GDC: Why OnLive Can't Possibly Work Comments by Richard Leadbetter

Things that I see OnLive being

OnLive won't be aimed at the hardcore gamer, more like the casual gamer whom has an average computer.

No doubt their pricing model will be based on a usage model i.e. Pay as you go, 10 hours a month, 20 hours month, unlimited etc.

No doubt they will introduce a pre-play system to cater for new releases, giving priority to say unlimited usage accounts.

@Unclebenny

Basically, this guy is claiming they can do virtual desktop in 720p at 60 fps for the latest games. For thousands of people. At the same time. You might be right to question this claim.

This comment thread is made up entirely of people agreeing that he's talking bullshit, except for one person who is deluded into thinking it can work.

That's both the article and the comments in a nutshell.

@beemoh

What you suggest would involve less work at the server end, but much more network bandwidth used.

Speed server side is a non-issue. They own the hardware, they can throw resource at it (they can do it fast by implementing it in hardware).

Video compression is all about sending the difference between this frame and the last one. The time consuming bit is working out how few differences you can get away with sending, and expressing that in as few bytes as possible. Then every so often you transmit a whole frame to fix any accumulated 'badness'. If you watch highly compressed video, you'll get a feel for it.

Troll king is troll.

We will need fibre optic net connections and the server has to be geographically near by since the speed of light isn't fast enough for near-zero-latency.

Rich is right, this stuff requires future generations of technology in several fields, im going to hazard a guess and say a decade+.

No one can read the future so no one can say this will never happen. But OnLive may be an idea to ahead of its time.

According to the GDC talk 'geographically near' means 1000 miles.

He actually admitted that an East Coast -> West Coast (US) deathmatch wasn't practical.

Finally, somebody thinking this though.

Now please restore that poor butchered irregular verb in the final sentence.

I find it strange for the best thing these masters of psychophysics could come up with as a game controller to look so much like a no brand 7 year old 10€ dualshock clone for the pc..

@ukslim: aha. Taverymuch.

Someone asked about bandwidth requirements for uncompressed video:

A 1280*720 pixel stream, using 24 bit colour, running at *30* FPS woud require 632.8125 *megabits* per second. That is a lot. Lossless compression even in a best-case scenario would not get that below 160 megabits per second.

A motion jpeg of the same stream would require (depending on quality but let's say "medium") about 30 megabits per second. Still a lot. This is what a previous quotation about the OnLive technology said they were doing by the way - not encoding using P and B frames, each frame only referencing itself. I don't believe it.

An H.264 stream can be got down to about 5 megabits per second - I have used multi-pass encoders that get great results at this bandwidth, but have no personal experience with a live encoder at this bandwidth. They needed about 9 megabits per second when I was using them.

So the compression ratio for h.264 over uncompressed video is about 126:1 which is quite impressive really.

@Darren: Exactly. And in ten years time, 1080p will seem just as crappy as 720p does now, anyway. This wont really work, at least for graphics-conscious PC gamers, unless broadband connection speed suddenly explosively outpaces screen resolutions.

If you don't really care how it looks, obviously this is a non-issue.

@udat

I think the statement about every frame being a key frame was a distortion of what was said in the GDC talk. What was actually said is that they have their own new codec, for which the decoder doesn't need to buffer more than 1 frame ahead.

@ukslim

That makes sense, and I am glad I didn't believe it :) Doing that will make it more responsive but it will have a negative impact on compression efficiency. Probably a worth trade-off in this scenario. Anyway, encoding latency is not the "difficult" bit of this technology.

OnLive is a brilliant idea. That is of course if it works. They plan to have beta testing as soon as this summer. After that it will launch in the winter of this year if everything goes as planned. They say there is no lag, and speeds are good, but what happens when there are 1 million people playing at once? I guess we have to wait and see.
Check out The OnLive Fans News Forum

i smell bullshit, but at the same time, i see the potential if it ever did work

the idea of freeing us all from escalating hardware is a brilliant idea. i just really dont think we live in that time yet

BobsUncle-I get the general idea. Essentially everyone playing games streamed of some giant monster computer in a bunker somewhere. so they dont have to own the hardware themselves. Then everyone seems to be saying its impossible.
I was just confused as to how no one else on the website was reading the article and not understanding most of the words used. Its all crazy techno babble to me.

thanks for taking the time to lower yourself to my level though :).

@Baranga

You are the worlds first OnLive fanboy! Congratulations!

I'm a big skeptic like most of you. However, if this thing somehow does get off the groud there are some games that would work ok with the input latency. Rule out action-based games and you are left with the following genres:

- Football/sports Management games
- Card and Board game recreations
- Quiz games
- Word games
- Puzzles
- Suduko

Sounds like a gold mine, huh?

