Stadia - Google's answer to consoles - streaming

[Erloas]

So I haven't noticed a post yet, it was just announced but not a very well kept secret, that Google was getting in on the console gaming scene.

https://www.businessinsider.com/goog...essions-2019-3
https://www.ign.com/articles/2019/03...latform-stadia

Here is a quick two links of at least the basic first impressions by the media.

There isn't actually a console, it is stream of the game being ran on Google servers.

They are recommending a 25Mbps connection speed and latencies less than 50ms. They haven't given pricing yet, but I think just that requirement is going to make it a non-starter for a lot of people. That is of course for 1080p content, assuming that the 4k they are touting is going to require closer to 100Mbps. I'm going to guess this pretty much eliminates it as an option for anyone not living in a big city.
I'm in a Seattle suburb and I don't think my latencies are that good. I don't know about to Google, but to the game I'm currently playing I'm only about 150 miles from their offices and my latency is about 90-100ms.

Although my computer is getting old and there are quite a few games I can't play, or at least not very well, I'm still not feeling like jumping into a streaming service like that unless the prices are pretty low. Especially if the games cost more on top of the service fee. I could maybe see it being worth a lot if all the games are included in that cost, but for the most part having 100s of games to play isn't that interesting to me. I mean I've already got dozens of games in my Steam library that I haven't played, and quite a few FTP games too.

They tout the ability to easily stream to YouTube, in fact it will do so directly while you're playing so you don't have to do anything extra, which is a feature I've got exactly 0 interest in, from both a player perspective and a viewer perspective. Granted I'm probably in the minority in that I've never watched any gaming content on YouTube and in general watch almost nothing there.

So what does anyone else think?

[Blackhawk748]

I'm sorry I would like to actually own my games thank you. On top of this, streaming a movie is simple, it is only about 1 gig on average. Most games are 10+ so good luck with that Google, you don't have the best track record wit hthis sort of stuff.

[Erloas]

On top of this, streaming a movie is simple, it is only about 1 gig on average. Most games are 10+ so good luck with that Google

It isn't streaming the code/installation, it is streaming the video. So streaming 1920×1080 at 30 FPS will take the exact same amount of data, if that is a movie or a game or whatever else. The main issue though is most movies are 24 FPS and they're talking about 60FPS, so 2 hours worth of streaming at 1080@60FPS is going to be about 2.5x more data than 1080@24FPS. There would also be the added data for controls, but that should be minimal compared to the video.

It would also be worth mentioning that at least for PC gaming, 1080p is pretty old, most newer "gaming PCs" are running much higher than 1080p screens. (Older or low/mid-range PCs are probably still mostly at 1080p)

[warty goblin]

It isn't streaming the code/installation, it is streaming the video. So streaming 1920×1080 at 30 FPS will take the exact same amount of data, if that is a movie or a game or whatever else. The main issue though is most movies are 24 FPS and they're talking about 60FPS, so 2 hours worth of streaming at 1080@60FPS is going to be about 2.5x more data than 1080@24FPS. There would also be the added data for controls, but that should be minimal compared to the video.

It would also be worth mentioning that at least for PC gaming, 1080p is pretty old, most newer "gaming PCs" are running much higher than 1080p screens. (Older or low/mid-range PCs are probably still mostly at 1080p)

It's actually worse than streaming a movie, because you know all the frames in a movie in advance. Because video compression basically works by sending you the differences between the current frame and the next frame, you get substantially increased efficiency knowing multiple frames in advance. This is also why for instance streaming movies tend to look worse right when you start, or right when a scene changes; there's just too many differences to give you a high resolution map of the frame differences. Also why videoconferencing suffers from worse compression problems than streaming movies.

Long and short of it is that for something like a game where they can't pre-calculate enormous amounts of stuff, it's going to take substantially more data. And because they have to keep the latency down, they can't spend all that long compressing either; which again isn't a problem for streaming because the compression can be entirely precalculated.

Beyond that, I struggle to see the appeal from a consumer's point of view. Even if one has a very low-latency connection, you're probably looking at a substantial hit to video quality due to compression. There's potentially some cool stuff you could do in terms of throwing enormous compute power at a game from a development point of view, but that's also going to be very expensive to develop, i.e. not worth it unless you already have a large customer base. So until there is a big user base, it's basically pay a monthly fee plus cough up for serious internet to... play the same stuff you could play already. Maybe it's just me, but I can't see the appeal of paying whatever a month for access to the 'console' unless it also included the game, and that cuts the profit margin for the developers down substantially.

(and that's assuming you even can cough up for sufficient internet. Where I live, that's flat out not an option.)

[Grey_Wolf_c]

There are also reports that the browser you use to play it’ll require something like 4GB of RAM, so don’t be so sure that it’ll work for older PCs not capable of running the game locally - I know my old laptop doesn’t have that much memory to spare

Grey Wolf

[Kitten Champion]

My Youtube experience is often quite crap - lots of freezing and having to reload to get through videos - so I'm going to have to give a hard pass.

Also, Google's got a long track record of failed digital services that were supposed to upend the market and be revolutionary, I'd rather my gaming money not go down with the new shiny.

[Zevox]

Wait, this thing's supposed to be a console? I did not realize that. I just saw the words "streaming service" used to describe it, assumed it was some new competitor for Twitch, and moved on because I wouldn't care about that.

Honestly, sounds like I still don't. If the games are all being streamed from somewhere else, that just sounds strictly inferior. Definitely for games where timing matters, such as the fighting and action games I play a fair amount of, it sounds like it would introduce extra input delay and/or straight-up lag, which is totally unacceptable for those. And even for ones where that doesn't make much of a difference, I don't see much of an advantage. I guess you don't need to have the game fully installed and take up that much space on your console's hard drive, but that's a minor annoyance for the most part, not something that's remotely worth the trade-off here.

Plus, well, it's a Google product, so it'll probably wind up looking a lot like the X-Box games-wise. All the multiplatform titles, sure (if it catches on at all at least, maybe not even that if it just crashes and burns), maybe a few exclusives if Google makes any games themselves at some point, but probably getting ignored by all of the Japanese developers. Which is where a lot of the games I prefer come from, so that's a big mark against it out of the gate for me.

...wait, did I just read in one of those articles that this thing "is not a piece of hardware?" So, what, it's just something you download to a PC, not an actual console itself? Sounds more like a competitor for Steam then, not a console at all.

Yeah, very much still in the "I don't care" category for me.

[Erloas]

...wait, did I just read in one of those articles that this thing "is not a piece of hardware?" So, what, it's just something you download to a PC, not an actual console itself? Sounds more like a competitor for Steam then, not a console at all.

Yeah, "console" is more used as a "this is the market we're targeting" rather than the actual device. They act like it can be used with anything that can run the Chrome browser. So you could use it on your phone if you wanted, assuming at least that your phone can handle the network demand (ie your on Wifi). Although they do have technical specs of the equivalent hardware that will be used on the server side to run said programs.

And while you can use almost any controller to play, their Stadia specific controllers connect directly to wifi to pass data to the service without having to go through your streaming device of choice, which should help reduce latency.


In theory this should be able to run anything that can be ran on a PC, at least from the technical standpoint, licensing is another story. Although I think they said it is actually going to be a Linux based server so they can run all Linux based games...

[Kaptin Keen]

So streaming 1920×1080 at 30 FPS will take the exact same amount of data, if that is a movie or a game or whatever else.

Well - it's not exactly the same. But, sure, extremely close. But input from the controller and so on needs to be be transmitted back and forth, so there's ... somethings besides the visual data. Also, maybe, there's something about encoding? I dunno. Can you live-encode game data to video format? Never mind - my point is that 'exactly the same' isn't 100% accurate.

[Rodin]

From what I gather from the articles, the appeal of this is mainly people who don't have high-powered gaming equipment. The example they give is playing Assassin's Creed Odyssey, which you could play on its highest graphics settings on a PC that wouldn't be able to run the game at all, and then you could go mobile and play it on your iPad. They're basically trying to design a Nintendo Switch, only more powerful because all the processing power is being done somewhere else.

