Double Buffered

A Programmer’s View of Game Design, Development, and Culture

Why OnLive Won’t Work

Posted by Ben Zeigler on April 5, 2009

So the big topic of general interest at last week’s GDC was OnLive. Lots of people, both gamers and developers, were excited about the possibility of playing PC games without a PC. It’s supposed to solve piracy and system requirements issues in one go. Well, I tried it out in person at GDC, and I can say conclusively that OnLive’s plans will never work. But, it’s not for the reason you may think. The technology is viable, it’s just the economics that make no sense.

There was a bit of a line at the OnLive booth, but I waited patiently before I got to play BioShock over it for a few minutes. The experience was actually better than I had expected. The image quality was perfectly fine in my opinion, although it got worse when you moved the camera rapidly. There was noticeable lag, but it was still playable. It felt almost exactly like playing an FPS on too-high settings. Basically it was like playing BioShock on high quality settings at about 15 frames per second. It’s not an experience I would consider paying money for, but I can see how it would be acceptable for slightly less picky gamers. So, OnLive can successfully play games remotely 50 miles away from San Francisco to Santa Clara (where the servers were), using a dedicated line, and to a quality level acceptable to an average user but not to the hardcore. That’s actually a pretty impressive accomplishment.

But it isn’t enough to actually make money. Time to do some math. According to what they’ve said publicly about their business plan, they plan to stream games over the general internet, as long as you are within 1000 miles of a server (1000 miles would be significantly worse lag then I experienced), and have a 1.5 Megabit sustained connection (5 Megabit for HD, which we’ll just ignore because it’s even sillier). Ignoring the issue with ISP bandwidth caps (which could be a serious problem) this is mostly reasonable. However, it leaves out a big part of the equation: the servers and bandwidth on OnLive’s end are not free. In fact, they are very expensive.

Thick-client Western MMOs are a space I am reasonably familiar with. Out of your $15 subscription fee, somewhere around $5 goes directly to hosting and bandwith (the rest towards development, support, and licensing). MMO bandwidth is split between patching (let’s say 1 Gigabyte a month per player on average) and runtime (usually pretty light, 100 kilabit/s MAX). At 50 hours of play a month (which is about average) that ends up being around 3 Gigabytes of bandwith per month. On the other hand, 50 hours a month of 1.5 megabit streaming is 30 Gigabytes per month. So, bandwidth costs for OnLive are 10x what they are for MMO hosting. The other part of the cost of running an MMO is purchasing and powering servers. An average MMO server can host a minimum of 50 concurrent players (a lot more in the case of games like WoW which are light on server use). These are fairly buff machines (but not special in any way), so let’s say that either you could buy 5 cheap desktops for that price, or host 5 client instances of OnLive games, giving a maximum concurrent OnLive users per server at about 5. This means that OnLive will need to buy and run 10x as many computers as an MMO.

The two components of hosting are both about 10X worse for OnLive than they are for MMOs (and that’s a very generous number). Multiplied by the wholesale hosting cost of $5 a month for MMOs, this gives an estimated wholesale cost per user of OnLive to $50/month. Remember, this is wholesale and before they make ANY money off of this. So assuming a minimum margin of around $10/month, anyone who wants to use OnLive is going to be paying at least $60/month for it. This is why they haven’t announced pricing yet, because they know what the reaction would be. So, the audience for OnLive is people willing to pay $60/month for pc gaming, but not hardcore pc gamers who care about latency.

This business plan doesn’t make any sense. On the high end, it fails miserably compared to Steam, which is a lot cheaper and more satisfying for the hardcore. On the casual end, why pay $60/month when you don’t care about graphics and are happy playing browser games? There are only two ways to make this work: either contract directly with ISPs (which would negate the bandwidth costs) and get it subsidized through already-expensive cable bills, or sell the company for lots of money up front. Considering that OnLive was created by Steve Perlman, who is the same guy who sold the largely useless thinclient technology WebTV to Microsoft for $425 million, I think it’s obvious what the corporate strategy is: Get bought before you have to launch the service and hope no one digs too deeply into the realities of the business plan.


8 Responses to “Why OnLive Won’t Work”

  1. Joe Ludwig said

    I’m not convinced OnLine is going to work out either, but I want to poke a couple holes in your argument anyway. šŸ™‚

    The bandwidth and hosting costs for aren’t anywhere near $5.00/month. For a game with 100,000 subscribers they are more like $1.50/month and as the number of subscribers increases the cost per user drops considerably.

    In addition, many people are saying that OnLive’s business model is going to turn out to be providing an upsell opportunity for ISPs. The servers will be hosted in the ISP’s own datacenters which will help with the latency, but most importantly keep the load off the internet itself. ISPs don’t have to pay for the bandwidth between their datacenter and their own customers.

  2. JZig said

    Definitely the $5 could be off depending on the game (Cryptic games tend to be a bit more hardware intensive than some other MMOs), but in that case the concurrent-per-server would also go up in that case, making the relative ratios for hardware and bandwidth OnLive needs to buy significantly higher, so it ends up being just as bad. Feel free to run your own numbers based on what you know from PotBS and such.