Basically, this guy is claiming they can do virtual desktop in 720p at 60 fps for the latest games. For thousands of people. At the same time. You might be right to question this claim.

This comment thread is made up entirely of people agreeing that he's talking bullshit, except for one person who is deluded into thinking it can work.

That's both the article and the comments in a nutshell.


THIS

If this works Microsoft will buy this comapny & it will be the end for Sony & Nintendo.

As a full time networks & distributed systems researcher (Eurogamer, feel free to contact me if you want a formal quote), I can't see any way this be feasible. I mean, sure, on a controlled network you could do it... for far more than it would ever cost to just host the stuff locally. In reality, I'm not even sure where to start with problems...

If they genuinely have a demo working with the servers 50 miles away, that's impressive already. I'm not an expert on video encoding, but the comments by the person quoted in the article seem accurate to me. Getting control input from the controller, into the computer, through the network stack and out, down the network connection, into the server, processed and back before the user notices? Difficult already, as I believe most MMOs will demonstrate on request. Even if they can get it to all work in time under perfect conditions, the first packet loss, damaged packet or routing hiccup will almost certainly bring it crashing down like a house of cards.

We can then start on the data centres. High-end servers already struggle with their power consumption and heat output being major issues, and they want to stick graphics cards into them? If you think your desktop runs hot, imagine a room with a few thousand of them in. So, they somehow get data centres full of gaming systems and enough cooling to run them (this is all a ploy to get funding for their Antarctic base, isn't it...), how do they got enough bandwidth? As much as ISPs complain about the bandwidth costs from iPlayer etc. this is going to make iPlayer look like a drop in the ocean, and that includes someone having to pay for server-side bandwidth for this stuff.

If they've got video encoding hardware that good, I'm sure many MANY companies would pay them for it, for camcorders, live events, etc.

If they've got latency managed that well, Blizzard and many other MMO producers would readily pay for anything that would get their costs down.

To claim they have technology that does both, and this is how they want to use it? Insanity.

Oh, for latency; speed of light isn't your constraining factor, it's routing and processing time. If you can run a single lump of fibre from one end to the other, you're sorted (sub-millisecond latency to anywhere on the planet). However, for any real network your traffic spends most of its time being checked, processed, batched together with other packets, and generally being stuffed in the right place by network hardware. It's like comparing carrying a parcel from A to B yourself, and sending it by post.

Edit: My bad, speed of light is an issue. Light in a fibre is 2 x 10^8 m/s, which works out at 200km/ms. The earth is 40,000km in circumference, for reference.

I knew this was bollocks. again, I've done plenty of work with video encoding and streaming, and it all seemed like a pipedream considering the demands placed on the server side and the network connection.

Again, I'd love to be proved wrong here, but I'd like to see this working outside of the lab. My home connection is 5.5Mbps uncapped at the moment. I should be able to run the HD service - but there's no chance I reckon both due to my connection (and its contention ratio), and that it's unlikely that this will scale up very successfully to multiple thousands of users.

/waits

Oh ye of little faith.

Once in a while technology leaps forward. After reading the article, I have an image in my head of a circa 1950 switchboard operator trying to comprehend how a single switchboard could ever handle more than 5000 calls each day. She would have to grow fifty new arms in order to connect all those calls!

To me, the 30 fps video looks just fine, and I feel convinced that this is what the guys are going for. Maybe they dont plan to offer all games, but maby focus on "casual" stuff, leaving the twitch gamers to sony and MS. For now. Maybe they actually have a well though out strategy laid out already.

Oh, and in the future, bandwidth will be free. I've read it so it must be free. In a world where virtually limitless bandwidth is free, something like this will be the dominant gaming platform (although you'll always have enthusiasts). Except, maybe the publishers will stop releasing titles on "enthusiast platforms", seing as "enthusiast" is pretty much synonymous with "software piracy" anyway.

How has gaming changes over the last 10 years?
How will it change over the 10 years to come?
Is OnLive the shape of things to come - maybe not this decade, but surely in the next one?

I said it had to be an april fools joke because it was impossible to actually do and got called names. Funny how suddenly everyones a sceptic once an article pops up that noticed the exact same claims I made.

ho hum...i guess that's what I get for always being one of the ones on these forums that's ahead of the crowd.

It has very little to do with bandwidth and a lot to do with scalability and latency. Generally you want things to get easier the more popular they become (e.g. mass producing things gets cheaper as the thing becomes more popular) and this will have the opposite problem.

