While no one is ever going to claim that the OnLive experience is as good as glorious can’t-hear-yourself-think-over-the-GPU-fan desktop and console gaming, for many gamers it is good enough. For hardcore gamers, OnLive’s 150-250ms latency and 720p resolution is akin to gouging your eyes out, but for almost everyone else — i.e. most console gamers, and almost every social gamer — it is just fine. If you haven’t given cloud gaming a try, visit OnLive and sign up for a free trial; it still blows my mind that it takes just a split second to send your mouse movements to an Onlive data center, compute the next frame, render it, encode it, and send it back to you.
Bear in mind, too, that OnLive games are competitively priced — and because this is all cloud-based, you don’t have to download 15GB of data before you can play (a huge boon for casual gamers), and you can play OnLive (and access your game saves) from any internet-connected computer. If all that wasn’t enough, because OnLive is ultimately a software-as-a-service (SaaS) provider, it also has two more awesome offerings: For many games you can buy a 3- or 5-day pass for just a few dollars (which is often long enough to finish a game, if you’re a hardcore gamer); and for $10/month, you get unlimited access to some 200 games.
In short, if the latency could be reduced and the resolution could be pumped up to 1080p, OnLive — cloud gaming — would be almost perfect. Which leads us neatly onto Nvidia’s new Kepler GPU, which, according to Jen-Hsun Huang at the GPU Technology Conference yesterday, contains new virtualization and encoder tech that makes it the ideal building block for cloud gaming installations. Nvidia has taken its new Kepler GPUs and built it into what it calls the GeForce Grid, which as far as I can tell (details are scarce) is a complete cloud gaming solution. Gaikai (a cloud gaming company similar to OnLive) and Citrix are launch partners.
Currently,
according to Nvidia, in OnLive and Gaikai data centers, one server is
required for each gamer (or “game stream”). Each server has a single CPU
and a single GPU and consumes around 150 watts — i.e. current cloud
gaming setups aren’t efficient in terms of space or power. With GeForce
Grid, however, Nvidia squeezes four GPUs into each server, giving each
server the juice to accommodate up to four users, at just 75 watts per
stream. Nvidia hasn’t stated which GPUs will be used in this new cloud
gaming system, but the hardware specs on the Grid website suggest that they’re GTX 690s with extra RAM, or simply re-branded K10s (the upcoming Kepler-powered Tesla GCGPU).
Not
only will the GeForce Grid be more efficient, but Nvidia is also
claiming that its new system will register a huge drop in latency over
existing cloud gaming installations. Take a look at the graph to the
right: Yes, Kepler is so powerful that the GeForce Grid will be a hair
faster than your Xbox + TV. Keep your eye on the Kool-Aid, though: While
Kepler’s hardware H.264 encoder will certainly help things along, the
“Game Pipeline” and “Network” improvements are peppered with caveats.For a start, the huge reduction in game pipeline processing will only be realized if software developers actually design their games “specifically for the awesome graphics processing power that these cloud machines deliver.” As it stands, game developers really don’t care if an engine takes 5ms to render a frame or 15ms — it’s all the same, on a local console or PC — but when we make the jump to cloud gaming, every millisecond counts. In a day and age where games are riddled with small, latency-inducing bugs, Nvidia is too optimistic on this front.
Then there’s network performance. Basically, it seems like Nvidia’s solution is to install so many GeForce Grid data centers that network latency never accounts for more than 10ms — either that, or it wants to roll out its own fiber infrastructure. Neither of these things are likely to happen in the near term. Nvidia has the right idea with GeForce Grid, but it’s simply five years too soon.
