UtilizingMachineLearningTechniquesfortheAssessmentof ConcurrentKernelExecutionPerformanceonGPUs

Abstract

The increasing demand for high-performance computing has positioned Graphics Processing Units (GPUs) as critical components in various computational fields, including scientific simulations, machine learning, and graphics rendering. This thesis explores the utilization of machine learning techniques to assess and optimize the performance of concurrent kernel execution on GPUs. Concurrent kernel execution allows multiple kernels to run simultaneously on a GPU, improving resource utilization and overall throughput. However, this approach introduces challenges related to resource contention and performance variability. By leveraging machine learning models, we can predict performance metrics, identify bottlenecks, and optimize scheduling strategies. This research presents a comprehensive framework that integrates predictive modeling, classification algorithms, and deep learning techniques to enhance the assessment of concurrent kernel execution performance. Through extensive experimentation and analysis, we demonstrate the effectiveness of our approach in improving GPU performance.