community for Technology Leaders
RSS Icon
pp: 1
Haohuan Fu , Tsinghua University, Beijing
Lin Gan , Tsinghua University, Beijing
Robert Clapp , Stanford University
Huabin Ruan , Tsinghua University, Beijing
Oliver Pell , Maxeler Technologies, London
Oskar Mencer , Maxeler Technologies, London
Michael J. Flynn , Stanford University
Xiaomeng Huang , Tsinghua University, Beijing
Guangwen Yang , Tsinghua University, Beijing
Seismic migrations dominate around 90% of the computation cycles in oil and gas industry. With the demand to handle high-density data and more complicated physics models, migration applications always call for more computing power, and adopt new architectures quickly. The current multi-core and many-core architectures have significantly improved the density of computational resources within a chip, but have also made memory bandwidth a bottleneck that stops the scaling of the performance over the increased number of cores. This article presents our reverse time migration design based on reconfigurable data-flow engines. Combining both algorithmic and architectural optimizations, we manage to achieve a balanced utilization of various resources (computational logics, local buffers, memory bandwidth, etc.) in the system, with none of them becoming the performance bottleneck. Our data-flow design provides an equivalent performance to 72 Intel CPU cores, and achieves 10 times higher power efficiency than the multi-core CPU architecture.
Haohuan Fu, Lin Gan, Robert Clapp, Huabin Ruan, Oliver Pell, Oskar Mencer, Michael J. Flynn, Xiaomeng Huang, Guangwen Yang, "Scaling the Reverse Time Migration Performance Through Reconfigurable Data-Flow Engines", IEEE Micro, , no. 1, pp. 1, PrePrints PrePrints, doi:10.1109/MM.2013.111
19 ms
(Ver 2.0)

Marketing Automation Platform Marketing Automation Tool