GDC09 interview: OnLive founder Steve Perlman, continued [page 2]

What if you want to trade a game you've purchased? Is there any kind of second-hand functionality built into the system at all?

No, there isn't. The problem with second-hand sales is that what you're sort of doing is you're taking a game when it's new – obviously there's no second-hand version of it – people pay full price and then they play it through and they put it back into retail where it's sold at a lower price. Essentially what you're seeing is the value of that game is lower because it's not as new. Ok, that's fair. But the problem with the second-hand model like that is the money that's paid for that second-hand game at a lower price doesn't go to the publishers. You talk to the publishers they don't have a problem with the fact that as the game gets older the price should go down. But the dollars for that second sale at a lower price if it goes back to the publisher they can go and make better games. Although there won't be second-hand sales there will be tiered pricing as games get older. What you currently are experiencing when you go to GameStop or go to Amazon and get a used game you'll certainly get the benefit of the lower price as things get older. The difference is the money's going to feed back into the ecosystem and you're going to get better games.

Has this been a part of your pitch to publishers? Removing this threat of second-hand retailers? Or companies like GameStop. Their profits are ballooning at the expense of getting that money to publishers.

"Second-hand sales became a big deal during our development and sort of by serendipity, it turns out that this is a good solution for it."

So, here's the interesting thing. When we started on this journey there really wasn't a big second-hand on the market about seven years ago, and our pitch to publishers then was, "Oh we get high-performance games, it's easy to upgrade, and there's no piracy." Which really was an issue back then as it still is today, okay? But what started happening as we'll reveal more in gains when working would be publishers, "Oh, good progress, let's see where you are and everything." They start telling us, "Hey, this thing eliminates, you know, second-hand sales." We didn't develop OnLive seven years ago to address second-hand sales. Second-hand sales became a big deal during our development and sort of by serendipity, it turns out that this is a good solution for it.

Right, I can imagine from a publishing perspective that it's like a golden goose.

They love it. It freaked me out, but there's Reuter's article about a second-hand sales game something in Europe that I didn't know existed that was downgraded because of the announcement of OnLive. Their stock was downgraded to a "sell." I'm like, "This is weird. I've never heard of these guys. They probably have never heard of us, and their stocks are being affected by an announcement at the Game Development Conference? Not E3?." And so, yes, it definitely is having a profound impact on what the perception is of where the future of where second-hand sales would go.

MMOs? The platform will work with MMOs I imagine, seeing as it's a kind of virtualized PC. It wouldn't know the difference, you're tricking it anyway. But MMOs obviously are a big market. They're already subscription-based there.

Yes.

Have you talked to any MMO developers, publishers?

We have.

How did those conversations go on?

Speaker 1: They like it. I mean, if you take a game like World of Warcraft, it doesn't need OnLive. What they've already addressed is a way to make the game work in a low-end PC with no GPU. For example, in China's internet cafe's, where they don't have GPU very often, it works great, okay? So, we can run World of Warcraft on OnLive, I don't know if it'd be useful. Maybe it you want to embed it in the environment, do the spectating or something, but that kind of game – or a flash-based casual game – they weren't great, you know? We're not here to go and say, "You don't need these other things. Those are fine."

Now, MMOs that have very high performance graphics and very complex things going on in the world, they love OnLive. An example of an MMO that has had a tough time being matched with its audience is Second Life. Second Life is kind of a casual game, it's much like The Sims. It's a world to go in, meet people and do things, and so on, but it's got very high-end graphics requirements, so the only people who could really play it have pretty high-end computers, and that kind of limits their audience. Well, a game like that – not necessarily Second Life specifically – is something that we can host in pretty high performance servers in the OnLive Server Center that are hooked up together with each other with gigabit Ethernet. And they have huge disk drives, and the latency from user to user – remember, the users are in the service center – is sub-millisecond, okay?

So, I'll give you a simple example: Suppose I've got a character, totally decked-out with user-generated or purchased content, so I've got all this geometry. I've got all this physics and all this kind of stuff, and suddenly I come into view. In a traditional MMO environment, all of that data – all the textures, geometry, all the procedural stuff – has to download through your DSL or cable connection into your computer, and let's just hope your GPU has enough texture memory, and let's hope you're not sitting there waiting to look at a guy, and he walks away before you actually have the download done to view him. With OnLive, obviously, that goes away. He comes into view, zip: It comes into your server, and your server is known to have the capability to handle it. And in fact, if you're coming into an environment where there's people or situations where you're underpowered, we swap you. We lift you, and we move you to another server. OnLive switches on frame boundaries for the video, and it's so cool. It's a lot of work to get that to happen. When you see the demo you can watch with OnLive from game to game, or user interface or anything, you might've been on twelve, fifteen different servers. You don't know it. With MMOs, the fact that we can swap you, you're always on a high performance enough server, you always have a fast enough link, and you always have a low enough latency.


So what brings me up to my next question is that of the server, right? Each of these servers can be very expensive with the GPUs , so it's a little different than maybe what Google is doing, there they just need a bunch of processors, so the industry is still going to iterate and create new GPUs so what happens to A.) All this old hardware and B.) it's an expensive proposition to continue to upgrade for that bleeding-edge gamer and get enough servers to be able to push Crysis 7, or 8, or 12 in 720p?

So there's a couple things about servers. Nobody looks at servers individually. Google doesn't. Yahoo doesn't. What you look at is the how much revenue the server produces per the number of people that are using it, right? So if you're dealing with a impression-based, an advertising model, CPMs and one server may need to host 10,000 people over the course of a day in order to pay for itself. Because you are getting these tiny fractions of a cent for each time a person goes to the web site hosted by the server and sees an ad impression. The servers that OnLive is hosting is running games for people who paying many dollars per month, comparable to what they're paying right now for software in retail, right?

