Lessons from the fallen: the obstacles to cloud gaming

OnLive Bankrupt Header


The idea that we are moving towards a fully cloud-based gaming world is not new. In 2009 cloud gaming pioneer OnLive announced to the world that it would revolutionize gaming. The vision was inspiring: the ability to play any game, anywhere, on any device. Consumers would no longer need to fork out hundreds of dollars for a console or constantly upgrade their PC hardware. It was a proposition that posed a serious threat to Sony, Microsoft, Nintendo, and PC brands. In 2011 valuations for OnLive went as high as $1.8 billion. Just a year later OnLive spectacularly crashed and burned, going bankrupt in August 2012 and laying off most of its 150-200 employees. So why did OnLive fail? Was it bad management or is there a fundamental issue with cloud gaming itself? I won’t pick apart where OnLive specifically went wrong, though the fact that they were burning through $5 million a month running 8000 servers for only 1800 simultaneous users (at peak) points to a faulty “build it and they will come” mindset. OnLive CEO Steve Perlman seemed to be far more focused on the final vision (probably due to hyperactive fundraising) than on letting the company grow organically. But putting the specific issues with OnLive aside, what are the obstacles inherent in cloud gaming that all entrants to this industry face?

The biggest hurdle is latency. In gaming latency is basically the time it takes for an input on your game controller to be output on your display. On a console a typical first-person shooter will have a latency of 150-200ms. Any slower than 200ms, the average human response time, and a shooter tends to feel sluggish. So local console games don’t actually leave much latency margin anyways, and most actually go over 200ms. Games such as GTA IV and Killzone 2 for instance were criticized for feeling unresponsive.


Console latency chart


So when you add in network time, encoding time on the server side, and decoding time on the user side it’s easy to see how latency can quickly become an issue. OnLive claimed to have latency that outperformed consoles, but in practice their latency was upwards of 200ms and often reached an unplayable 300ms. Let’s go through what happens when you press the trigger on your controller while playing a streaming game.


Cloud Gaming Latency Chart


First, the trigger signal from your controller is sent to the gaming server through your internet connection. While broadband connections are faster and more stable than ever before, when dealing with such unforgiving latency margins, even a small increase in traffic can seriously degrade game performance. The average internet speed for a US household is about 7mbps, which is just slightly higher than the required 5mbps to play a 720p HD game. However, during peak times this bandwidth can easily dip below the 5mbps minimum and if someone else decides to play a YouTube video on your home network kiss your gaming performance goodbye. So the viability of cloud gaming largely depends on improvements in network infrastructure.

However, some companies are employing strategies to get as much performance as possible from current technology. Gaikai is investing in many small servers spread out across the United States, in contrast to OnLive’s large, centralized servers. This way they can minimize network delay caused by distance between the gamer and the server.


Gaikai Server Map

Gaikai server map shows 12 servers across the US

Game Processing

Once the game client receives your trigger command, the server CPU and GPU need to process the command and generate the next frame. On your console playing a game at 30fps this will typically take about 100ms. However, cloud gaming companies have worked out that if they run a game at 60fps they can cut this processing time in half. It seems counterintuitive, but essentially what they’re doing when running the game at 60fps is forcing the game engine to process each frame twice as fast. It works the CPU and GPU harder but decreases latency.

The downside of this is that 60fps ends up being more data that needs to be encoded and streamed back to the user, which will most likely mean degraded image quality. One possible solution to this is to convert the video feed to 30fps before encoding. However, the most effective solution will be to have games designed for the cloud. With consoles and PCs developers simply haven’t needed to optimize their game engines, but as the cloud becomes more established as a serious gaming platform, they will have an incentive to decrease processing latency as much as possible.


After processing your trigger command and generating the next frame, the game client will then need to capture and compress that frame so that it can be sent back over a 5mbps connection (for HD). This is no mean feat. YouTube encoders for instance still take a long time to process a 30fps HD video. While live streaming video can be done successfully (though usually isn’t), it’s because the video is not actually “live”. With broadcasting there is usually a “profanity delay” of about 7 seconds to allow for on-the-fly censoring. On top of this the online feed will usually have a client buffer to prevent interruptions caused by sudden changes in bandwidth. Even with these built-in delays, live streaming video still tends to lag behind the TV broadcast by about 10 seconds to a full minute. While this delay doesn’t matter too much when watching a video, it would render a game completely unplayable. So with that in mind, it’s really quite impressive that OnLive was able to run playable games at all.

Cloud gaming companies need to either wait for encoding software and hardware to get better or develop it themselves. Fortunately, they may get some help from hardware developers like NVIDIA, who in 2012 announced their GeForce GRID which aims to tackle the latency issue with cutting edge hardware encoding and decoding.


NVIDIA GRID Latency Chart

Business Model

OnLive was not able to provide game performance that rivalled consoles, but charged more for a game than regular retailers…and you didn’t even end up owning the game! You just ended up with the license to play it, and only while your OnLive subscription was active. Gamers did the math and it didn’t make sense to pay more per game for a subpar gaming experience. However, even when the technical challenges are eventually overcome, cloud companies will still face issues with the entire business model.


OnLive PlayPass Purchasing


Companies will need to ensure they have enough capacity for everyone who wants to play. However, as it’s very costly to run a gaming server they need to avoid excess capacity. Managing this balance through a flexible scalable business model is key. OnLive grossly overestimated market demand and the resulting cashflow problems brought them to bankruptcy. However, server advancements are decreasing the risk by bringing down costs. Another benefit of the NVIDIA GeForce GRID for instance is that it can allow 4 gamers to play on one processor. It also consumes less power, generates less heat, and in turn reduces cooling costs.

The OnLive subscription model also caused friction with publishers over the question of who owned the customer. That’s why cloud gaming startup Agawi is positioning itself as a B2B business, providing its services and technology to the publishers. They believe the ideal model is that when you a buy a cloud game you get access to it forever without having to pay a subscription.

Also, to avoid the customer acquisition issues OnLive faced, new entrants are partnering with cable companies and ISPs as they already have a billing relationship with customers. This way customers only need to press a button to activate cloud gaming for their account. The fee will just be added on to their existing cable/internet bill at the end of the month.

Learning from OnLive

Steve Perlman of OnLive

Steve Perlman of OnLive

Steve Perlman saw the opportunity to create a game changing platform (pun intended). Through first mover advantage and significant investment he sought to claim enough market share to establish OnLive as THE gaming platform, in the same way Facebook is THE social platform. Unfortunately, the technology and market weren’t quite ready for OnLive. Perlman probably bet that if he could keep the wheels of OnLive turning through more and more financing, that the economics would eventually swing in their favor. That bet didn’t pan out. However, they did show to the world that cloud gaming was indeed possible, effectively paving the way for a myriad of new cloud gaming startups.

But these startups are proceeding with caution. They speak of scalability and strategic partnerships. Gaikai, OnLive’s biggest rival, started out as a B2B service for publishers to demo their games in the cloud. Now under the wing of Sony Gaikai will be able to test their service in a contained environment with minimal risk (streaming PS3 games to PS4 users). Read more about Gaikai-powered services for the PS4.

Leave a Reply

Your email address will not be published. Required fields are marked *