So, I can see why people would want to use it. In theory, it's both cheaper and more flexible than buying a high-end gaming rig or a console (or both). I'm mostly just dubious that we're there in terms of reliable Internet. Video game responsiveness is a thing and I have a hard enough time keeping my streamed TV watching consistent. The other thing of course is game lineup, which from the IGN article doesn't look great at this time. That's really the problem all new entrants into the "console" market have - who are you going to get on board that isn't satisfied with the current options?

[snowblizz]

From what I gather from the articles, the appeal of this is mainly people who don't have high-powered gaming equipment. The example they give is playing Assassin's Creed Odyssey, which you could play on its highest graphics settings on a PC that wouldn't be able to run the game at all, and then you could go mobile and play it on your iPad. They're basically trying to design a Nintendo Switch, only more powerful because all the processing power is being done somewhere else.

If you don't have a high end computer will you have a good enough connection?

And if you have any interest in gaming, then normally you make sure you have adequate gaming equipment.

I can't really see who their market is on this.


I've streamed my own games via Steam at home from gmaing PC to a Surface3 in the living room but eventually comes down to the experience not beign good enough. This sounds simialr but with a couple orders more magnitude of issues.

[Hunter Noventa]

Yeah I heard about this and thought 'Nope'. The technical limitations aside, the vast majority of the US doesn't have high enough speed to support this. Maybe if everyone had Google Fiber, but that's basically dead in the water too.

[snowblizz]

Maybe if everyone had Google Fiber, but that's basically dead in the water too.

Aha! The masterplan takes shape...

[Maryring]

This'll be interesting to see. I expect it to fail, because unless you're playing something turn-based you'll really notice the latency. Especially in things like fighting games or precision platformers. But maybe they'll make it work. The base idea isn't bad, and I could certainly see some games transform into this kind of service after the early attempts crash and burn. But until then, I'm satisfied with my Switch.

[Kaptin Keen]

the vast majority of the US doesn't have high enough speed to support this

You don't? Then who are all those people streaming twitch and youtube and so on? They can't all live in Denmark, I'm guessing.

[Hunter Noventa]

You don't? Then who are all those people streaming twitch and youtube and so on? They can't all live in Denmark, I'm guessing.

I'm lucky enough to have very fast internet. But there are far far too many places in the US where your max download speed is 25Mbps if you're lucky, and you still have a 500gb a month data cap on top of it. Anyone with a datacap won't be able to use this if they want to do anything else with their internet.

[warty goblin]

Well - it's not exactly the same. But, sure, extremely close. But input from the controller and so on needs to be be transmitted back and forth, so there's ... somethings besides the visual data. Also, maybe, there's something about encoding? I dunno. Can you live-encode game data to video format? Never mind - my point is that 'exactly the same' isn't 100% accurate.

So, very roughly speaking, there's two ways you can send game data. You can send the calls to the renderer, i.e. Character moves here, or you can send the rendered video of the action, i.e. the frame where the character moves there. Sending just the calls is very efficient in terms of network requirements, because it's not very much data, but the person receiving the data has to actually do the rendering. This is how multiplayer games work; what's transmitted isn't video but the data the game needs to generate the video. Sending the video is much less efficient, since it's a lot more data.

Basically what Google are proposing is that you send them the game input, or some function of that, they tell the game to process that input, render the video, compress the video, and send it back to you, where your client decompresses and actually displays it.

The advantage of this is that they can throw a lot of computer power at the rendering step, so they can run things at very high resolution and graphics settings. Because cloud hardware is cheap compared to owning your own, you can potentially throw a lot more computer power at it than you'd be able to afford otherwise.

The two major downsides to this are the video compression, and the latency. It doesn't really help if you're running the game at maximum resolution and settings in the cloud if it gets compressed way down and picks up a bunch of pixelation by the time it gets to you. As I mentioned earlier, compressing things live is less efficient than compressing them frame-by-frame, and you can't spend as long running the algorithm either. But it's quite possible that they've more or less solved this.

