This dissertation conducted an in-depth analysis of graphics processing unit (GPU) performance models and configurations. It profiled GPU systems across a range of hardware configurations and workloads, employing machine learning techniques such as support vector machines (SVM) and Random Forest algorithms to develop predictive models and identify key system parameters impacting performance. This work also presents the Genomics-GPU benchmark suite, comprising ten critical algorithms representative of genomics analysis tasks. It performed extensive evaluations using both hardware and software simulators, accompanied by microarchitectural characterizations examining memory access patterns, thread scaling behavior, pipeline stalls, and interconnect network impacts. It proposes a novel optimization: a lightweight process-in-memory (PIM) data pre-processing unit (DPU) integrated within the GPU’s memory hierarchy. This architectural enhancement offloads preliminary filtering operations to memory units, thereby reducing data movement between memory and processing units—a significant bottleneck in handling large genomic datasets. The specialized data pre-processing unit (DPU) performs sequence alignment pre-processing directly within DRAM, improving data throughput and overall performance by 1.23X-8.37X.