Energy-Efficient GPU-Accelerated Solvers for High-Dimensional Simulations

Abstract

Modern scientific computing heavily relies on high-dimensional simulations, which scale to large computational and energy requirements especially when implemented on large-scale hardware platforms. Although computational throughput has increased significantly due to the use of the GPU acceleration approach, the power consumption is a major issue that needs to be addressed to achieve sustainable high-performance computing. This article introduces a solver framework that is energy efficient and uses a GPU to solve simulation problems that are computationally and power intensive in high-dimensional simulations. The approach offered combines parallelization strategies through energy-efficient techniques, such as optimizing access to memory, kernel fusion, and mixed-precision computing, to maximize performance without consuming more energy. It comes up with a mathematical representation of the solver and includes an energy model that assesses performance on the basis of execution time and power usage. The architecture is executed with CUDA-based architectures and tested with high-dimensional simulation problems on benchmarks. The effectiveness of the proposed optimizations is demonstrated by experimental results showing the proposed technology to be significantly faster and more efficient in performance per watt than conventional solutions based on GPUs. The results indicate that by incorporating energy-conscious design concepts into the GPU-based solvers, sustainability and scalability of the high-dimensional scientific simulations can be significantly enhanced.