In their latest email newsletter, GameIndustry.biz takes a far less alarmist approach to dissecting OnLive than I did.
There is simply no readily available comparison for what OnLive is suggesting here. Streaming high-definition video services, even hugely popular ones like YouTube or the BBC’s iPlayer, are several orders of magnitude less demanding – because all they’re really doing, at the end of the day, is serving a pre-rendered file to their users. They’re not doing anything in real-time, unlike OnLive, which is proposing to 3D render an extraordinarily taxing scene, encode it as a high-definition video and send it across the network, anything from 30 to 60 times a second – for every single connected user.
It is truly an enormous undertaking, bordering on snake oil–especially when one considers the controlled environment OnLive’s handlers used to sell the thing to wide-eyed journalists at GDC.
Just to be clear, I’m also doubtful that OnLive can seriously deliver even 1/10th of what they’re advertising at GDC this week. Nevertheless, it’s a first step towards a legitimate paradigm shift in the way we play video games, and I still think Nintendo, Sony, Microsoft and any other company that relies on both software AND hardware to make a buck should take notice.
The concept of what they're talking about isn't THAT far-fetched, and in fact bears no small resemblance to Sony's original idea for the PS3, i.e. networked computing for real-time rendering solutions. The problem is that this method mandates a stable permanent high-speed connection and with many providers starting to stack on bandwidth usage fees, the practicality of On-Live is what will ultimately hold it back…..for the time being. Five to ten years from now, I think what they've got will be perfectly viable.
Nothing changes the fact that it won't work. It just can't. I don't care how good their compression algorithms are, there are basic laws of physics they are claiming to violate here if they're going to state they can deliver lag free streaming play of any game that requires fast reactions. I don't care what they do in a controlled environment, I've been to more than my share of trade shows and seen demos done at a developers booth that looked NOTHING like the finished product. Out in the wild, things change, a lot. Unless you're actually playing the product that's shipping, take it with a grain of salt, and in this case, I'd take it less seriously than that, even.
There's also the fact that you don't truly "own" anything. I don't want to pay a fee to stream a game. I want to own a game.
This whole concept is a joke. I am old enough that I can remember Sega's experiments with VR systems in the early to mid 90's. The processing power available meant that the system would have a slight lag between when the user clicked a button and when something happened on screen. The result: motion sickness, vomiting, etc.
You cannot stream data fast enough from a server to the video display on a remote terminal for 3D graphics. The PCI-Express 16x can barely keep up with today's games (hence SLI configurations). Imagine switching to a socket interface similar to 8-bit ISA and then picture the graphics quality you will get. That is what onlive is facing. Sure once the data is processed by the video card it can be severely compressed but you cannot compress an elephant enough to fit in a carry-on suitcase!!
I imagine this will be filled along with the Phantom console, a cool idea to dream about but in reality nothing but an investors nightmare.
Cloud computing is great for processing large data sets that do not need to be finished in a finite amount of time, and the time must be fractions of a second. This limits the practicality of cloud computing to processing data sets (Seti@home, et all), and some low quality GUI's (word processing, etc). There is no way it will work for games.
Kale: in 5 to 10 years the graphics requirements of the games of the day will require even more bandwidth then what the network can deliver and as such the network speed and latency will never catch up. Don't even get me started on the lag created by encoding and compressing the signal.
On-Live = Biggest Joke from GDC 2009
The rendering and compression are one thing, but I don't think control latency plays as big a role as some people think.
Yes, there will be a delay between when you pull the trigger and when your character fires his gun in the game. The thing is, there is already the *exact same* sort of delay in every on-line game. The action doesn't happen until your input travels from the controller, gets processed on the central server, and then is sent back and rendered on the screen.
That's the *exact same* round-trip, as far as latency goes that we're already accustomed to.
Now, if you throw in the fact that video needs to be compressed on one end and rendered on the other, let me put it to you this way: they claim that they've addressed the latency issue. My guess is that, instead first of running a process that generates a screen, and then compressing that screen, they are directly rendering the compressed screen, or at least some sort of hybrid thereof.