So the other thing about the servers for OnLive is that if you're using a server playing Crysis and you go offline, that server will be used for somebody else playing Crysis. So we get multiple reuse of the same hardware, so now we have multiple people paying for the use of that server.

And then, not every game is Crysis, you know. We have lower-end games, Lego Batman is a good example. But there are other ones that are for grownups that are also lower performance that don't necessarily need a GPU, or they don't need a very high-end GPU. And we virtualize them so we might have ten simultaneous versions of Lego Batman running on one server. So then other thing that you asked about was what about when these servers get old and we get new ones? Well every six months we will get the latest GPU the latest CPU and so on. But the older servers will they continue to run those games that used to run on them and also will run the lower end versions of the future game. So the Lego Batmans of the future which now require high performance may run on the Crysis server from two years before. Or a year before, so we have constant reuse of hardware.

And it's being determined dynamically on-the-fly? If you are going to do it you may shuffle them off to a lower-powered server?

What we are showing at GDC and when you use the thing, as far as I know it is the first time there's been a virtualization which is literally switching video on frame boundaries on servers so it's imperceptible. I mean, it looks like you are running on one super-duper server when you're in front of that thing but you're not. Most of the time, unless you are smack in the middle of a game, most of the time you are probably sharing a server with many other people because we don't need the level of graphics you need for your game.

How much does this all cost? You're doing this right now without any customers and as far as I can tell, it sounds like an enormously expensive and complicated process.

When we developed the compression algorithm on ridiculously expensive hardware: dual quad core Xeons. At some point we upgraded to Skulltrails. I don't know if you've seen those things. Skulltrails are this limited edition thing that Intel did; a special motherboard that was overclocked and could handle GPUs. We didn't use GPUs because the compression algorithm was just running in software. We needed that to go and refine the algorithm; we needed a completely flexible thing to do the development. We are talking about $8000 per stream, ok? We just distilled that all down to a very, very inexpensive piece of silicone that uses just a couple of watts. And its very, very flexible and programmable with the kinds of things we learned that we need to do with these ultra-high performance machines during the research phrase.

So now the cost of getting a stream out on a wire is de minimis, when you amortize the life of this board we built over three years, it's nothing. Just pennies per month. So that and the hardware itself – remember when I was saying how much people pay per month? If we expect a useful life of a server – say three years – and remember the server that is the coolest high-end server now, three years from now will be running the Lego Batmans of the future. We still have use for them. Or we can sell them. We can sell these servers in other markets that do not demand high-end stuff. So there is an active market for used equipment and so forth.

"And our servers ... are far more profitable than the servers for example that Google is running, or Yahoo's running, in terms of what we make per-server per-month."

So what we look at is, just like any other business that looks across a three-year window for amortization of hardware you go and say, "Ok, how much revenue is coming in per month?" And our servers, when you go look at it, are far more profitable than the servers for example that Google is running, or Yahoo's running, in terms of what we make per-server per-month. So it just pays for itself and then you have other people that are pretty happy. People are pretty happy even in today's market to finance things that will go and produce, that will go and pay for themselves in a few months.

GPUs? You guys are building custom silicon for the compression chips. Are you working with nVidia, ATI, or both?

Yes.

And are they building custom silicon or custom parts for you?

No. For now they're not, I mean we're a startup company and we have a zero million unit installed base. (laughs) So, part of the things I've learned over the years about getting your feet underneath you, you don't demand from big companies that they've got to do something custom for you. The first early versions of OnLive did require, you know, customization of the games to achieve the low latency. When Mike McGarvey came on board from Eidos – he's our COO – and he basically said, "Look, there's no way that they are going to go and design for a new platform that's from a startup. Either you guys can work with existing versions of the games or forget it."

Part of what we do in the custom board is make it work with an existing game. So the same thing with GPUs, we are not going to go and ask nVidia or AMD to design, or ATI to design custom GPU boards for us. At least early days. I mean look, once we do have millions of subscribers we'll talk to them about optimizations. I want to bring the GPU onto the motherboard rather than as a card. And I want, well ... not announcing right now but let's just say there were some applications that could use multiple GPUs to create very highly realistic worlds. You run into the practical issue of fitting all these damn boards there and on a motherboard there's certain kinds of sharing of the data which is far more efficient than going through a PCI slot.

Right.

But we aren't doing that at launch because we don't want to do anything custom. We want to feed off of the existing supply chain of computer components and devices and later on when we get a sufficient volume we can move on.

So two part question here: First, you're showing sixteen games here. Those are all new games. Do you think that there is a market for, and an interest on your part, to have a library of older games, classic games? And as the service gets older, if you're not going to add older games now, in ten years, are you still going to keep that version of Crysis on the service?

So we are not going to launch with any back title games really – I mean maybe this or that if it's not too old. There's a couple reasons for that: One is marketing reason. There's been so many download services and so many things that call themselves game-on-demand services that have these back catalog games. We want to make sure that people recognize that this is top tier games, retail release windows. All the agreements with the publishers are all that if they don't wanna give us their newest games, during the release window, then OnLive's not the service for them.

As these games themselves age, we'll see if the publishers want to keep them around, then it's up to them, really. Generally speaking, you'd imagine that they would, right? Why not, if somebody wants to use the game and they are going to pay something for it you might as well collect the revenue from it. But it's hard to say. There may be situations that they decide that they've upgraded to the next version and want to move over to that. Now you're getting outside of the realm ... it's just pure speculation on my part. That decision will really be made by them. Down the road, once we establish ourselves with an install base and we're not afraid of people thinking we are a back catalog company, will we start having classic games? Maybe, I don't know. They aren't hard to host, they're inexpensive. So if there's a demand for them, why not?

Thanks so much for your time, Steve!

<< BACK COMMENT