Good read that, I've got my own thoughts on OnLive here at my blog. It's not as detailed as this though which makes this pimping rather pointless (Y) toodles

http://videogamesatemybaby.wordpress.com...

@Olemak: "Once in a while technology leaps forward." - no it bloody doesn't. The transistor may have enabled computers, but was designed in 1947, not a couple of days ago. The Internet is rolling in for 40 years ago. Even the web is over a decade (first formal spec is dated 1996, first browser 1993, and there was Gopher before that). Space flight is as close as you get, and that was because two superpowers through everything they had at the problem, not a couple of dozen guys no-one has heard of before.

"the idea of freeing us all from escalating hardware is a brilliant idea."

So you'd like to rent all your games with a subscription, would you? It's not brilliant, it's a crappy idea.

Olemak, as has been pointed out both in the article and the comments, what they are claiming is just not feasible. Yes technology moves forward, but not at that pace. They're talking about server farms stuffed with high-end PCs streaming massive amounts of data all over the world. It's ridiculous. Google and others have enough problems cooling and maintaining the server farms that keep their search engine and other ventures running, and they're not fitted with cutting-edge graphics cards.

This is like DRM's love child.

@anomagnus: "the idea of freeing us all from escalating hardware is a brilliant idea."

Yes, it is, but it's just escalating somewhere else. You might as well set up a subscription with Dell to have them deliver you a new PC every two years...

If you can run a single lump of fibre from one end to the other, you're sorted (sub-millisecond latency to anywhere on the planet).

That's not true, is it? Light can only travel ~300KM in one millisecond.

I think lots of commenters haven't seen the GDC presentation:
http://www.vg247.com/2009/03/25/gdc-onli...

It answers a lot of the questions being asked. And you can watch Crysis playing through the service.

Now, they might be lying. Or they might be hopeless optimists. But if they go to beta and it doesn't work, the whole thing's a disaster - and they don't seem stupid enough to do that.

@wush: Speed of light in fibre (not in vacuum) is 2 x 10^8 m/s, so 2 x 10^5 km/s. So... 200km/ms. Yeah, I was way off, sorry!

@ukslim: Well, I'll be here with my hat, and the BBQ sauce, waiting for that beta :)

@ukslim - "I think lots of commenters haven't seen the GDC presentation:
http://www.vg247.com/2009/03/25/gdc-onli...

It answers a lot of the questions being asked. And you can watch Crysis playing through the service."

That's in a controlled environment though for demonstration purposes and is hardly conclusive proof that it works in the real world where there may be ten of thousands, possibly millions of users.

If it works, then great; if it doesn't, then that's too bad. These technical considerations go way above my head and I don't pretend to know anything about them.

Aside from the question of whether it will work or not, another interesting question to raise is one over ownership. Will people be happy just renting games from a service rather than owning the disk, the manual and the case?

@Baranga

You might be a 'tard, but some of us study and work with these technologies daily.

We're sceptical due to the laws of physics, and the limitations of real technology as it exists in the real world in 2009.

I expect that you're one of those Sonytards still waiting for teh c3ll and blu rayz to unlock 4d 120fps 32:9 graphics.

Tard alert.

Technologies such as these will be fine for games that don't depend on an immediate response. It'd be fine to play Uno or RPG & RTS games here.

Racing games and FPSs though? I need convincing. That said - I'd love for my home Internet to be capable of supporting this in full quality with the lowest latency. Once we're at that point I'll expect several other excellent services to be rolled out off the back of the streaming tech.

@malixu @darren of course it's a controlled environment. You don't take risks on a presentation like that. But the demo's impressive anyway (as Leadbetter says), and the intro and the Q&A refute some of the points people have made here.

Bear in mind this summer's beta is for Americans only. Us Eurogamers will have to go by what our American friends tell us.

I can't tell how likely this is to be true, because they aren't telling us what the technology powering it actually is.

So fair enough for some scepticism but to claim it won't work by citing specific technology that wasn't designed to do this and saying this existing technology can't do it therefore OnLive won't work is daft.

I mean really - they'd have to have a high spec gaming PC per connection - it's a daft thought, for all the reasons people are saying. So why assume that is how it works?

I thought there was a couple of interesting things from the interview linked above - the resolution scales with connection speed, you can instantly save the last 15 seconds of play to your clips, and that any player can be spectated in real time - spoke of a million spectators. That just sounds like a different way of transmitting game visuals than I can comprehend.

Now I am fully aware that this kind of thing is rather unprecedented and it may be VC securing guff, but the list of publishers signed up was pretty interesting.

My take on it is if it will work as they are talking about then they are doing something fundamentally different or new from the techniques talked about in Richard's article. Otherwise I'd agree with the article but I think it was written too soon and without the facts.