Latency is much harder to solve, because there's some fundamental physical limits involved. There's gonna be more delay when sending your controller input to a Google server farm than to your home console/computer, because the server farm probably isn't sitting in your living room. Unless they're sending data FTL, the latency is going to be higher. This isn't really about how fast your internet is either, in the sense of how long it takes to download a megabyte of data fast, but how long it takes the signal to reach the servers. What sort of cable connects you with the server matters, but so does physical distance.

Of course this could be another case of a tech company forgetting that people live places besides San Francisco. *coughSelfDrivingCarscough*


And unless I'm going totally crazy, I seem to recall this being tried like two or three times before. Remember OnLive? Remember how Crackdown 3 was gonna do a whole bunch of rendering in the cloud?

[Blackhawk748]

Remember the Phantom and the catastrophe that was? Honestly, technical limitations aside, I want to actually own the flipping game I am playing. The only reason I feel ok using Steam is because they have to allow you access to your game library even if you're banned from everything else, so they can't just take it all away.

This mess just gives all the power to Google and it requires you to always be on the internet. So ya, those of us who live out in the Not-San-Fransisco area of the country are going to have terrible lag and more incosistent internet speeds, making this a flop before it launches

[Kaptin Keen]

So, very roughly speaking, there's two ways you can send game data. You can send the calls to the renderer, i.e. Character moves here, or you can send the rendered video of the action, i.e. the frame where the character moves there. Sending just the calls is very efficient in terms of network requirements, because it's not very much data, but the person receiving the data has to actually do the rendering. This is how multiplayer games work; what's transmitted isn't video but the data the game needs to generate the video. Sending the video is much less efficient, since it's a lot more data.

Basically what Google are proposing is that you send them the game input, or some function of that, they tell the game to process that input, render the video, compress the video, and send it back to you, where your client decompresses and actually displays it.

The advantage of this is that they can throw a lot of computer power at the rendering step, so they can run things at very high resolution and graphics settings. Because cloud hardware is cheap compared to owning your own, you can potentially throw a lot more computer power at it than you'd be able to afford otherwise.

The two major downsides to this are the video compression, and the latency. It doesn't really help if you're running the game at maximum resolution and settings in the cloud if it gets compressed way down and picks up a bunch of pixelation by the time it gets to you. As I mentioned earlier, compressing things live is less efficient than compressing them frame-by-frame, and you can't spend as long running the algorithm either. But it's quite possible that they've more or less solved this.

Latency is much harder to solve, because there's some fundamental physical limits involved. There's gonna be more delay when sending your controller input to a Google server farm than to your home console/computer, because the server farm probably isn't sitting in your living room. Unless they're sending data FTL, the latency is going to be higher. This isn't really about how fast your internet is either, in the sense of how long it takes to download a megabyte of data fast, but how long it takes the signal to reach the servers. What sort of cable connects you with the server matters, but so does physical distance.

Of course this could be another case of a tech company forgetting that people live places besides San Francisco. *coughSelfDrivingCarscough*


And unless I'm going totally crazy, I seem to recall this being tried like two or three times before. Remember OnLive? Remember how Crackdown 3 was gonna do a whole bunch of rendering in the cloud?

I think you're saying the exact same thing I was - just from the point of someone who actually knows what they're talking about. All I have is a basic understanding that Process A (rendering the video) can be done either at Location Z (your own CPU) or Location Y (a giant pile of Google Hardware).

And then, I'm pretty sure, there has to be some data transfer besides the raw video - although most of the data would likely be the video.

But I totally agree that latency is a real problem - while the additional data isn't.

[DavidSh]

It sounds like the original version of Playstation Now, where Sony had a server farm which would host games, treating your console as a remote terminal. I never heard how that worked out in terms of required bandwidth and latency for a playable session.

[FireJustice]

It's the Google Glasses all over again.

Streaming Video is easy. You can download ahead. You can buffer.
A game? not so much. You have to read the inputs too.
only 50ms of latency? yeah right..
The "game" will have the input on your controller/keypad/etc send through the internet to the google megaserver that will in turn execute the command to the game, render the video output, then send back to you.
Let's assume it have the best codec possible to compress and send it back to you, and your "console" will uncompress and show the video only.
IMO its one or two generations ahead of current streaming technology. If you have hardware good enough it could have some hardcore video prediction so you can use some kind of buffering. Then again, if you have
hardware for that, why not just render the game locally.

