Paper Info
Title
Acceleration of Generalized Minimum Aberration Designs of Hadamard Matrices on Graphics Processing Units
Abstract
The process of applying generalized minimum aberration criteria (GMAC) to non-regular fractional factorial designs is extremely computationally intensive. Constructing and ranking all designs can take hours if not days; therefore, exploitation of the massively parallel nature of modern graphics processing units (GPUs) are used to perform the task. The computation is not just ported to the GPU, but is implemented as to optimize performance based upon the modern GPU architecture. Optimizations include using bit operations and table lookups to reduce the number of addition and multiplication operations performed. Tables are housed in GPU constant memory with almost no latency for access. Using a statistical proof from previous research reduces the memory required for the calculation. Optimizations regarding memory storage and transfer in NVIDIA's Compute Unified Device Architecture (CUDA) are also explored, as well as advance features such as streams and multiple GPUs. Experimental results have demonstrated the effectiveness of the proposed approach.
Year
DOI
Venue
2012
10.1109/HPCC.2012.191
HPCC-ICESS
Keywords
Field
DocType
multiple gpus,generalized minimum aberration designs,modern gpu architecture,fractional factorial design,compute unified device architecture,modern graphics,hadamard matrices,memory storage,bit operation,graphics processing units,generalized minimum aberration criterion,gpu constant memory,parallel algorithms,sorting,statistical analysis,indexes,error correction,kernel,statistical proof,memory management
Graphics,Massively parallel,Computer science,CUDA,Parallel algorithm,Parallel computing,Multiplication,Memory management,Graphics processing unit,CUDA Pinned memory
Conference
ISSN
Citations 
PageRank 
2576-3504
0
0.34
References 
Authors
2
4
Name
Order
Citations
PageRank
Jon Calhoun1474.75
Josh Graham200.34
Hong Zhou300.34
Hai Jiang462.56