OTOY In Action

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
Remember that announcement from AMD and OTOY back in January (here) about AMD's Fusion Render Cloud? Well, you can see the tech in action in this video. Pretty damn impressive if you ask me.
 
Last edited:
So he gets to play gta at work? I have no sound so I could not heard what he been saying.
 
I hope that someone will take the video, remove the audio and start a rumour saying that GTA IV is coming to OSX :p
 
Sorry if I don't have this right, I'm not clued up on how the picture transmission will work, but...

To play a game at 1024x768 with 32bit colour, you're talking...

786,432 pixels multiplied by 32bit colour ... 25,165,824 bits per frame ... 8 bits per byte ... 3,145,728 bytes... exactly 3MBytes per frame...

You want 30 frames per second for minimum smooth gameplay ... so you need an internet connection which can handle a down-speed of 90MBytes per second ... basically you need gigabit internet. Nice to know he can play GTA4 on his work-LAN...

Of course compression could lower this amount, but that adds on lag time, which isn't practical in games ... great :(

Please enlighten me if I have this wrong, but I can't see how it could be realistically practical.
 
I am betting that they are streaming down a much smaller image that is scaled up locally by a flash engine of some sort. Still it is bandwidth intensive because he said basically his cable modem at home was good but the connection via wifi was not as good as it would need to be to play somewhat smoothly.

I just want the URL for work! ;)
 
I'm wondering what the back end looks like to this and how much they plan on charging? Are they rendering the graphics with GPUs or CPUs? I could see CPUs scaling much better and costing less on the backend... Very curios as to how scalable this is (i.e. how many people they plan to support simultaneously) and if this is actually going to be profitable for them given the hardware / bandwidth requirements for the backend.
 
There's gotta be a big server farm somewhere with a lot of GTX 285s Sli configs or something to be able to render games for multiple people. There's no way CPUs on a server are able to render games like that then send them out to multiple clients without some beefy GPUs to help them. This will definitely have good potential if they can keep the connections reliable and it's not too expensive. It seems like the same thing as OnLive though.. are these going to be competing?
 
This is wonderful, so you need a 10mbit downstream connection to the internet to get fluid video at 1920x1200. Don't forget your ISP is gonna knock your ass off when that bandwidth cap gets hit. ISP's are going to love this type of thing as an excuse to sell overpriced bandwidth.
 
To play a game at 1024x768 with 32bit colour, you're talking...

786,432 pixels multiplied by 32bit colour ... 25,165,824 bits per frame ... 8 bits per byte ... 3,145,728 bytes... exactly 3MBytes per frame...

You want 30 frames per second for minimum smooth gameplay ... so you need an internet connection which can handle a down-speed of 90MBytes per second ... basically you need gigabit internet. Nice to know he can play GTA4 on his work-LAN

Video is compressed/decompressed and has massive savings bandwidth wise, for instance blurays which are 1920x1080, only require on average 30Mbit/3.75Megabyte a second to deliver their framerate
 
Its not new, there are a couple other companies out there doing it with Mpeg4-avc. The bigger issue is the ability to roll it out. Think about this. say a 4 megabit payload video (which would give you 720p type resolution) at a delta of 4-6 frames (33 ms per frame / 30 fps) would be 130-200ms of lag for video compression plus send time (say 40-60ms) and some response send time for your actions to go up (40-60ms) and we're looking at 210-320ms of lag.. not that bad, but remember... thats a best case scenario if they are hooking up to your provider's head end with no bottlenecking. And of course you would be looking at UDP unless they were guaranteed great quality of service to prevent packet splitting so TCP wouldnt have to wait to reassemble.

Now for the real numbers. say you use comcast as I'll explain how DSL would be harder (like most internet people here in the US). And say 40 people out of your entire community's coverage area for that particular node are playing a game (because there will be more than 40 people hooked up to the converging pipe surfing the web, watching porn / whatever). So those 40 people are eating 4 megs each, so thats 160 megabit + everyone else's transfer converging. Now lets start adding up some of these neighborhoods, and we start getting into the tens of hundreds of gigabits per second sustained in a decently populated city (say 10% of a 1 million population city like to play games?). So thats 400 gigabit for 100k people. Pretty sure we're not going to be streaming this video across the net, its going to take some private peering action especially with the response time, and the other say 25% of the population using the internet (since i think we're at what 30-35% penetration now?). You're talking about saturating 30% of the overall bandwidth with a constant data stream above and beyond what it is now.

Think comcast was pissed about P2P, wait until they have to upgrade their infrastructure to handle this onslaught of juice.

So why did i leave DSL out? well.. unless they plan to pull gigabit lines to the DSLAM co-los, they wont be able to sustain that level of service. same environment with the distance from each copper head would need 160 megabit minimum PLUS whatever is needed to allow the other people (not part of the 40 game players at 4 megabit), to actually be able to surf the web.

From there you can figure out what the costs would be to do this. And it would be pretty apparent that a very aggressive partnering business plan would be needed with some serious umph from a couple 900 pound gorillas to make it happen.

That said however, in say Korea where its all you can eat at breakneck speeds with top notch penetration rate, this would be the prime way to go.