Also, bandwith and bitrate, the main problem of streaming today.
Also- also. Traffic Shapping from your ISP. If the squirm with Netflix and video streaming,..

Somehow google will solve the problem and you will get 60fps low latency gaming streaming and make it affordable too as a service.
Always online to play policy. One of the worst features.

and people are calling this the "Hardware killer..."

[Kaptin Keen]

only 50ms of latency? yeah right..

Even so, 50ms is 'high', right? It's unacceptable for pro play, I know, and I'm pretty sure you need the pro e-sport scene to pull in the early movers, who'll drag along the slowpokes like me, eventually.

Doesn't that sound ... about right?

[tyckspoon]

Even so, 50ms is 'high', right? It's unacceptable for pro play, I know, and I'm pretty sure you need the pro e-sport scene to pull in the early movers, who'll drag along the slowpokes like me, eventually.

Doesn't that sound ... about right?

Serious pro events for twitch/response-time sensitive games are usually done in a LAN environment when at all possible and often with provided systems to try to ensure as even a base as possible .. 'course, it also means the sponsoring company has a lot more control of the layout and presentation, which is really appealing when they're trying to make money on advertising and want to make sure their official commentators aren't going to go wildly off-script.

So I suspect if you have an event that is 'hosted' via Stadia what's going to happen is the show runner will make a big deal out of all the competitors playing on thin client systems/out-of-the-box lower spec Dells or something.. and then the farm that's actually running the games is gonna be right next door and directly wired into the player's systems.

[danzibr]

Eek. Yeah the idea is nice, but dat input lag.

[Erloas]

Even so, 50ms is 'high', right? It's unacceptable for pro play, I know, and I'm pretty sure you need the pro e-sport scene to pull in the early movers, who'll drag along the slowpokes like me, eventually.

Doesn't that sound ... about right?

If their target market is "pro gamers" they may as well call it a failure now. That isn't going to draw in developers, and probably not even worth it for the "pro gamers" either since they've already got the top end computers and part of the draw is playing games that people know and can play themselves, so not exclusive to the Stadia games and if they're not exclusive then the pro gamers would probably be better off on dedicated hardware.

As mentioned before, I'm just outside of Seattle proper, (I can drive past Amazon HQ or Microsoft in about 20 minutes from my apartment), which is one of the top tech hubs in the world. I've been running a few speed tests from work and from home, and while there is some variation based on which test site I use, pretty much the only servers I can get sub-50ms (around 30ms) pings are located in Seattle. Even going to CA or SLC Utah servers (the next closest choices) my pings get into the 60s, and anything on the east cost are around 100ms. That is both from my home line and my work. So if Google wants to keep pings low they're going to have to have these servers in pretty much every state, maybe multiple locations in larger states. Which basically means large parts of the country are never going to have good access to it, and there are going to be entire parts of the world and countries that aren't going to have servers close enough for this to work for them.


Someone said that server hardware is cheaper, but it isn't. Generally it is more expensive. The advantage to server hardware though is that it can be better utilized. So if your average player only plays a few hours a day that "console" you sold them can be used by someone else all the rest of the time. The $1000 worth of hardware on the server is a lot more than say a $300 console (cost to make, rather than retail cost), but that $1000 worth of hardware can probably cover 4-10 players worth of sales... except that only works if you can balance things correctly. But to keep latencies low most of your "west coast players" are going to have to be on "west coast servers" so they're all going to be playing at mostly the same time, you can't use that same hardware to cover European players that play at different times because the latency will be too high. Normal cloud operations work that way though, because taking 30ms or 200ms to stream a show, or load a webpage, doesn't really matter. You can load balance across the whole world (if you're a big company like Google) for the majority of server requirements and don't need server farms everywhere. But in this case your high demand and low demand times can't be balanced by geography because that will kill the user experience.

[Zevox]

Yeah, "console" is more used as a "this is the market we're targeting" rather than the actual device.