The computationally expensive (i.e., latency-inducing) part of the process is in compressing, not in rendering the compressed content. If they've found a way to directly render compressed video, then they've cut off a huge slice of latency.
My guess is that they're not going to focus on games that require quick reaction times, but even if they do, everyone will have the same sort of delay, and it won't be an unprecedented experience.
I for one look forward to this new innovation in gaming. Since the cost threshold for trying it out won't be very high, I'll probably give it a spin when it comes out.
Burndive: Even if you combine the compression with the scene rendering there is still an additional latency over that of simply rendering the scene. To compress video you need to have an entire frame rendered, this way you know all the information prior to compressing. MPEG2/4, JPEG it is all just a single frame broken into 64 pixels (8×8 blocks) and then compared to the previous frame and then compressed. This all takes time.
Not to mention the fact that an internet connection of 5Mbps does not imply that you will get sustained 5Mbps just peak. Each time a computer (switch, server, etc) touches the data additional latency is introduced.
Even your youtube clips have a buffer on your local machine because they cannot sustain the data rates required to stream the video. So if you have a 500 ms buffer on your end, then all the latency I covered before things compound quickly into a crappy experience.
All this adds up to a cool idea but BS in reality. There is no way this will work in the real world. I hope this thing goes away quickly so we can talk more about the Wii system update, DSi etc.
@burndive: Let me be the first to tell you that you don't know what you're talking about. The client/server relationship in games doesn't work how you're describing, at least in any of the games I've ever worked on or seen. What actually happens is that you do something on the client, based on a combination of prediction code and data received from the server that is combined into a world model that the rendering and client side physics are calculated from, and then the data from your client is sent and reconciled with the interpolated data on the server. You don't send controller inputs to the server. At most, in say the average shooter, you'll send an updated position, a facing vector, and maybe a server synced timestamp. Nothing on your screen is rendered with any latency beyond that in your machine that's running the client. None of this even accounts for the fact that you're talking about online games, which hardly encompasses the whole gamut of games people play. As for "directly rendering compressed video", I don't even know what that means. You still have to process the frame and compress it. Can you do those things simultaneously? Sure, but it's not a magical latency reducer. I don't care if they said they "solved" any issue, plenty of people have claimed such things and they haven't worked in practice. This is bunk.
The obsoletion of the console may be upon us…
I see no insurmountable technical problems EXCEPT network latency. They're claiming 1ms compression latency. With low enough ratios, simple enough algorithms, direct enough rendering to compression handling and dedicated compression hardware I see no reason to claim it's impossible. Broadband can certainly handle the load (they only claimed SD on 1.5 Mb/sec connections and HD on 4-5 Mb/sec in any case). Latency is definitely a problem and something they can't work around if it's on the ISP/Network/User end that ends up routing things inefficiently. Perhaps part of their user license agreement is that the user must expect a few dropped image macroblocks, or they have really good error tolerance, and they get around odd data slowdowns by just forgetting about them altogether.
Bear in mind that the speed of light is pretty fast and that if everything was working smoothly it should arrive pretty quick from a 'local' serverfarm. If not now, then in the near future it seems sustainable when the networks are more reliable.
The economics are REALLY interesting. Sure I like owning games now, but a subscription model also interests me. Say I pay a flat fee for all my obsessive high end gaming. Or I pay for what I play, to the second. I don't like a game? I stop playing. They get less money and the game maker gets a smaller cut. Thus devs are rewarded directly for the quality of their software. Not to mention the MASSIVE savings on CPU/GPU compute costs. Say you use a Cell processor 10% of the day. You paid $100 to not use it 90% of the time. The same processor used 100% of the time does 10X the useful work for the same up front cost. Not to mention all the associated costs with full system hardware and retail…etc…etc..
Having tried it at GDC, i can say that it SEEMS very solid – however, we were only connected from San Francisco to Santa Clara, where the servers are housed, and there were only a dozen or so connections from outside of the Santa Clara location connecting to the service. Who knows how it will work in practice? I'm more than willing to give it a shot – my laptop will only run "new games" on "lowest settings" for another half year at best.
