Atlas shrugged : device-agnostic radiance megatextures

Abstract

This paper proposes a novel distributed rendering pipeline for highly responsive high-fidelity graphics based on the concept of device-agnostic radiance megatextures (DARM), a network-based out-of-core algorithm that circumvents VRAM limitations without sacrificing texture variety. After an automatic precomputation stage generates the sparse virtual texture layout for rigid bodies in the scene, the server end of the pipeline populates and updates surface radiance in the texture. On demand, connected clients receive geometry and texture information selectively, completing the pipeline by asynchronously reconstituting these data into a frame using GPUs with minimal functionality. A client-side caching system makes DARM robust to network fluctuations. Furthermore, users can immediately start consuming the service without the need for lengthy downloads or installation processes. DARM was evaluated on its effectiveness as a vehicle for bringing hardware-accelerated ray tracing to various device classes, including smartphones and single board computers. Results show that DARM is effective at allowing these devices to visualise high quality ray traced output at high frame rates and low response times.