Predictive Modeling To Increase Responsivity of Streamed Games 120
jones_supa (887896) writes Streaming game services always bump up against a hard latency limit based on the total round-trip time it takes to send user input to a remote server and receive a frame of game data from that server. To alleviate the situation, Microsoft Research has been developing a system called DeLorean (whitepaper) using predictive modeling to improve the experienced responsiveness of a game. By analyzing previous inputs in a Markov chain, DeLorean tries to predict the most likely choices for the user's next input and then generates speculative frames that fit those inputs and sends them back to the user. The caveat is that sending those extra predictive frames and information does add a bandwidth overhead of anywhere from 1.5 to 4 times that of a normal streaming game client. During testing the benefits were apparent, though. Even when the actual round-trip time between input and server response was 256 ms, double-blind testers reported both the gameplay responsiveness and graphical quality of the DeLorean system were comparable to a locally played version of the game.
Re:Why? (Score:5, Interesting)
Why on earth would you want to do this? Run the damned thing locally
Let me count the guesses: A publisher paranoid about prohibited copying may be willing to license its game at a lower price if the game program never leaves the server. Or it might be cheaper and faster to send a video stream than to send sufficiently powerful hardware and 50 GB of game at once. Or sufficiently powerful mobile hardware might not even exist. Or it might want to ensure that all players connected to the same server have comparable lag and the same inability to install cheat mods.
Microsoft did something like this once before (Score:5, Interesting)
Back in the 1990s, Microsoft developed something similar. Their idea was to render frames in layers, with the more distant or less active layers rendered less often. if the viewpoint changed, the background layers were scrolled, rotated, or transformed to match, rather than being re-rendered immediately. It never caught on, because graphics hardware became fast enough to re-render everything on for each frame.
This new thing is similar. Mispredicted frames are viewpoint-warped as a temporary measure so the user sees something. The image is wrong, but close enough to look OK until a new rendered frame is sent. It looks OK for Doom, on which it was tested, because Doom is mostly about the shooter and the opponents moving; there's not much general activity in the background. GTA IV/V would probably look much worse than normal.
The whole concept represents a desperate attempt to make something "cloud-based" that shouldn't be.