Negative Latency works, people just don’t understand it
So there’s been a ton of backlash, apathy and head scratching over Google Stadia. The new streaming service from Google designed to stream games is a weird pivot for the company. One major hangup, among many, that people have with the idea is the networking side of things. Stadia needs a pretty beefy connection to function, sure, you can play AAA games on your phone in theory, but the networking overhead is pretty nasty. Google did announce a new innovation that will help the system actually running the game to optimize the delivery of data and reduce overhead, mostly by predicting player actions.
The concept is actually fairly common in some genres of video games. Hardcore fighting game fans will likely be very familiar with one iteration of this idea known as Good Game Peace Out (GGPO). This technology was introduced in 2006 and is used in many different fighting games to take the inputs players put into the game, and the server or game software hosting the matchup will compare those inputs and then generate a predicted outcome. This predicted outcome will be used as a baseline for what happens next in that particular match, with data being simulated and predicted down to the individual frame in some cases. This works by taking variables into account such as what inputs are being generated by players and uses a technique called “rollback” to figure out what is most likely to happen next. In short, GGPO predicts the inputs players send and simulates the next frame without delay using that assumption.
This can create a very seamless system and clients which use this technology for their games netcode have seen a noticeable increase in game quality, and although only some fighting games use this technique, it makes for a better overall experience. Of course there are other versions of this idea, some of which use client-side optimizations rather than server-side simulation, but GGPO and similar ideas are the most commonly referenced when talking about predicting player actions for the sake of lag reduction in games.
There is still an issue though. The core problem is, the idea may not translate beyond the 2D fighting game world where the number of inputs and outcomes is strictly limited. More open games, especially open-world titles, will need to be much more selective in what they try to predict if they want to remain stable. The technology of negative latency as Google is pimping it seems like it won’t work for many games, with major desyncing issues being likely, at least if they just try to copy-and-paste it onto the base code of each game. It’s simple fact that games have to be built with optimizations like GGPO in mind, so you can’t just translate the idea to an already existing game. It will be very intriguing to see how Google gets around this kind of limitation.
This whole kerfuffle is probably one of the reasons Stadia is going with the ownership model that they are, as these games will require specialized clients to even function at all. All this means that some people who want more finite control over the games they buy through Stadia aren’t likely to get it. This isn’t a SimCity 2015 debacle though, Google Stadia actually does need online connections and specialized code to function.
And all of this completely ignores the laughably absurd idea that Stadia will ever be faster than local gameplay. The physical limitations of networking and computing cannot be broken, and innovations that push boundaries take years to develop, there’s very little chance that this particular claim is legit and anything other than marketing buzz.
ISKMogul is a growing video game publication that got its start covering EVE Online, and has since expanded to cover a large number of topics and niches within the purview of gaming.