But in the US, it would appear to be more of a cost and infrastructure issue for deployment and operation.

On the tech side, it could be done with 2 frames of encoding in UDP to reduce overhead as well as a form of RS overhead correction. the ping time up and back cant really change too much unfortunately because you are at the mercy of the network and the traffic running at the time. In the end for it to be profitable, it needs to scale big, and that has a pretty healthy overhead cost. Not from lighting up dark fiber, but getting it into the localized community (which has been the achilles heel of the video on demand industry for awhile).

But thats just my 2 cents.
 
Cool! So with this, you can actually play games...on a Mac! :p

It will certainly require a fast connection, but I don't see how something like a standard 10Mb line would have any problems. That can already be used to stream relatively high quality video.

I'm more interested in the CPU, RAM and GPU requirements of the servers if such a service was to be made available to the general public... Imagine trying to stream Crysis to a couple of million players at the same time... :eek:
 
If/when this gets released to the public I hope they have distributed render farms and not just one big one, otherwise I'm pretty sure the IR satellites could see the heat signature from space :)
 
If this tech actually picks up beyond "browser based" gaming... In 10 years I could see this being the way that PC gamers get their games. Wow.
 
Love the idea, the implementation is going to be horrid for the first go around.

So the question at that point, who wants to absorb the first failure of this platform in order to make the 2nd round potentially viable?

AMD? Nvidia? Neither will sell hardware in this instance, so at what point does this all rely on game developers to foot the costs, and at what benefit to them....less piracy is an obvious benefit. But look @ Steam and Blizzard, both are virtually hackless, and are doing just fine (I'm sure there's flaws here and there, nothing is 100%).

I just don't see the benefit on the surface right away. Cool concept, yes, great idea, yes, potential to make more money that is currently made...I just don't see it. Bandwidth is being trivialized in terms of costs compared to what it used to be, but its still a cost that has to be absorbed, and with cable providers fighting and wanting to cap bandwidth, I don't know how this type of system is going to work with that, unless you pay your provider more per month, and then you have the issue of what benefit do YOU get in the long run from that...

The list of issues they have to overcome are not all technical either, you still have to convince someone that its worth it. Look how long it took Steam to take hold and be a viable platform for delivery....3 years....5 years....? Remember "Steam 1.0" if you do, you know what I mean by it all :)
 
This is an awesome step in the right direction. Streaming Crysis 2 on my eee pc will surely turn heads.
 
As far as streaming video, I'm sure some sort of quality option will be available to compress the video as best possible...but another thing is that when I was playing COD4 (for about 2 hours) the game's bandwidth to communicate with the server was 55mb. Imagine have a few people on a lan playing a game on OTOY, by then it wouldn't be long till an ISP's bandwidth cap will be reached.

Though when this is released to the public, I guess there would be that few platforms that game devs would have to work on. But were probably like 3 or 4 years away for a Crysis like game to be played via cloud computing.
 
ITs amazing to see how ideas get reinvented. Think about it. When MS came out to do Business with IBM the PC was a meager concept. It was all about mainframes and terminals. Gates had the vision of a Microcomputer on every home. But who would have thought that we will be going to basically the same idea: client and server. In this case, instead of a mainframe, The cloud.
Ummm.
 
Imagine trying to stream Crysis to a couple of million players at the same time... :eek:

Exactly what I'm thinking. They'll have to invest on an insane amount of servers, one's with very high-end video capabilities, at that. Also, say that they do get the crazy servers somehow - what about ISP's? It will be one thing load-wise for the servers running the games, but imagine the impact it will have on ISP bandwidth when tons of their users are constantly streaming huge games. I can't see this whole thing being a great idea. There are just so many factors to look at.
 
Video is compressed/decompressed and has massive savings bandwidth wise, for instance blurays which are 1920x1080, only require on average 30Mbit/3.75Megabyte a second to deliver their framerate

Oh good! Just what I need! Lag AND compression artifacts. {Thanks for the offer, but I'll pass}
 
The internet infrastructure in the US is so far behind that this would only work for about 0.05% of the population... So unless these people are going to pull fiber to every client I don't see this working to well.
 
This will work fine in the future. Probably not their intent to have it ready to launch this year or anything.

one word: Lag

Which can be improved (never fully eliminated), with enough investment into technology.

If you saw the video, there was very littel slow down. GTA 4 looked quite playable on that unit.
 
This will work fine in the future. Probably not their intent to have it ready to launch this year or anything.



Which can be improved (never fully eliminated), with enough investment into technology.

If you saw the video, there was very littel slow down. GTA 4 looked quite playable on that unit.

Umm.. lets do some math. First lets start by saying 100ms + ping is unplayable.

I press key, key goes to cloud, cloud sends response displayed on my screen @ 50 ms if you get a decent connection to the could. But wait! there's more! Cloud sends key to multiplayer game server at another 50ms. This brings you to 100ms if either those connections average 50ms (more likely you'll get 2 60 something connections for a 120-130ms lag).

GTA4 was in single player mode, furthermore he even talked about the lag on thier demo unit. Even more so, looking at the size of the window he was using, they are already ahving bandwidth problems while doing it on a lan.
 
Back
Top