Google Reckons Stadia Will Have ‘Negative Latency’

Google Reckons Stadia Will Have ‘Negative Latency’
Image: Google Stadia Launch at GDC 2019.

In one of the more intriguing statements around cloud gaming, one of Google’s chief engineers has said that Google’s predictive work will be able to deal with any potential lag between the client and server. Their term for it? Negative latency.

In an interview in the latest Edge UK magazine, which PCGamesN grabbed at the newsagent, Google’s VP of engineering Madj Bakar reckons Stadia will be faster and more responsive than gaming on local hardware “in a year or two”. I’d normally call bullshit, but here’s the full quote from Bakar going into bat for the cloud.

“Ultimately, we think in a year or two we’ll have games that are running faster and feel more responsive in the cloud than they do locally regardless of how powerful the local machine is,” Bakar is quoted as saying.

So how do you make a cloud game “feel” more responsive than something on local hardware? The answer, apparently, is through “negative latency”. Apart from the bit where it sounds like Google has reinvented the passage of time, negative latency is effectively a buffer where Stadia applies certain techniques to combat lag. It might include predicting user inputs in some scenarios, or a rapid increase in FPS to lower latency, but the general idea is that the cloud gaming model will actively work in the background to identify and combat areas of high lag before they happen.

There’s certainly some quibbles about how rooted in reality the idea of “negative latency” actually is as a term. The principle isn’t too dissimilar from some techniques and netcode implementations we’ve seen in video games, though. The GGPO rollback netcode that was built by the EVO fighting game championship creators predicts and simulates frames before they happen, rolling the game back to the most recently accurate state if the predictions are wrong:

GGPO uses speculative execution to eliminate the perceived input delay for each local player. Instead of waiting for all inputs to reach a player’s simulation before executing a frame, GGPO will guess what remote players will do based on their previous actions. This eliminates the lag experienced by the local player created by the traditional frame-delay method. The local player’s avatar is just as responsive as when playing offline. Though the actions of the other players cannot be known until their inputs arrive, the prediction mechanism used by GGPO ensures that the game simulation is correct most of the time.

Another technique was the one used by hacknet creator Matt Trobbiani for Wrestledunk Sports, where the game runs in a 240fps deterministic space. “Every frame, you send your inputs across the network directly to the partner’s machine. Whenever you inputs from your partner in, I roll the entire game simulation back in time to when that button was actually pressed, add in all the inputs, then re-simulate back up the the present,” Trobbiani told Kotaku Australia.

Why The Creator Of Hacknet Is Making A Sports Party Game

Hacknet is one of the most successful indies ever made in Australia, and to this day probably the best game ever made about hacking. But having spent so many years in the text-heavy world of Hacknet and its expansion, Adelaide creator Matt Trobbiani wants a change. So for his next game, Trobbiani is changing tack completely -- to sports.

Read more

So the idea isn’t as farfetched as it sounds … although it probably wouldn’t hurt if Google could come up with a name that didn’t sound like they were reinventing physics.

[Thanks, PCGamesN!]


  • Bullshit. Predictive lag compensation is always going to make errors at least some of the time, and I think it would make them a LOT if the input device was a mouse, considering the range of possible movements. Keep in mind fighting games generally only have 8-directional, single speed movement and there is no player controlled look – so the thumbstick is essentially 8 buttons, not actually analogue.

    “feel more responsive in the cloud than they do locally regardless of how powerful the local machine is” is obviously bullshit. If I have a computer powerful enough to run my game locally at several times my monitor’s refresh rate with no meaningful framerate drops, that’s… how could it possibly be made more responsive by running in the cloud? The absolute best case scenario with low latency and perfect lag compensation would be that I couldn’t tell the difference.

    • Mouse movements are of course predictable. Theoretically they can go in all directions but practically there will be a higher probability to go in certain directions. They just take a few million hours of ingame mouse movements, throw them inside a machine learning cluster and they get a prediction model.

      You know, in VR it’s already used to predict head movements. That’s more difficult than predicting mouse movements and it already works.

  • So games with predictive networking code already are going to be making predictions based upon predictions? What could possibly go wrong?

    • It will be interesting to see if the same multi-player games are released on Stadia as other platforms. The model is so different that a lot of common practice might not make sense.

      If you are going to do input prediction, you may as well just have it done once on the multiplayer server for all users’ inputs. You don’t have to worry about wall hacks or aimbots with a cloud gaming system since players can’t influence the system running the game code in that way.

  • Interp has been around for a long time, lol.

    Stadia does have some interesting benefits for multiplayer game and latency, but still going to be dictated by latency to your screen and input latency

  • “I’d normally call bullshit”. Then call bullshit Alex because there is nothing new here.

    As someone who battles against the magical “cloud” all day at work, I’ll take a PS5 thanks.