Is there any kind of announcement of when this thing is going to be released? My guess is they announced it really early and it'll turn to vaporware or it'll be released a few years from now when everyone has higher download speeds.
On Live is cool and all but we all know when it comes down to it, this will have absolutely NO impact on the Big N's full-nelson it currently has on the video game market. This thing simply won't be able to make it.
Do you think that any currently Wii-owning gamer is going to spend several hundred dollars on a scary halloween-colored console because it will allow them render crysis in 720p at 60 fps on an HDTV?
Nope.
Do you think a kid is gonna wanna drop Mario Kart in favor of some random PC shooter?
Nope.
Do you think the core Nintendo-gamer is gonna drop Brawl, House of the Dead, Galaxy, Punch-Out!!, or MadWorld for this thing?
Another NOPE.
its a cool concept. but no more threatening than the Gizmondo was to the DS.
I think the point, gojiguy, is that this is not a console, and will not cost hundreds of dollars. It's a cheap video decoder, basically.
And people can play more than one console and game. I play Mario Kart AND "some random PC shooter" (whatever the hell that entails). If this thing actually worked (which it won't), yeah, I could see a lot of people buying one and putting it next to their Wii, and then not buying the next generation of console from Nintendo or anyone else. Would I? No. But that doesn't mean a horde of others wouldn't.
That said, it's a moot point, since this won't work.
Once again I must stress the economics in response to the Eurogamer article. In terms of power the system is remarkably cost effective. Say out of a total theoretical userbase of (lets pretend) 100 million users, only 10 million people use it at any one time at average maximum daily usage. And say 20 million for peak usage on the unusual occasion of Metal Gear solid 5 and Halo IV coming out at 6pm on the same Friday simultaneously everywhere in the world. They would only need rendering/compute capacity for those 20 million X larabees. This is as opposed to every one of 100 million users purchasing the necessary future processors. If every gamer paid what they'd usually pay for a console to OnLive, but OnLive only pays 1/5th of that for the units needed to cover them, then that's 4/5ths of nice profit. More reasonably each user pays less than a full system and OnLive makes more money than full system makers do now. What's more, it means we never need to upgrade ourselves again, the servers can scale economically both with power demand from advancing game technology and with popularity, and power from old hardware remains useful for future compute cycles. Finally, with shared resources such as hard drive space, data in RAM, power, cooling units, casing, lack of disc drives etc. efficiency gains would yield even more power/$ and possibly more power/watts.
elmer, I think you're seriously underestimating the power requirements here. Even with their supposedly custom built servers doing the computing, they have to buy that hardware and keep it upgraded, which is a massive investment. Electricity, cooling and space are all major issues, even with this (mythical) technology. The overhead is immense, even after factoring in increased efficiencies from centralization.
Not to mention, if they've been setting up all these machines, where is the hardware coming from? Where are the massive spikes in orders from major chip manufacturers? There hasn't been a peep on that front, as far as I know. Which means they're waiting for money from venture capitalists (aka suckers) to try and go through with this, which makes the whole thing even more dubious. If the orders haven't even been placed, that means waiting even longer for them to be filled, assembly to happen and the service to be up and running. And all this by the end of the year? Insanity.
The service will scale with users, so obviously they don't need 20 million units now. They'd need 20 million when they have 100 million customers. If they plan on having 1 million customers at the end of the year (probably an overestimate) they might only need 200,000 units. If they've really been going for a while, maybe they have some serious quantities of hardware, though I agree that may not be the case. Either way, operating chips in servers units is more efficient (much more efficient) in energy, plastic, centralised cooling, components, space, cabling etc. per chip then it is in a home desktop or a PS3. One problem is that users may have to pay for their (say) 1/5th of a full spec machine up front until they have enough capital to sustain growth themselves, which is an issue. Also, if the system reaches dominance then total bandwidth may drive isps to destruction.