Uh huh. Yet, the more I read about this, the more I think they don't understand console gamers. Because a lot of this is reminding me of a console that crashed and burned so badly that it never even released: the X-Box One. Microsoft's original plans for it, anyway, which they scrapped a mere week after announcing them. Always online and no physical media, check - and this goes one step further on the latter and has you not even own a digital copy of the games you're playing.

And you can add onto that the fact that it's not even an actual device, so at least some of the convenience of consoles as dedicated game devices will be lost there; Google not having the first-party exclusives that Nintendo, Sony, and Microsoft have to offer for their devices; and the potential for problems with lag/input delays that don't exist on consoles unless you're playing online multiplayer modes. Yeah, it's really hard for me to see why they think console gamers would want this.

[warty goblin]

It seems to me that one place where this could be advantageous is very large playercount MP games. The difficulty with getting lots of players into the same instance with the current model is that you have to send every client the information it needs to draw every player's screen, which is loosely speaking all players within some radius of each player. It's a lot of work for a server to figure this out quickly, and it needs to transmit all that data to each player's client constantly. If there are P players, and A agents (including other players) that each client needs to know about, then you need to transmit data describing PA things to P players, which grows very fast in P.

If all the rendering is done client-side however, you just need to be able to render each client's view, which, since it's in the cloud, means you can throw lots of compute at it very easily. In other words a fully server-side EVE Online could potentially not slow down massively during big battles; same with a server-side Supreme Commander, in a fantasy world where people still throw millions of dollars at developing giant cutting edge RTSs. You could also get a lot of gain for complex physics in a model like this, where the computation is done once, then rendered from multiple points of view and sent to each client.


But games with enormous numbers of agents aren't really where it's at these days. MOBAs don't need anything like this level of tech, ditto Battle Royale games, and MMORPGs really aren't such a thing anymore. Sure WoW still has a couple million players, but I think we're past the point where publishers are tossing giant stacks of cash into trying to make a WoW-killer. So sure this could solve some problems that would allow some theoretically interesting sorts of gameplay. And pretty clearly you already can do pretty substantial physics with reasonably large numbers of players; just look at Battlefield 1, so you'd need to push to genuinely extreme cases to show substantial advantages. I mean my computer's a potato, but it still makes Battlefield 1 look better than streaming realtime video off the internet.

[Kitten Champion]

It occurred to me that the one of the biggest stumbling blocks for this - at least from a games publisher's perspective - is that it won't have any access to the Chinese market.

Which is the largest market in the world presently for the medium, and a major influence on how publishers have been directing their content.

[Erloas]

If all the rendering is done client-side however, you just need to be able to render each client's view, which, since it's in the cloud, means you can throw lots of compute at it very easily. In other words a fully server-side EVE Online could potentially not slow down massively during big battles; same with a server-side Supreme Commander, in a fantasy world where people still throw millions of dollars at developing giant cutting edge RTSs. You could also get a lot of gain for complex physics in a model like this, where the computation is done once, then rendered from multiple points of view and sent to each client.

Actually it wouldn't help EVE. The problem EVE has is that in huge battles even the server can't keep up. TDI is slowing down everything so the part of the server handling that node doesn't crash, it isn't in any way driven by client side performance. The way the systems/nodes are set up means that the rest of the server doesn't slow down, but a single system (processing task) can't be distributed any more and the processor running that task is doing as much as it can. Granted in the huge battles, both the server and the client are struggling to keep up, so someone without a powerful machine is going to only be running at a few fps, but that only kind of matters because the server couldn't handle it faster even if the client could.

Cloud computing is only powerful if you've got a task that can be broken up into many small tasks and games can only be parallelized so much. You're average server isn't any more powerful than your average gaming PC. It is just that the tasks a server handles are different types of tasks compared to your average PC. So if you asked the server to do "regular PC types of tasks" they aren't going to be any better at it than a regular PC.


*That is of course a simplified generalization, but I think I've got the main point illustrated.

[Kaptin Keen]

If their target market is "pro gamers" they may as well call it a failure now.

If you're not going to reply to what I actually said, then please don't quote me.