  • Literally impossible? (without being on ultra fibre turbo internet).
    I got good cable NBN (rocking a solid 100/40 basically 99% of the time even during peak). Best ping i have seen ever on mine is like 8ms.
    Assuming the stadia game stream managed to get to me in exactly that its already going to be 8ms slower than running it locally.
    Because either way you still have the screen latency of 5ms+ on IPS or 1ms on TN. Either way you will still have the latency of the PC running the game, because at stadias’s end they obviously have to run the game like any other PC, so again whether local or there, its included. So 16.6ms if at 60hz.
    So even if they can make some predictions on your movement (which seems like a bad idea anyway for any game thats even remotely fast paced and/or not super simple), i doubt its going to save more time than is lost by having to get the stream to you over the internet.

    So with that math its
    Local: 16.6 + 5 or 1 = 21.6ms or 17.6ms
    Stadia: 16.6 + 5 or 1 + 8 = 29.6ms or 25.6ms
    So whats that 36% more latency.
    Then the fact i run all my games at like 120fps which shaves another 8.3ms off the local calculation, when i doubt you will be able to do that on the stadia stream much (whether by limits of the service or by limit of internet.)

    • Your math doesn’t appear to take into account the concept of negative latency.

      And you’re really trying to argue the downside of 30ms vs 22ms latency? I’d throw in some pithy comment about 1st world problems, but it’d take me too long to type out. Perspective, my friend, go seek perspective.

      • And you’re really trying to argue the downside of 30ms vs 22ms latency? I’d throw in some pithy comment about 1st world problems, but it’d take me too long to type out. Perspective, my friend, go seek perspective.

        Depending on the game you are playing that kind of difference is very noticeable.

      • I did mention “negative latency”. Saying it sounds like balony.
        Seems like a bad idea for anything but like turn based games or other non time critical games. Predicting the movement is crazy for a shooter for example. When you do shit like flick shots, or sudden decisions like a grenade throw (so aim up and away from the enemy), also how will it know what i want to shoot and where to predict said aim to?

        And that was AT LEAST 8ms on the internet and i got lucky having better than 95% of people in the country. Plus also as i said thats on 60hz. Again if we are talking games where this latency matters you’ll likely be playing at much higher FPS. So as i said take another 8ms off for 120fps. So its more like 14ms vs 30ms (at least). So we are getting over 2x times slower.
        And i still feel im being generous on what stadia will actually be able to do.

        Don’t lecture me about perspective. We are literally in an article about a streaming gaming platform, having this at all is yes ‘first world’, does that mean i can’t be critical of it? Piss off with that stupid shit. (i dont actually mean that very aggressively, more just that what you are saying is incredibly stupid).

        • You mention negative latency, and then proceeded to completely ignore it in your maths.

          Sure, it may seem like a bad idea to you, but some very smart people appear to think it could be an effective solution. Perhaps you should experience it before being so critical?

          With the average reaction time of elite gamers over 150ms, adding an extra 8ms will not lead to a noticeable increase in performance. Especially for those that are less than elite. This is what I mean about perspective. In chasing every precious millisecond, you’ve lost the idea that anything 30ms or below is actually really great for 99.9% of all gamers out there.

          • I address it as much as one possibly could right now and with the only logical conclusion (given current knowledge). That negative latency sounds like gibberish. and might might be useful for slow games. Tech companies always make fantastical claims that dont turn out (and google has killed tons of ideas/projects that didnt end up panning out how they planned).

            IMO it makes more sense to be critical of it than accept the magic sounding fix just because it would be cool if it did work. Like i can’t think of any reason why it WOULD work, so im gonna be critical until i see it happen.

            I’ll be honest im not even a Hz/FPS snob, i play most singleplayer games at like sub 60 by cranking up settings as much as possible and really only aim for shooters to be a solid 60, with some at 120 because they run easy and my PC is OP.
            But i just dont think this game streaming is gonna work very well at all, and this particular feature sounds like marketing bull.

            Also, sorry for the rudeness at the end of my last post. Now that you cleared up how you meant that “perspective” part i get what you mean.
            But the differences clearly are very noticeable, even after the reaction time. I notice it just the difference between 30vs60fps, 60vs 120fps.
            So i guess there must be something about it being more noticeable when the slowness is out of your control.

  • We’ve already seen implementations like this in games over the years. They’re what lead players to cry bullshit when shot by someone who, on their screen is facing the other direction.

    Honestly, I don’t want the game to “predict” what I’m (or other players are) going to do. It’d be far more game breaking for me to be trying to do something and have the game “glitch” and roll back because it’s prediction got it wrong (hello Diablo3).

  • I can see the value in the research but I don’t see them getting it right enough to satisfy players who know they’re being tricked. On the one hand it’s the only way they’re going to make the games playable and it will probably be the most successful attempt at that to date, but on the other it’s going to make playing obnoxious at times when part of the appeal is to give big processing results with low hardware requirements.

  • Lets be real here…the target market for this is likely casual gamers who don’t care about an extra bit of latency and/or don’t notice it regardless. Anyone who does care about it will spend the money on their own system.

    So comparing your gaming rig or even a standard console to Stadia is pointless, it’s apples vs oranges as far as the market is concerned.

    This is a budget cloud gaming solution for people who don’t want to or can’t afford to pay for their own system, that’s it.

Show more comments

Log in to comment on this story!