I think you're seriously underestimating the possible cost benefit. The plan is audacious as it essentially relies on the complete displacement of the existing purchase.upgrade.upgrade.upgrade.purchase paradigm of existing home computing, but a user/available CPU time saving of potentially 5:1 or more, and more saving from resource and sharing centralisation make this concept too brilliant to rubbish, and something seriously worth a tentative investment.
This is actually an old idea (Gamasutra wrote about it some years back). Most technologists said it would be the future of compute intensive processing if it weren't for the latency issues as it's so much more cost effective for power. Frankly I'm surprised people are totally taken aback. Apart from the latency it should seem obvious.
As a concept it's obvious, but once you start getting into implementation, it all breaks down. I think you're overestimating the efficiency gains from this approach. Ask Google how their server farms run. Centralized cooling is offset by the fact that you have a massive farm of chips all in close proximity. Energy consumption is still a major factor, because you're still running large, hot GPUs in great quantities. The cost benefit is negligible to me – I like maintaining and upgrading my computer, but if you're looking big picture, beyond the scope of the company even, then yes, the idea has merit. But latency issues kill it, even ignoring all the rights issues. And no one seems to be touching on the fact that, if they have managed to accomplish these massive technological leaps, why are they squandering them on a virtualized gaming service? I mean, they could be licensing this technology out of billions if it actually worked.
I can't even sum up my hatred for this concept as a whole, I f I thought for a second this was the future of gaming, you could count me out.
"You'll never need to buy a game again; you'll just rent time on the ones you want to play."
Does this even at a concept level scare anyone to their core? As brian said, I don't want to pay a fee to stream a game. I want to own a game.
…not even getting into the fact you're technically playing an FMV not a real, tangible game on your end. I don't know about you guys, but I like the feel of playing a real game on real hardware, knowing the console/PC I built in front of ME is calculating and drawing everything in real time. As it is I get really disappointed when a good game's cutscenes turn out to be pre-rendered, especially when it's clear they used the game engine for those scenes *cough*UT3*cough*. Here it's this all the time, oh sure somewhere systems are doing it for me but meh.
I hope even if this somehow defies everything and works, that people stay far away from it.
@gojiguy
You talk as if gamers are only allowed to have one system – as if being an OnLive customer means one would have to give up their Wii – but it's not like that. Additionally, OnLive could potentially cost significantly less than any system on the market today – AND have higher graphical output. From what I understood from the person who deomed the technology to me – they intend to offer it as a subscription service, and the "TV Unit" might be subsidized. This means access would be comparatively cheap and possibly a better value if there was no additional charge for playing new games. (You could play more games per year for less cost, although you wouldn't own any of them.)
@Reynard
I understand the feeling, but i've slowly changed my opinion. I used to be against digital distribution – it bothered me too much not to "own" a copy of the game – but as long as I know I COULD if I wanted to, I find that i don't care anymore. In a world of Netflix and Gamefly, it's a "rental" world, and I've noticed that younger generations tend to care less about "owning something" rather than simply consuming it. While it still strikes me as odd, I find that I don't care to own most of the games I play, opting to rent them online instead – saving my money only to purchase the game I truly enjoy and want in my permanent collection – and these days that's a scarce few.
(Also DS games, as player data saves on the cartridge rather than your local console.)
@ Ritcher
First: Heat exchange can be done more more efficiently when all your heat is in one place. In fact (not saying they do) a smart engineer could use the heat in one place to drive an engine and recycle at least some useful energy.
Second: The inverse to our Google analogy of holding a centralised index of the internet is having every user's computer have their own reverse index of the internet. While the power demands of Google are shocking, the demands of not having it centralised are absolutely astronomical.
Third: What other application would you have this be used for? Streaming video already exists and isn't nearly as demanding. Application serving can be even more latency demanding (accurate mouse movements on toolbars/pixels on say art packages may require > frame rate temporal resolution), generally take fractional computer power compared to games so may as well be run local, already exist in limited data exchange forms, and even already exist in video form (ever try Citrix or a VNC package? I had one running on my DS). While it seems esoteric, there really is no other application quite like gaming that demands this specific set of capabilities. At the moment anyway.
