Is GPUs Ready to Boost Genomic Alignment Computation

This paper presents a comparative study of CPU and GPU performance in sequence alignments for the Next Generation Sequencing (NGS), specifically in the context of short reads mapping to a reference genome. Bowtie2 and BWA which runs on CPU only and its equivalent, NvBowtie and BarraCUDA which runs on both CPU and GPU, is chosen as the genomic tools benchmark for this studies. The CPUs selected for our study is Intel Xeon Processor E5-2695v2(x2) and Intel Xeon Processor E5-2699 v4(x2). The GPUs that we have selected for our study are the NVIDIA Tesla K80, Pascal 100 (P100) and Volta 100 (V100). Our results show that GPUs performs better than CPUs for long read's length and large number of reads. However, taking the price/performance ratio into account, the results suggests a case of diminishing return for the newer generation of NVIDIA GPU.