Q-VR: System-Level Design for Future Collaborative Virtual Reality Rendering

Since the release of the movie Ready Player One, consumers have been longing for a commercial VR product which could provide a truly immersive experience without mobility restriction and periodical motion anomalies. In other words, users require exceptional visual quality from an untethered mobilerendered head-mounted displays (HMDs) that is equivalent to what high-end tethered VR systems (e.g., Oculus Rift [15] and HTC Vive [7]) provide. Although the current mobile’s processing capability has been significantly improved [1,16], they still cannot fully process heavy VR workloads under the stringent runtime latency constraints. With the development of high performance server technology, server-based realtime rendering of Computer Graphics has been introduced by the recent architecture studies [8, 10, 23, 24] and major cloud vendors [5, 14]. However, under the current network conditions, remote servers alone cannot provide realtime low-latency high-quality VR due to the dominating communication latency. Fig.1 shows the breakdown of the end-to-end latency (i.e., from tracking to display) for executing several high-quality VR applications under two commercial mobile VR designs: local-only rendering and remote-only rendering. The blue lines represent the frame rate (FPS) achieved on the VR HMD while the red dash lines illustrate VR system latency restriction (i.e., commercial standard of 25ms). The figure shows that the performance on the integrated GPU is the key bottleneck for local-only rendering, while the transmission latency in remote-only rendering contributes to approximately 63% of the overall system latency. Thus, neither local-only nor remote-only rendering can satisfy the latency requirements for high-quality mobile VR: there is a clear mismatch between hardware’s raw computing power and desired rendering complexity.