Look, Scientists do it all the time. They lease out compute time on IBM's supercomputers because they can't afford to make one themselves. When they do buy one for a little department, it goes unused when no one has anything to do and it's wasted potential. Almost all universities now have supercomputer clusters running full tilt all day, with individuals requesting time.
As for owning/upgrading, well very few of us still have the stomach for the upgrade game, and plenty of us were insulted at the idea that Crysis was built for a machine that didn't yet exist. Moreover, I can only afford one modern computer at a time and I choose a laptop as I'm sure is true for many. I'd consider this a godsend. Finally I'm sick of purchasing games, not really getting round to playing them, and then feeling guilty for having spent the money, particularly if I didn't enjoy the bits I played as much as I hoped. I know I can't be the only one here. Similarly I've played a game for so many hours I thought it was disproportionate the makers got so little money for it. This system could solve some of these issues. I like owning games for real. But I already lost something when I moved from functional silicon cartridges to generic pressed discs. I've already given up ever having a cassette tape or a CD ever again. I hardly use a pen and paper outside of diagrams and short notes at this point. Heck, I gave up my answering machine for voicemail. Face it, we're all moving on, slowly, kicking and screaming. Besides, there's no reason publishers couldn't sell you the disc with the game on it so you can play it traditionally, but provide an OnLive license with it if you want to play on a random netbook once in a while.
But yes. There are technical hurdles nearing impossibilities. I won't be the one to claim its un-tenability before they've rolled out however.
@elmer:
"While the power demands of Google are shocking, the demands of not having it centralised are absolutely astronomical."
Sure, but in the second case, Google isn't paying. These guys will be, and their demands will exceed Google's. That's a massive roadblock to me.
"What other application would you have this be used for? Streaming video already exists and isn't nearly as demanding."
Yeah, and streaming video sucks right now. Let's be honest, YouTube is garbage in terms of quality, even the HD stuff. If their tech worked, streaming video alone would be revolutionized, and IPTV could become more than a scattershot-deployed niche market. I won't even get into art packages, I'd go crazy trying to work with one in a virtualized environment.
I also get the supercomputer comparison, but the thing is, my home machine gets used at times other than when I'm demanding top performance with my choice of application. It's not going unused. It's either off, doing something in the background or lending computing power to somewhere else. I don't think it's quite right in this case to compare academic computing resources with the massive server farm they'll be running.
"As for owning/upgrading, well very few of us still have the stomach for the upgrade game, and plenty of us were insulted at the idea that Crysis was built for a machine that didn't yet exist."
I'm sorry, I just can't understand this sentiment. While I can understand some people don't have the financial wherewithal to upgrade their computer, the effort it takes is so vastly overstated, I really just find it annoying. A child could put together a modern PC, it's so simple. As for Crysis being built for a spec that didn't yet exist, if that offends you, you should probably stay away from PC gaming, since that's been a common practice since, oh, around Quake 2's release.
"I hardly use a pen and paper outside of diagrams and short notes at this point. Heck, I gave up my answering machine for voicemail. Face it, we're all moving on, slowly, kicking and screaming."
Hey, I was the among the earliest adopters of these alternatives. I use a tablet notebook, I have virtually nothing on paper any more. I was banging the drum for Steam long before most, and I still buy most of my PC games through it. Not having the physical product kind of sucks, but I can overlook it if I have faith in the service delivering the product, and I can retain a digital copy locally. OnLive seems to take even that away, that we're just renting time on their servers, we don't own anything. Issues of censorship and consumer rights immediately spring to mind when I think about their service. It's distasteful, and I'll go kicking and screaming to my grave if I have to, but I'm going to resist this kind of change, because it's a move away not from the past, but from personal responsibility. I don't think publishers are going to have any interest in providing hard copies if this works, because it will cost them money they wouldn't have to otherwise spend, and eliminates the advantages of the platform, namely piracy prevention and scaled payment options. If this works and becomes popular, the end of physical game media is upon us totally.