Digital Foundry vs. OnLive UK • Page 2

So then: can it possibly work?

Video Quality

When it comes to image quality, the raw mathematics speak for themselves. OnLive operates at 5mbps, which translates into 640K per second in terms of throughput. Bearing in mind that the system targets a 60FPS update, that means that on average there's just 10.67K per frame available for image quality. And that doesn't even factor in the fact that we also need to include audio into the equation.

To achieve the quality it does, it's believed that OnLive utilises the same h.264 video compression that is pretty much the standard with the vast majority of internet video, albeit with a tweak to the usual GOP (group of pictures) format that sees the implementation of periodic intra refresh. Typical internet video uses I-frames, or intra frames, to store reference information with subsequent frames based on differences to the reference frame before a new one comes along. Periodic intra refresh sees the image cut into rectangles and reference data provided on a "per rectangle" basis, updating much more frequently.

It works, but the bottom line is that it often isn't pretty. On static scenes, OnLive works well enough and is comparable to an offline experience, but introduce any kind of motion and the system has real issues retaining quality. The more motion there is, the bigger the differences from the reference data and more image integrity is impacted. The problem is that generally speaking video games have a lot of motion, and a lot of bright, constantly shifting colours.

"We compress different types of games differently. Literally, when you go and play Borderlands, it's using a different compression algorithm than if you're playing Lego Harry Potter. The different algorithms favour different things," says OnLive front-man Steve Perlman.

"We wish there were a silver bullet: one approach that would solve all problems, all games, all ISPs and all internet impairments. But we weren't able to find that. If we see a game that does not perform well, then we work on it. We go and tune up the algorithm."

All of which brings up an interesting point: has OnLive video quality improved over the year since we carried out our first analysis? Assassin's Creed II was an interesting example of how variable image quality could be, ranging from almost pristine to farcical. Well, here's a quick comparison video, comparing ACII on US servers from our original captures, up against some fresh new grabs from the same game running on our OnLive UK micro-console.

Assassin's Creed II compared on OnLive's US launch and earlier this week on the UK service. All our image quality comparison videos are hand-encoded and are running at 25 per cent speed in order to retain as much quality as possible in a streaming environment.

Aside from gamma differences, perhaps down to the lack of full RGB support on the micro-console, the level of compression looks almost identical. Still as good, or as poor, as ever it was with no signs of any extra "tuning" in the 14 months since we first saw the game.

And the real question of course is that if you have just 10.67K on average per frame, just how can you "tune" in better quality? In essence it becomes an exercise in attempting to fool the human eye and OnLive's solution to this is to introduce copious amounts of blur to smooth off the macro-blocking.

We don't know whether OnLive's claims about tuning by game have any kind of merit, but let's give the firm the benefit of the doubt. The point is that even if there are bespoke encoding profiles per game, there isn't enough bandwidth to sustain reasonable image quality at 60 frames per second. Two examples from our testing spring to mind. Let's take a look first at DiRT 3, pitting the OnLive version up against the Xbox 360 game.

In order to sustain image quality as much as possible within a streaming video environment, we've dropped the speed of playback to 25 per cent, and hand-encoded each video.

DiRT 3 running via OnLive, up against the Xbox 360 version. You'll note the increased frame-rate the Cloud service offers, but the impact to image integrity is very significant.

Colours are fairly muted in DiRT 3 (good for compression, generally speaking) and you'll find that the pre-game level reveals generally look fairly good. However, game performance is approaching the 60FPS level and with so much detail moving past at such a high speed, the unfortunate reality is that OnLive has to dial up the blur to very high levels in order to mask the very real lack of the detail in the scene the player is receiving.

Probably the nadir of OnLive visual quality we've seen so far is THQ's Warhammer 40,000: Space Marine. It's a detail-rich game with a fair amount of intricate content and likes to surprise the player with nice big splats of blood, flashes of lightning and huge explosions at any given point. There's so many highly detailed objects moving around on-screen that OnLive just can't cope. In no way is this anything you could describe as a high-definition experience.

It's long been our contention that video encoding can't cope with all gaming eventualities - as you'll see here, basic Space Marine gameplay manages to turn image quality into a mush of blur and macroblocking.

Of course, games with little motion and muted colour schemes will work out fine with OnLive. While it's not the most optimal example, the look of Deus Ex: Human Revolution wasn't bad at all. Yes there are compression artifacts, yes there's a fair bit of blur, but check out the look of the game in the RPG sections, or in the initial engine-driven "walkthrough" with Megan. In a typical living room environment, with the player set back from the HDTV, it'll generally look OK.

Criticisms of image quality here move beyond the issues of compression artifacts and into the basic rendering setup. Curiously, there is no anti-aliasing whatsoever - something we seem to see in a great many OnLive titles. It's all the more curious in that the post-process AA solutions employed with the game - FXAA in particular - would probably work rather well in concert with the video encoding process. Picture quality is good enough to discern exactly where the developers have pared back the visuals in order to run on OnLive's target PC platform, but more on that later.

Deux Ex: Human Revolution - OnLive at 720p vs. PC at absolute max settings, operating at the same native resolution. Given neutral colour schemes and not a huge amount of motion, OnLive image quality can hold up well in 'living room' conditions.

OnLive likes to talk about its state-of-the-art servers, but the overall impression we get is that there's a fixed platform running on fairly low-level graphics cards (something along the lines of a 9800GT or 9800GTX remains our best guess based on performance analysis) in concert with a mid-range dual-core CPU. Titles aren't running at anything like max settings (think: low/medium) but the question is whether increased GPU power would actually amount to any real difference if intricate detail is being lost in a sea of compression artifacts any way.

Our gut feeling is that OnLive will improve its base PC specs, but will only do so when it feels ready to migrate on to the next hardware platform. In the recent Eurogamer interview, Steve Perlman talks about a forthcoming 1080p60 upgrade requiring 10mbps of bandwidth - what we'd really like to see is the option to retain 720p resolution but to use the higher bandwidth level. Realistically, short of a major rethink of its bespoke hardware encoder, OnLive image quality will only be improved by addressing the issue with a very significant boost in the video bitrate.

Comments (77)

Comments for this article are now closed, but please feel free to continue chatting on the forum!