Cathode Tan - Games, Media and Geek Stuff
logo design by man bytes blog

Tuesday, March 31, 2009

Dave Perry On OnLive

The good news is that as each user logs off, another can log on, so (using their numbers) if you take the average month has 30.417 days, and you multiply by 24 hours, you get 730 hours of server time per month. They estimate 60 hours per user, per month, so one server covers 12 gamers, with a little bit spare.

So the cost of each server is divided by 12. Let’s say the actual server + GPU cost is $1500 each, using AMD blades for example. They get 36 months to amortize the cost. So they are down to $42 per month, that amount needs to be covered by 12 people. That’s $3.50 per month. With their massive bandwidth discounts added on, they will probably end up around $7-ish per gamer, per month.
-- My "feelings" on Streaming Video Games (via Wonderland)

Dave's co-founder of Gakai, essentially an OnLive competitor. Apparently he got a lot of flak at GDC from the OnLive guys, which seems kinda bad since he comes off as pretty reasonable here.

I'll agree with Dave on a couple of points. Namely that cloud gaming will happen, in some form, at some time. As I said before, there's nothing technically impossible about the concept, most of it not even really being technically difficult - until you factor in the scale of it. So the impressive part isn't one computer logging into one server and playing Mirror's Edge at 60 frames per second ... it's a bit of a neat parlor trick, but it's basically a credit to bandwidth and compression.

The real trick is getting these mythical twelve people per server. Now Dave has done a lot more research on this, since it is apparently his job - I'm just writing based on conjecture and guesswork. But my guess tells me that twelve people on a $1500 piece of hardware seems like pretty curious math.

Server hardware is designed to do some pretty specific things, and it is definitely not allowing twelve people to play games at even 30 fps at the same time. The thing is, I'm not sure what hardware is designed to do that. Remember that once your fancy graphics card is done with all that processing, the data from it has to go somewhere. This is why things like the bus on your motherboard still matter, RAM has to not just be large, but quick, for applications like gaming. An $800 computer with a $300 graphics card can probably reasonably do it well for one person, but most $1500 server blades aren't going to be good at even that task.

The thing people seem to be missing is that games are the most intensive things personal computers can perform. The reason why so many other applications are making successful moves to the cloud is that even the cheapest computers on the planet are about a hundred times more powerful than the apps require. Word processing? I did word processing on a what, an Apple II? My iPhone is what, a gazillion times faster than that? Of course I can do word processing on my phone, or even my browser.

And the truth is - browser still don't even do that very well.

Dave uses this math and comes up with a $7/month subscription figure. That's roughly what I pay for web hosting, so I'm going to have to hit that number with the doubt hammer. I think to be reasonably profitable, we're probably looking at a number about ten times that figure.

I remember being pulled into a meeting about how video on the web was going to be next big thing and we needed to figure out how to be a part of it. That was about a decade ago, and five years later YouTube would launch. But a lot changed in that five years, including high speed adoption, Flash software and compression technology. I'm getting a sense of deja vu from the debate over OnLive.

But as I said before, I'm armchair QB'ing this. So as usual, anyone who wants to correct my logic should feel free with great haste.


Thomas said...

Your reasoning makes sense. But for me, it just raises the question: if the graphical/sound quality isn't going to be top-notch (and it isn't, really), why not just scale the graphics down and do it on the client in the first place? Adobe's done some wild stuff with Alchemy and AIR, or you could create your own plugin/VM. At that point, you could practically pay people to port their games over to your lo-fi platform, and it would cost less than running these incredibly high-performance gaming server farms all over the country.

Josh said...

And of course, there's a very direct historic example for that. YouTube couldn't exist without Flash.

sterno said...

My sense is that the technology is close enough to feasible that this might actually be doable. If it was solid, I'm sure you could get people to comfortably shell out $30/month to play whatever game they want. I mean hell, I'd try it out for a month out of pure curiosity.

The problem I think they are going to face though is that "solid" issue. This is tremendously dependent on network latency and consistency. I mean it's one thing to get a little lag in a networked game. It's an entirely different beast to have the entire game, inputs and all, experience that kind of lag.

What could be really interesting with this though is the potential to spin the concept into different directions. For example, instead of releasing a demo, provide a few free hours with OnLive to try out the real game. Probably a lot cheaper to do that than to put together a separate demo release.

Also, invariably they can do different tiers of service. For example, they could offer a cheaper service that would offer unlimited gaming but only during non-peak hours. They could charge for extra hours spent, etc.

I mean really, if they can get it to work well, I think it could become a license to print money. I'm certainly up for giving it a try.

Josh said...

I wonder if it won't niche itself. Like I could see this as the ultimate demoware platform. I don't have to wait for a download, there's no risk of me stealing the software, I won't be too picky about performance because I'm just trying it out, etc. I could even handle playing a lower res version just to get a feel for the game.

Course, I don't know how you pay for that model, because I'd be willing to spend very little for a subpar trial. Ads maybe? Cut of sales?

sterno said...

How you pay for the demoware model is by charging game publishers for making their games available. With the consolidation of the publishers, they should be able to do this pretty effectively if they can just sign on somebody like EA.

The main advantage to the studios is that they don't have to screw around with all the work it takes to put together a playable demo. They don't have to do separate QA, etc. They can just unleash people on the real game and save some hassle.

Also, consider this. What if they had beta testing on this system. Have a few racks of different configurations and as players hit the servers, they'd end up testing out a variety of combinations. It would give you a much easier to manage environment and you could more readily log the results and find glitches in the system before hassling with packaging and shipping anything.

How many millions do you think the big game companies blow on that kind of work? If they could effectively push a big chunk of that work out to the customer, it would be totally worth it to them. And while yeah, you don't want to have your customers seeing horribly bug ridden software, if you manage expectations and limit access, you could probably get a far better QA than you'd normally get.