Simulation of bevel gear cutting with GPGPUs—performance and productivity

Abstract

The desire for general purpose computation on graphics processing units caused the advance of new programming paradigms, e.g. OpenCL C/C++, CUDA C or the PGI Accelerator Model. In this paper, we apply these programming approaches to the software KegelSpan for simulating bevel gear cutting. This engineering application simulates an important manufacturing process in the automotive industry. The results obtained are compared to an OpenMP implementation on various hardware configurations. The discussion covers performance results, but also productivity of code development realized in this effort.

Accelerators for Technical Computing: Is It Worth the Pain? A TCO Perspective

“…The emergence of accelerators in high-performance technical computing increases the TCO complexity and makes a fair comparison to nodes with commodity processors challenging. In previous works [9,10], we have seen the importance of development productivity especially on (GPU) accelerators, also dependent on the used programming model and the kind of application. CAPS' case study pamphlet [11] gives a short overview of the economics of GPU code migration and draws the conclusion that GPUs are worthwhile when gathering at least a two-fold speedup.…”

“…25 % of the serial runtime. Since its industry costumers have had GPU hardware at their disposal anyway, we still started accelerating this kernel [9] (∼ 150 lines in serial code) using the portable OpenCL. However, for our TCO calculations, we will assume that this module accounts for 90 % of the whole application runtime to illustrate our statements and not be restricted by Amdahl's law.…”

“…In previous works [9,10], we investigated the impact of development effort in accelerating code regions and expressed it by the number of lines of kernel code added or modified. Now, we try to quantify this effort in man-days (see Tab.…”

An investigation of the performance portability of OpenCL

“…Other work has investigated the use of OpenCL on different platforms as a means of assessing power usage [5] and productivity [6].…”

A Performance Study of a Dual Xeon-Phi Cluster for the Forward Modelling of Gravitational Fields

“…Each new technology carries with it different programming paradigms. Nevertheless, the effort to migrate scientific applications to CUDA (for GPUs) or OpenCL (i.e., programming that requires programming low level kernels) is often much higher as compared to directivebased programming like OpenMP (for CPU or MIC) [2]. Prior experiments which test the functionality of the Xeon-Phi coprocessor show that migrating scientific software is relatively easy, thereby making the MICs a promising tool for HPC applications [3].…”