An Image-Space Split-Rendering Approach to Accelerate Low-Powered Virtual Reality

TL;DR

This paper introduces an image-space split-rendering method for mobile VR that leverages both server and client resources to reduce latency and improve performance compared to traditional streaming and client-only rendering.

Contribution

It proposes a novel networked split-rendering approach that divides rendering tasks between server and mobile client in image space, enhancing speed and efficiency.

Findings

Achieves faster end-to-end image presentation rates.

Reduces latency compared to existing streaming solutions.

Utilizes server and client resources more effectively.

Abstract

Virtual Reality systems provide many opportunities for scientific research and consumer enjoyment; however, they are more demanding than traditional desktop applications and require a wired connection to desktops in order to enjoy maximum quality. Standalone options that are not connected to computers exist, yet they are powered by mobile GPUs, which provide limited power in comparison to desktop rendering. Alternative approaches to improve performance on mobile devices use server rendering to render frames for a client and treat the client largely as a display device. However, current streaming solutions largely suffer from high end-to-end latency due to processing and networking requirements, as well as underutilization of the client. We propose a networked split-rendering approach to achieve faster end-to-end image presentation rates on the mobile device while preserving image quality. Our proposed solution uses an image-space division of labour between the server-side GPU and the mobile client, and achieves a significantly faster runtime than client-only rendering and than using a thin-client approach, which is mostly reliant on the server.