Multi-user predictive rendering on remote multi-GPU clusters

Note: We don't have the ability to review paper

PubDate: February 2019

Teams: Université de Reims Champagne-Ardenne;PSA Peugeot Citroën;PSL-Research University

Writers: J. Randrianandrasana; A. Chanonier; H. Deleau; T. Muller; P. Porral; M. Krajecki; L. Lucas

PDF: Multi-user predictive rendering on remote multi-GPU clusters


Many stages of the industry workflow have been benefiting from CAD software applications and real-time computer graphics for decades allowing manufacturers to perform team project reviews and assessments while decreasing the need for expensive physical mockups. However, when it comes to the perceived quality of the final product, more sophisticated physically based engines are often preferred though involving huge computation times. In this context, our work aims at reducing this gap by providing a predictive rendering solution leveraging the computing resources offered by modern multi-GPU supercomputers. To that end, we propose a simple static load balancing approach leveraging the stochastic nature of Monte Carlo rendering. Our solution efficiently exploits the available computing resources and addresses the industry collaboration needs by providing a real-time multi-user web access to the virtual mockup.