Pairwise Biological Sequence Comparison is an important Bioinformatics application, which is executed many times daily all over the world. In order to accelerate Biological Sequence Comparison applications, GPUs have been used for more than a decade, with very good results. In this talk, we present CUDAlign, a fine-grained multi-GPU parallel strategy to compare huge DNA sequences in hundreds of GPUs. CUDAlign uses parallelogram-shaped parallelism, overlapping of computation and communication, an innovative speculation technique and has pruning capabilities. We show that CUDAlign is able to attain the best performance in the literature for comparison tools with GPUs. We also discuss CUDAlign's energy consumption. We present ongoing work on pruning with multiple GPUS and show some results in an IBM Power9 + NVidia Volta platform. Finally, we discuss open challenges and research directions.
Continue the conversation in Slack