Indirect Illumination in the Clouds
Games are an interactive form of media and thus, very sensitive to latency. Cloud based services have been successfully used for file storage, computing, web hosting and other tasks, but in gaming things are not looking too good so far. One of the most well-known cloud gaming services, Onlive went through bankruptcy and reacquisition, while its quality left a lot to be desired, with bad image quality and lag. Still, many companies see a future in real-time cloud 3D rendering, including Nvidia and AMD. Recently, Microsoft has advertised their Azure cloud servers and their availability for Xbox One developers as a way to improve game quality. How much is true in their promises of better cloud assisted graphics rendering on X1 remains to be seen.
Latest demonstrations on SIGGRAPH 2013 from Nvidia graphics team (paper, slides)is an interesting approach to cloud graphics. The offering for tablets, like iPad, is just a standard full-frame server streaming from Nvidia GRID. When it comes to more powerful platforms with dedicated GPUs, like laptops or desktops, a mixed approach is shown.
Indirect Illumination is a computationally demanding task and thus is not used in real time in most games. In cases of static lightning or limited dynamic one, game developers computed global illumination in advance and baked it into textures (Mirror's Edge is a very good example). The Nvidia team uses a similar approach for laptop GPUs, with the remote server calculating global illumination maps and streaming them. For more powerful desktop GPUs, servers can do a shorter pipeline and send less processed results, allowing the desktop to do the remaining map calculation locally with higher quality.
The benefit of this mixed approach seems to be higher graphics quality without an awful effect on latency. Everything else is rendered locally with minimal latency, only the more subtle global illumination is a bit behind. The team has shown some latency perception tests, where global illumination lags behind the rest of the image. Even the 200 ms latency in advanced illumination seems quite decent, while everything else reacts quickly.
Technologically the idea of remotely-calculated dynamic global illumination in games seems sound, but the economic question remains. This technique is only worth it when the server is more powerful than the local machine. On PC it may not be worth paying for a service that supports only a few games, if the GPU upgrade has similar cost. Another possibility is the appearance of home gaming servers which will stream locally to mobile platforms and assist weaker laptops and PCs. On consoles with fixed hardware – cloud graphics DLC subscription may become a thing.