NMF-mGPU: non-negative matrix factorization on multi-GPU systems

• DOI: 10.1186/s12859-015-0485-4
• Publication Date: 2015-02-12T04:57:12Z
• Authors: Edgardo Mejía-Roa; Daniel Tabas-Madrid; Javier Setoain; Carlos García; Francisco Tirado; Alberto Pascual-M...
• Keywords: Non-negative Matrix Factorization; Coprocessor; Speedup; Linear algebra; Graphics processing unit; Implementation

In the last few years, Non-negative Matrix Factorization ( NMF ) technique has gained a great interest among Bioinformatics community, since it is able to extract interpretable parts from high-dimensional datasets. However, computing time required process large data matrices may become impractical, even for parallel application running on multiprocessors cluster. this paper, we present NMF-mGPU, an efficient and easy-to-use implementation of algorithm that takes advantage high performance delivered by Graphics-Processing Units GPUs ). Driven ever-growing demands video-games industry, graphics cards usually provided in PCs laptops have evolved simple graphics-drawing platforms into high-performance programmable systems can be used as coprocessors linear-algebra operations. these devices limited amount on-board memory, which not considered other implementations GPU.NMF-mGPU based CUDA Compute Unified Device Architecture ), NVIDIA's framework GPU computing. On with low memory available, input are blockwise transferred system's main GPU's processed accordingly. addition, NMF-mGPU been explicitly optimized different architectures. Finally, multiple synchronized through MPI Message Passing Interface four-GPU system, about 120 times faster than single conventional processor, more four device (i.e., super-linear speedup).Applications getting attention due their outstanding when compared traditional processors. relatively price represents highly cost-effective alternative clusters. life sciences, results excellent opportunity facilitate daily work bioinformaticians trying biological meaning out hundreds gigabytes experimental information. "out box" researchers little or no expertise programming variety platforms, such PCs, laptops, high-end freely available at https://github.com/bioinfo-cnb/bionmf-gpu .