You remember OnLive, right? The service, which lets you play any game remotely on a distant server, has produced much skepticism and much interest, and is now in public beta. We got a good look at it back in March when we were at GDC, and it appears that things are much the same. However, the combination of crowd noise and my bad playing made for a less-than-optimal viewing experience. This video is much clearer and much longer (it’s essentially a guest lecture at Columbia), so if you’re still interested in the OnLive thing, it may be for you.
This video deals with some of the technical issues that have been brought up. I haven’t watched the whole thing (skipped around to get the interesting bits) but he does address some of the compression and packet loss issues they have to deal with. I remember being told it’s about 4-5Mb/s for 720p/60FPS, which actually seems a bit low for streaming video, but with a specialized codec and stream they seem to have made it work, even with tricky bits like crisscrossing lines and slow gradients. They have a routing technique that they claim reduces latency as well, but can they really guarantee <20ms pings for everyone using the service? Seems optimistic, but overall pretty convincing.
Here is the "business model" slide:
I notice they leave out a very significant number. They say they’re leasing servers, but I assume that’s for crunching video data and streaming it. They need a whole other set of devices to actually run the games. You want to run a game at 1280×720 and 60FPS? That’s a serious investment in hardware. Even with sophisticated planning algorithms for determining peak times and load sharing, you’re going to need thousands and thousands of machines to keep your service running. If I’m wrong and they’ve really avoided this, then I’ll eat my words gladly. Let’s just ballpark some hardware here:
The GPU will have to be at least mid-range, same with the CPU, or it won’t be able to run the newer games. Extra cooling will probably be done on a large scale, but is too squirrely a number to factor in here. Assuming there’s no case and they’re using onboard audio, then they’re looking at a bare minimum of $700 if they buy smart, probably more like $500 if they buy in bulk. Let’s call it $500.
He talks about running things on CPU only, and virtualizing things across servers, but really, when you’re advertising playing the latest games on release, like Assassin’s Creed 2 and Modern Warfare 2, people aren’t going to choose Tetris. The bulk of games people will want to play are going to use real hardware. You can’t sell a product for one purpose and spec it for another.
If each machine costs $500 and they have to serve 100,000 users, let’s say they need to have a third of those available at any time. That’s $500 x 33,000 = $16.5 million. I don’t see that figuring into their calculations anywhere. And I doubt Intel, AMD, or NVIDIA is likely to pony up that much hardware on credit. Depending on how much they charge for month, it might take users a year to “pay off” the hardware that enables their account. And don’t forget, OnLive will have to upgrade regularly, like us poor PC gamers.
I’m still skeptical of the whole service, or at least its scalability, but the fact that it’s publicly displayed and discussed makes it far more real than, say, the Phantom. I assume they’ll be at CES, and maybe we can put some of these concerns to the man himself.
Update: I just want to note that I am actually amazed by the work they’ve done so far; I’m just concerned about the costs involved with scaling it to the levels they want.