View in EDS

Exploring Efficient FPGA Acceleration of High-Order 3D Iterative Stencil Loops on Large Data Grids.

Saved in:
Bibliographic Details
Title: Exploring Efficient FPGA Acceleration of High-Order 3D Iterative Stencil Loops on Large Data Grids.
Authors: Ibrahim, Alhussain1 (AUTHOR) aibrahim@palmlab.com, Elrabaa, Muhammad E. S.2 (AUTHOR) elrabaa@kfupm.edu.sa, Alsaleh, Saleh2 (AUTHOR) salehs@kfupm.edu.sa, El-Maleh, Aiman H.2 (AUTHOR) aimane@kfupm.edu.sa, Tonellot, Thierry3 (AUTHOR) thierrylaurent.tonellot@aramco.com
Source: Arabian Journal for Science & Engineering (Springer Science & Business Media B.V. ). Jan2026, Vol. 51 Issue 1, p205-222. 18p.
Subjects: Field programmable gate arrays, Iterative methods (Mathematics), Resource allocation, Tiling (Mathematics), Time
Abstract: An efficient methodology for FPGA acceleration of high-order 3D iterative stencil loops over large 3D-grids was developed. Spatial (tiling) and temporal (combined iterations) blocking are used to circumvent the FPGAs' limitations and maximize throughput. Implemented as a fully asynchronous SW-HW pipeline, it can compute high-order stencils on 3D grid without any limitations on the grid size or the number of iterations. An 8th-order, 25-point 3D stencil was used to demonstrate the methodology and possible optimizations of performance, resource utilization, and power efficiency. Results show that throughput is only limited by the FPGA off-chip memory bandwidth. Comparisons with published results for the same stencil showed that the developed methodology can achieve a throughput equivalent to ~ 43-Haswell cores (22 nm technology) or ~ 13 Milan-x cores (7 nm technology) running at 2.3 and 2.45 GHz, respectively. Compared to an A100 NVIDIA GPU implementation of the same stencil, it achieved ~ 41% better power efficiency (Watts per GB/s). The methodology was extended to support multiple FPGAs. With two FPGA boards, the total latency was reduced by ~ 27%. [ABSTRACT FROM AUTHOR]
Copyright of Arabian Journal for Science & Engineering (Springer Science & Business Media B.V. ) is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:An efficient methodology for FPGA acceleration of high-order 3D iterative stencil loops over large 3D-grids was developed. Spatial (tiling) and temporal (combined iterations) blocking are used to circumvent the FPGAs' limitations and maximize throughput. Implemented as a fully asynchronous SW-HW pipeline, it can compute high-order stencils on 3D grid without any limitations on the grid size or the number of iterations. An 8th-order, 25-point 3D stencil was used to demonstrate the methodology and possible optimizations of performance, resource utilization, and power efficiency. Results show that throughput is only limited by the FPGA off-chip memory bandwidth. Comparisons with published results for the same stencil showed that the developed methodology can achieve a throughput equivalent to ~ 43-Haswell cores (22 nm technology) or ~ 13 Milan-x cores (7 nm technology) running at 2.3 and 2.45 GHz, respectively. Compared to an A100 NVIDIA GPU implementation of the same stencil, it achieved ~ 41% better power efficiency (Watts per GB/s). The methodology was extended to support multiple FPGAs. With two FPGA boards, the total latency was reduced by ~ 27%. [ABSTRACT FROM AUTHOR]
ISSN:2193567X
DOI:10.1007/s13369-025-10919-y