While you can't do client side prediction when getting a video stream from the server what about doing server side prediction instead?
I.e instead of server sending you frames rendered from you action 100 ms in the past, it could predict "the world" 100 ms from now, render and send that world.
I guess this would have the same artifact as client side prediction (but you could of course smooth things out better if everything is rendered on the client).

Other positive side effect with server based rendering is that it eliminates all cheats that is possible by modifying rendering code on the clients.

I still think that the server costs are too high atm to make it a viably option, but "cheap" gpu based solutions such as AMD's proposed solution might make it work.

I.e if an average gamer are willing to spend $100 a year on these games, then the server cost (hardware, staff, internet, power etc) for serving a single player may not exceed $500 a year in order to make some profit (assuming that 6 players can "share" the same resources by only playing an average of 4 hours a day).
I don't think it's feasible today, unless the graphics rendered is very simple (but then what's the point?), on the other hand a really powerful video compressor is needed to produce good low bandwidth video and that will probably eat a lot of power (CPU, GPU.. whatever) per player.

Well consider the market, retargetable high quality interactive video streams can be served up to mobile handheld, netbooks, laptops, desktops, consoles, through web browsers. I think if they think outside the box they have plenty of customers. The utilization rate will probably be something more like 20 : 1 per cpu. Most of them won't be playing Crysis but perhaps playing WoW on their netbook, or Fallout in the browser, or some XBLA game on the PC.

As for quality they can always reduce the resolution dynamically to scale with demand / performance for the appropriate platform. They control the GPU completely so they can basically do anything with the rendering.

If this catches on I'm sure developers will support OnLive directly and have latency compensation technology built right into their game with regards to remote input.

-ddn

Chalk me up as another person who will believe it when I see it. Around 2000 the big thing in computing was distributed computing, and in many ways, it was a huge success. Instead of having (arbitrary numbers) 1 server that runs 20 teraflop, you have 10 smaller servers running at 2 teraflop, which would be much cheaper. It is pretty much that way through most of computing. It is cheaper to have 2 decent video cards running in sync rather than 1 super video card. And now we are going the other way? It makes sense on one level, since almost no one's computer is running at 100% CPU usage 24 hours a day, and it is actually a great idea for business and productivity software, where you are only processing data periodically, and the majority of the time, you are staring at static output. But most modern games use computers nearly at capacity, and constantly while they are being run. Add that on to the fact that you need large computers to do the work of many small computers (which means steadily increasing cost per flop), and I just don't see where the benefit of this is.

I am assuming this is another hyped trend in computing that will come out, live for a bit and then die a cold death. Just because we can do something doesn't mean we should. Kind of like ordering your groceries on the internet around 1998 :-P It just didn't end up how they thought it would.

I just don't think this will do it for me as long as they can't deliver really hq graphics and sound. Just like I don't enjoy 64kbps MP3s or crappy pirated movies I don't want to play games with low visual quality... and pay for that.

But I don't see the "classical" way of playing computer games harmed, so... who cares. Can't wait to see this service in action under real-world conditions, actually .

Quote:Original post by Madhed
I just don't think this will do it for me as long as they can't deliver really hq graphics and sound. Just like I don't enjoy 64kbps MP3s or crappy pirated movies I don't want to play games with low visual quality... and pay for that.

You are not primary audience.

People like to point that WoW is a huge success. Yet there are FaceBook games which have tens of millions of users. Let alone outside of there. I think Audition broken 50 million years ago in China alone.

Crysis-like ultra resolution is a tiny tiny niche. At least as far as publishers go, and they need to worry about money.

Second is the demographic that console makers are targeting. A bunch of friends, hanging out on weekend night, instead of going to the movies, they rent a game. To play once, for two hours, and never again. On their TV. No download, no installation, no setup, just click on the icon and you are in. Imagine a service like this for $15 a month. What is the price of a movie + popcorn + driving there?

Obviously, playing WoW this way is somewhat pointless. But that is not primary target demographic.

Personally, I'd be very happy if I had the ability to play latest hits, when they launch (not wait X weeks till they are made available for whichever regional marketing reason), without the need to buy exclusive console it runs on, without the hassles of installing on PC (console ports are a disaster in this respect), etc....

Because incredibly good chances are that Game - The Sequel Part 4 IX will simply not hold my interest for more than an hour, and paying 60-something for that is too much.

Most people do not get all achievements, most don't even finish the campaign or equivalent, or replay on different difficulty, or do much at all.

As for quality - youtube (and others) didn't succeed because it offered 1080p Dolby surround experience. I mean, 400x300 video was said to be a dead end as well in the 90s, caricatured everywhere as tiny window in a huge screen. Yet apparently, it's not about fidelity alone. Same story with Blu-ray and similar. The only advantage DVDs offered over VHS was not video quality - it was removal of rewinding.

And historically, accessibility trumped technical quality just about everywhere.

Quote:Original post by ZeRaW
Correct me if I am wrong but 100-200ms on a network game is acceptable if the client is using some kind of client side prediction, but in this case there is none,the client has to remain still for 200ms and wait for next frame from server.

I think you've hit upon the major problem here...there's just no way to hide the latency when you're not rendering locally.

Quote:Original post by direwulf

I think you've hit upon the major problem here...there's just no way to hide the latency when you're not rendering locally.

Although, this article says that local rendering can have as much as 200ms input latency, and basically never less than 67ms. And apparently, people just don't care.

I think people do notice and probably prefer games with lower latency but won't mind something even as high as 200 ms latency. The lag breakdown at the start of this post only comes out to something like 150 ms latency for a 100 ms ping connection to the server, which I think is well within reason.

Recently Intel has invested in another startup in this field (

http://www.gamasutra.com/php-bin/news_index.php?story=25401

). It saids something when there are already 3 well known and funded competitors for a unproven market. Some people with money believe in this idea enough to invest millions of dollars to make it happen.

-ddn