I was very, very excited when I first heard about this, my thoughts at the moment in terms of video quality...

PC vs 360
PS3 vs Onlive

;)

FFS, you're implying it's a "scam" Leadbetter? Seriously?

@Jos

If it is anything other than effectively a large cluster of uber-PCs with good graphics cards then it becomes another platform to port to. Seeing as the majority of software houses already make a shit job of porting stuff to the PS3 or whatever from their lead platform, what gives you any confidence they could add another to the mix?

I agree that would not be a good thing. I just think if it is going to work it would have to be done differently.

They say it works and so I am waiting to hear how, assuming there is some missing information that squares this circle.

My issue is that the article has jumped the gun in assuming they are claiming to be able to use standard tech in a way that is impossible.

@jos: they're actually telling us rather a lot. Watch the video. For high end games, it really is one server per player. The servers are very specialised - with GPUs and custom video encoding hardware. The software isn't just PC code dropped onto the servers: developers port their game to the system.

You know having thought about the massive problem of input latency I do actually see a market for this with Wii owners as they are used to sloppy controls in their games so maybe this is who they are aiming at?

@ Ryze - 32:9 is kinda impractical, don't you think?

I think you should have some common sense. Again, why would they create something like this if it doesn't work? Do you really think they fooled all those companies? Do you really think this is a scam, while the world is watching them? I'm still waiting for a response to those questions. All I see is that you don't even consider they admited limitations and answered some of the points raised here, as posted above with the East Coasr vs West Coast deathmatch. There was a post that seemed to imply they'll only have one server farm. You ignore even the article's ideas as why and how it could work.

And you also ignore the news about the competition. What, is this a brain cancer that only hurts some developers and companies and makes them think OnLive's concept can be applied in reality?

Give me a break.
If the beta fails, good riddance, it was fun while this story lasted. If it's a succes... They'll be sued by every hardware company for every dumb thing they can think about.

Malixu: You seem to be mising my point, and back it up with a series of bad examples of how technology evolves slowly. Except - some of the time it moves in fits and starts.

"The transistor may have enabled computers, but was designed in 1947, not a couple of days ago."

So? Personal Computers were a standing joke, crazed enthusiasts only, and then Windows 3.1 happened and PC's were suddenly everywhere.

" The Internet is rolling in for 40 years ago. Even the web is over a decade (first formal spec is dated 1996, first browser 1993, and there was Gopher before that)"

Again, in the space of just a few months in 1994 the Internet gained it's momentum, rising from a few thousand homepages to millions, and professional, commercial services too. WIthin a few years, people were suddenly doing their banking from home, getting dates on the internet and outsourcing callcenters to India. Huge changes in a very short timespan, both technlologically and socially.

"Space flight is as close as you get, and that was because two superpowers through everything they had at the problem, not a couple of dozen guys no-one has heard of before."

OK, how about mobile phones then? How long did it take from the fist truly portable and reasonably affordable mobile phone appeard 'till everybody had one? Less than three years. In my book, that qualifies as a revolution.

Broadband bandwidth and processing bandwidth keep on developing, and once in a while we hit milestones that enable brand new services and products. Like call centers in India, or real-time video editing.

I think most people focus too much on one single ting, and that's 60 frames per second. Most likely it will be 30 FPS, locked. That will be fine - for the first generation of this tech, but the client side will be able to handle more FPS even if the server side solutions and the bandwidth wont allow it. It will still be very much playable.

Also, the image of each game running on a remote desktop computer with keyboard, monitor and mouse attached is just silly.

Light travels at 299,792,458 metres/second. That's 186 miles/millisecond. Living at the 1000 mile suggested limit from the data center it would take light just over 10 milliseconds to do the round trip. This is the minimum possible latency before you add in the encoding and decoding of the video and any other delays imposed by the medium.

On a console game we sample the input every one or two frames. The extreme case here is sampling every frame at 60fps, which is every 16 milliseconds. Games that run at 30fps and sample every two frames have 64 millisecond gaps between samples. Your button press is likely to take at least 64 milliseconds to show a result on screen when playing on your own console. We perceive this as instant. 60fps makes motion look smoother than 30fps, but input is far more tolerant of delay. I'm going to take a ballpark guess here and suggest that as long as the time between your button press and the result appearing on screen is less than 100Ms you won't give a monkey's.

I don't know why everyone is making the assumption that they are encoding every image as a bitstream. I think they must be encoding and sending just compressed textures and the updated polygon meshes (compressed aswell). And have the micro-console being some kind of 3D card. That could have several benefits (for example, as long as you get the new camera position, you could update and render a new frame).