Exploring Graphics Processing Unit (GPU) Resource Sharing Efficiency for High Performance Computing

Abstract

The increasing incorporation of Graphics Processing Units (GPUs) as accelerators has been one of the forefront High Performance Computing (HPC) trends and provides unprecedented performance; however, the prevalent adoption of the Single-Program Multiple-Data (SPMD) programming model brings with it challenges of resource underutilization. In other words, under SPMD, every CPU needs GPU capability available to it. However, since CPUs generally outnumber GPUs, the asymmetric resource distribution gives rise to overall computing resource underutilization. In this paper, we propose to efficiently share the GPU under SPMD and formally define a series of GPU sharing scenarios. We provide performance-modeling analysis for each sharing scenario with accurate experimentation validation. With the modeling basis, we further conduct experimental studies to explore potential GPU sharing efficiency improvements from multiple perspectives. Both further theoretical and experimental GPU sharing performance analysis and results are presented. Our results not only demonstrate the significant performance gain for SPMD programs with the proposed efficient GPU sharing, but also the further improved sharing efficiency with the optimization techniques based on our accurate modeling.

Multilevel Performance Evaluation of Nvidia Grid VCA Using Iray and V-Ray Rendering Engines in 3DS Max Design

“…The programmable parallel processors in GPUs exceed the computing power of multicore CPUs [15,16]. GPUs are being utilized at an increasing rate in scientific computing applications and GPU rendering algorithms have emerged [17,18].…”

A complete and efficient CUDA-sharing solution for HPC clusters

“…The NVIDIAproprietary device driver running in the GPU servers will manage the concurrent execution of the different active contexts using its own scheduler, in the same way as it does in a local GPU context. Our multiple servers sharing local GPUs could benefit from techniques designed to improve this usage mode, such as some of those employed in [36,37,38]. Furthermore, the recent Kepler GPU architecture [39] features enhanced multitask support with the Hyper-Q technology, which directly benefits our virtualization technology without the need of any further development or customization.…”

GPU Resource Sharing and Virtualization on High Performance Computing Systems

“…The approach followed by S GPU is complementary to [1] and our approach here, and may be combined with our proposed approach by simultaneously executing kernels from multiple processes for efficient GPU sharing. Furthermore, the efficient GPU sharing approach proposed by [31,32,33] is relavant to this work as well.…”