    Yeah, selling directly to ISPs is definitely a valid way to do it. But they explicitly stated that their INITIAL goal was to get it working over the general internet, and then only look at the ISP deals later. If their primary goal was to sell to ISPs, I don’t know why they would be mentioning the general internet plan at all, it’s going to be harder to sell it to them if it’s already available somewhere else. But, who knows. Business people are weird.

  3. How many concurrent users does city of heroes currently peak at? 20,000? 40,000? Based on your numbers, I would suspect that city of heroes has considerably less that 1,000 servers, if each supports 50 concurrent users. I imagine that City of Heroes colocates in several locations around the world then in order to better serve its customers.

    OnLive’s plan is very different and much larger in scope: they are building and running their own massive, custom datacenters, which contain tens of thousands of ‘servers’, probably built to scale up closer to a hundred thousand. Google scale. So I don’t you think you can estimate their costs compared to a smaller scale operation that colocates, but you can estimate their costs from first principles.

    For a longer analysis, see this post, but I’ll summarize here.

    Here’s what I know or can gather: Their ‘servers’ are really more like modern high-end gaming rigs, but just CPU+RAM+mobo+GPU. Off the shelf chips, custom rig, fairly cheap. They are employing some fancy new form of virtualization that allows multiple game instances to run on a server, but they still typically need one GPU and 2 intel CPU’s per client to run their games at 60+. And they need to run the games at 60+ to combat latency.

    From this you can work out the hardware cost to be <$400 per ‘virtual’ server, which is 2 modern intel cpus and 1 modern nvidia GPU. Amortize over 2 years and you get <$20 per month, which is split amongst X subscribers. The key number is the typical peak occupancy. I am betting they are shooting for a peak occupancy of 20% – this is probably a gross over-estimate though, as if you look at their games they are mainly single player (MMO’s average roughly 2X user weekly hours compared to regular games). With 20% peak occupancy, 5 subscribers pay for each server, bringing the amort cost down to $4/month for hardware. Then you have power/cooling – with modern datarooms that scales with average occupancy, not peak, and is (by design choice) probably close the amort hardware cost (but see my longer analysis for a more in-depth derivation).

    For bandwidth, I am pretty sure they are getting raw wholesale transit prices in obscene quantities, at the going rate of <10$ per Megabit/S for GigE direct connects. This cost in particular has been dropping rapidly in just the last year, and that I think is the real key ingredient. If you work out the math assuming an average connected user draws <2-3 Mbps, and 20% peak occupancy, and $10 Mbps/month dedicated, it works out to $4 per subscriber per month.

    So thats <$12 per subscriber per month as an upper bound, and actually I think it could be closer to $5 per subscriber/month, because 20% peak occupancy is probably unrealistically high, given the averge gamer plays just 10 hours per week. But the catch is that their business plan is hugely sensitive to the peak occupancy rate, and I am sure that heavily influenced their choice of launch games, and will also influence their pricing.

  4. JZig said

    Your server cost estimate seems really low to me. The cost of a virtual server does not JUST equal the cost of the cpus and gpu, it also must include a portion of the QUITE expensive server motherboards and other components that you need to be able to run virtualized game clients. Oh, and those server-scale GPUs that nvidia sells aren’t cheap either, because business customers will always pay more than consumers for that kind of thing. I think $600-700 is a way better estimate of the server component cost.

    I don’t have any direct experience with the bandwidth and power figures you’re throwing around, but it seems to me that you’ve made several assumptions on how cheap they can get their power and bandwidth, and don’t really source it in your original article. My numbers may end up being high (Joe above seemed to think so as well), but your article explicitly says how you take the lower estimate at every possible point, but claims that it’s the “middle” or “high” choice. I suspect the final answer is somewhere in between ours, but $5/month is just ridiculous

  5. Red Guy said

    I have to upgrade my system at least every 2 years for playing latest games (and I don’t mind it šŸ™‚ ). So far games got better and better but CPU, GPU and memory requirements grew as well. For example on gForce 5950 DOOM 3 at high settings is very sluggish but on 7950 runs smoothly.

    So, will they upgrade their hardware that often to support latest 3D graphics technology? I don’t think so.
    What will be the point of developing new hardware and 3D engines if this is going to be the future of gaming? All 3D technology development for gaming will die.

    Will they provide enough space for game saves? I have 12GB+ for The Witcher alone.

    “Get bought before you have to launch the service” I agree, this is a scam.

  6. […] […]

  7. me and a few people i know tried onlive, but the speed restriction wont allow us to use it, so the only way to be able to play onlive is to pay more for even higher speed internet connection (which the average person where i am can not afford), basically i would be paying $100.00 a month just to play onlive, not worth it, rather just buy the game online and play it on console.

  8. Mitch said

    I just found this article and recommend that you check out Onlive again. It very much works and is incredibly inexpensive. You can now even use it with speeds as slow as 2mbps.

Sorry, the comment form is closed at this time.

%d bloggers like this: