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

Abstract

High performance computing is increasingly common in technological industries and there are many different solutions available on the market. Determining which computing solution is most cost-effective can be difficult. This study outlines the performance between a single-user, traditional high-performance workstation and a multiuser , virtualized workstation. Along with this direct performance comparison, the impacts of virtualization on rendering performance, GPUs, and the technological industry is evaluated in this study. Through the repeated rendering of two different Computer-Aided Design (CAD) models under varying test scenarios, a pool of data including render times and image quality is collected and analyzed. Two phenomena are observed and explained. One is a diminishing return in GPU power output that is observed after allocating four or more GPUs to a single rendering task. The second is a noticeable point of image-noise convergence during a render that could potentially be calculated and exploited to make rendering more time-efficient. These discoveries may impact the effectiveness of virtual GPU scalability and make time-consuming rendering more efficient for industry users. The NVIDIA GRID Visual Computing Appliance (VCA) is found to be cost effective for research laboratories that have several users with diverse needs.