GDC: Why OnLive Can't Possibly Work • Page 2

Cloud computing or cloud cuckoo land?

How Did They Do That?

So, bearing in mind that OnLive is demonstrating at GDC, how is it achieving the results? It's difficult to say, but this is how I would do it. Firstly, I'd have a bank of whopper PCs behind the scenes running the games at 720p60. Each of them would be connected to a hardware h264 encoder which would in turn be connected via gigabit LAN to the clients. If the server-side PCs aren't on site, I'd have them at a very close-by datacenter. At the GDC demo, OnLive bosses Mike McGarvey and Steve Perlmen said that the company's servers were hosted 50 miles away. If this was a true test conducted over the internet, I'm betting that there was a whopping internet connection being used with oodles of bandwidth, even if only 5MBps of it was utilised.

Perhaps this suggests an element of smoke and mirrors, but if I were OnLive and about to give a demonstration of this importance, I'd definitely be looking to control as many of the conditions as possible. The main principles are being showcased, but in a best-case scenario. The thing is, actual performance has to live up to this demo and that's where things get tricky.

Factor in thousands more users, orders of magnitude more traffic at the datacenters, and all the vagaries and unreliability of the average internet connection and actual real-life performance must surely be in question. Much as we all want this to be brilliant, the fact of the matter is that even a Skype call over the internet is prone to failing badly at any given point, so the chances are that the far more ambitious OnLive is going to have its fair share of very tangible issues. Picture quality will be immensely variable and lag will remain an issue - but for the less discerning gamer, maybe - just maybe - it will work well enough.

How Could They Make It Work?

So, could this system actually live up to the claims being made for it? What sort of conditions are required to ensure optimal performance? Firstly, I don't think that the video encoding issues will be overcome and I don't buy into this 'interactive video algorithm' geek-speak. On high-action scenes, you're going to be seeing a lot of macroblocking; it's basically inevitable. I can't imagine Burnout ever being streamed in HD to acceptable standards at 60fps without at least two to three times the amount of bandwidth OnLive uses.

I can see 30fps video being the standard here rather than the mooted 60fps. It'll make the video quality look massively superior, and reduce the load on the client decoding it, plus it will help manage latency if the amount of frames being processed is halved. Plus of course there's the fact that 90-95 per cent of console games run at 30fps anyway. It's effectively the standard and it will lower the CPU/GPU requirements of the PCs server-side. But even then, don't think that this will result in lossless HDMI-quality video - far from it. Any game with fast-moving, colourful video is going to look very rough.

That said, the 1.5MBps standard-def option is intriguing and has a much better chance of working out. Here's the same h264 encoding profile I used earlier, reworked for standard definition. I can even run this in real time in the Eurogamer Flash player, no slowdowns or zoom-ins required.

SD 30FPS video

Let's give OnLive the benefit of the doubt for a moment and say that its encoder is better than the very best in compression available today. If its tech is the generational leap that Perlman and company say it is, maybe it could match that quality at 60fps. But still, blown up to full-screen, it's not going to be especially impressive.

Latency. I can only see one way to make this work and guarantee the necessary quality of service, and that's to adopt an IPTV-style model. The OnLive datacenters will be licensed to ISPs, who will have them at their base of operations. Latency will be massively reduced, the connection will be far more stable, plus the datacenters with the PCs and hardware encoders can be distributed worldwide in a more effective manner. ISPs will be cut into the deal the way that retailers are now with conventional game-purchasing.

But even in this scenario, practically, I still can't see it happening. Microsoft's IPTV venture still hasn't materialised anywhere outside of the USA, so what chance does OnLive have of brokering a deal? And with ISPs complaining about the load brought about by innovations like the BBC iPlayer, why would they want to be involved with a hugely congestive venture like OnLive?

And what about computer costs? OnLive is promising state-of-the-art PCs running your game experience. The costs in creating the datacenters are going to be humungous, even factoring in the assistance of a volume manufacturer like Dell or HP. And what happens when GTA or Half-Life comes out and everyone wants to play it simultaneously? Will we have to take turns on connecting to the available servers? Computer costs, bandwidth costs, development costs, publisher royalties... it's all starting to sound hugely, and prohibitively, expensive. Not surprisingly, OnLive is keeping mum about its cost structure to the end-user.

The Alternative

2

Let's say that I'm wrong. It's not completely unknown. I'm just a man (flesh and blood!) taking a pop at visionaries who reckon they have produced something truly epoch-making. But in order to make OnLive perform exactly as claimed right now, the company has to have achieved the following:

  • 1. OnLive has mastered video compression that outstrips the best that current technologies can achieve by a vast margin. In short, it has outsmarted the smartest compressionists in the world, and not only that, it's doing it in real-time.
  • 2. OnLive's unparalleled grasp of psychophysics means that it has all but eliminated the concept of IP lag during its seven years of "stealth development", succeeding where the best minds in the business have only met with limited success.
  • 3. OnLive has developed a range of affordable PC-compatible super-computers and hardware video encoders that are generations beyond anything on the market at the moment.

At some point, Occam's Razor, along with an ounce of basic common sense, has to step in and bring an end to this fantasy, no matter how much we want it to be true. OnLive boss Steve Perlmen remains adamant: "Perceptually, it appears the game is playing locally... what we have is something that is absolutely incredible. You should be sceptical. My first thinking was this shouldn't work, but it does."

So let's put it this way - I can't wait to be proved wrong.

Comments (363)

Comments for this article are now closed, but please feel free to continue chatting on the forum!