It's to cloud gaming latency what a warp drive is to the cosmic speed limit — a workaround
If you aren’t sold on cloud gaming, we don’t blame you. We aren’t either. Despite all its inherent promise, cloud gaming continues to be held back by the high latency generally associated with wide area networks — not to mention the ridiculously fast internet speeds and plenty of spare bandwidth that are the other desiderata for a quality cloud gaming experience. But what if you wanted to enable a better cloud gaming experience but didn’t know of a way to directly lower latency? How about masking it?
That is precisely what Microsoft is trying to do with its “DeLorean” technology. On the face of it, the idea is simple. In a normal cloud gaming setup, the game visuals are rendered on a remote server and streamed to a thin client, with the rendering and streaming being initiated based on user input. What DeLorean does differently is to predict user input and pre-render a set of future outcomes based on those predictions, “delivering them to the client one entire RTT [round trip time] ahead of time; clients perceive no latency.” According to a research paper published by the software giant, the technique, which has already been tested with Doom 3 and Fable 3, can be used to mask network latency of up to 250ms. (Generally, anything over 60ms is considered less than ideal and anything over 100ms downright unplayable.)
Besides predicting future outcomes, DeLorean also relies on state space subsampling and time shifting (to cut down on number of possible outcomes), misprediction compensation, and video compression. You would expect multiplayer to be particularly problematic for a system like this given the number of players, but Microsoft insists things are about just as bad as standard multiplayer.
There’s a major downside to this technique, though: the bandwidth consumption can be 1.5 to 4.5 times more than your run-of-the-mill cloud gaming setup. However, the researchers “believe this is a reasonable trade-off for service providers who are otherwise unable to offer users low-latency interactivity.”