Nvidia goes stereoscopic with GeForce 3D Vision

I’m going to assume you have, at some point in your life, tried out some sort of stereoscopic 3D glasses scheme. They come in all forms, from the early blue-and-red specs used in Jaws 3D to the virtual reality goggles with individual LCDs built-in. (Wow, I am old.) If you’ve seen a 3D movie in the theater more recently, you probably wore some simple cardboard glasses. And let’s be honest: all of these glasses have one thing in common: the cheese factor.

Yeah, they’re cheesy. You feel like a fool putting them on, and even if the 3D effect works pretty well, you can’t escape the sensation entirely. As if the people closest to you, the ones you care about most, could see you now—they would point and laugh.

If you’re at all familiar with the history of stereoscopic 3D schemes in computer games, you may have further reservations about these things. In my years covering this industry, I’ve managed to experience quite a few such products. I’ve seen the flicker, and I’ve come away with the vague headaches and sense of vertigo to match.

So when presented with these new GeForce 3D Vision glasses from Nvidia, you can imagine why my initial reaction was skepticism. I mean, what’s really changed between the last time we tried this and now?

A couple of things, it turns out. For one, incremental improvements in technology over the last little while have served to make stereoscopic schemes more viable. With each generation, GPUs have grown by leaps and bounds in terms of visual fidelity and performance. They can generate much more convincing images than they could just a few short years ago. At the same time, LCDs have supplanted CRTs as the display technology of choice. Display makers have adopted wider aspect ratios better suited to the human visual field and, crucially, the very newest LCDs are quick enough to refresh onscreen images at a rate of 120Hz—often enough to display a 60Hz image for each eye in a stereo vision scheme.

The second change is who’s making this latest push for stereoscopic visuals. This isn’t some start-up whose booth is tucked into the back corner of the convention hall at GDC. Nvidia is one of the leading graphics chip makers in the world, and it has strong relationships and substantial clout with game developers. Nvidia is also a mighty hard-headed—err, shall we say, determined?—company. If anybody can make this go, it’s probably them.

Not that the whole dorky 3D glasses thing is any sort of a slam-dunk proposition, of course. But I’ve spent some time playing around with Nvidia’s GeForce 3D Vision, and I’m at least intrigued by its potential.

What it takes
The GeForce 3D Vision package comes with two main components: a pair of glasses and a little, black, pyramid-shaped doohickey that happens to be an IR transmitter. Both of these devices plug into the computer via a USB cable. The glasses plug in only for the purpose of charging. Otherwise, they’re wireless and won’t keep you tethered to your PC, thank goodness. The IR transmitter must remain connected to the PC, of course, because it’s what sends signals to the glasses.

The 3D Vision scheme creates the impression of depth by showing different images to the right and left eyes, with the image for each eye adjusted for perspective. In order to be sure each eye sees a separate image, Nvidia uses a shuttering scheme similar to a number of past attempts at 3D. The display shows two different images in rapid succession, one for the left eye and another for the right. The glasses use active polarization to block the light going to the eye whose image isn’t being shown, and then, simultaneously, the display and glasses alternate to the other eye.

Without glasses, both images are visible—and confusing

This switch happens very rapidly—at up to 120 times per second with this new breed of LCD—and if all goes well, the resulting effect won’t make you blow chunks all over your living room floor. Instead, your eyes should struggle to focus for a second and then—BAM!—you have a splitting headache.

No, not really. At least, probably not. Even if you do, though, you should also have the distinct impression of depth inside of your computer monitor.

Interchangeable nosepieces offer a custom fit for your schnoz

The big, black, plastic glasses are key to making this magic happen, and Nvidia has obviously devoted considerable effort to their design. Despite their thick temples, which presumably house control electronics and a battery, the glasses are relatively lightweight, and the rubber nosepiece grips and rests comfortably on my outsized beak. Since a high proportion of 3D Vision’s potential customer base probably wears glasses, Nvidia has designed its magic shades to fit on your head while you’re already wearing regular glasses. To my great surprise, they seem to have achieved success in this department. At least, the 3D shades fit over and around my smallish, wire-rim glasses with no trouble whatsoever, without squishing them or compressing them into the sides of my head. If you wear gigantic horn rims, though, your mileage may vary.

As you may be able to tell from the pictures, the glasses manage to fit over a regular pair and stay on the face by, essentially, pinching your head. The green, plastic ends of the arms on the glasses are contoured and curve inward. On me, they grip right above the ears, compressing my enormous noggin tightly enough keep the 3D specs secure. Problem is, I have chronic, low-level TMJ syndrome. Pinching my head above the ears will not endear you to me. I really wish the glasses had better padding and a larger surface area where they meet the head. Most folks probably won’t run into the same problem, but you’d definitely want to try these things on before buying them.

Nvidia claims the glasses are good for 40 hours of gaming on a single battery charge. I didn’t quite test those limits, but based on my casual use and neglect for charging, I’d say the glasses do have a considerable run time.

What it takes — continued

The other bit of hardware needed for GeForce 3D Vision is this IR transmitter, which helps synchronize and activate the glasses’ shuttering mechanism. The picture above shows the back side of the transmitter, which houses a rather important control: that wheel adjusts the amount of depth in 3D games. Turning it up provides more stereo separation and greater depth, while dialing it back does the opposite. Although 3D Vision also supports 3D movies and videos, the depth wheel only works in games, because only games generate images on the fly.

The Samsung SyncMaster 2233RZ display. (Source: Nvidia)

In order for any of this to work, you’ll need a display capable of 120Hz refresh rates. In our case, we tried out a pre-production version of a 120Hz LCD, the Samsung 2233RZ. This 22″ display has a native resolution of 1680×1050, with a 16:10 aspect ratio and a 5ms rated response time. The $399 MSRP is mighty pricey for a 22″ monitor. Samsung’s own SyncMaster 2233BW sells for about 220 bucks at online retailers. List and street prices don’t always match up, of course, but you’re still likely to be paying quite a premium for a 120Hz display.

Beyond its 120Hz capability, our pre-production sample of the SyncMaster 2233BZ isn’t anything special, either. To make an entirely unfair comparison, next to the Dell 3007WFP-HC we usually have on our GPU test bench, the 2233BZ has noticeably inferior color reproduction, with visible loss of contrast and slight color shift at acute viewing angles. Perhaps production models will be improved somewhat when they arrive—they’re slated for release in April—but I doubt Samsung will be able to achieve the color reproduction of the best LCDs in combination with 120Hz quickness.

ViewSonic has also announced a 120Hz 22″ LCD with a funny name, the FuHzion VX2265wm. Happily, although its specs are similar to the Samsung, the FuHzion has a bad spelling discount of 50 bucks, bringing its suggested retail price to $349. That’s it for LCDs at present, although there’s promise on the horizon. Nvidia informs us that LG has a 23″ 120Hz panel planned for later this year, and that display will have a 1920×1080 native resolution. That’s more my speed, considering that the 3D Vision kit itself costs $199. Seems to me like this is something of a premium product, and 1680×1050 isn’t really a premium resolution.

If you’re into rather larger displays, 3D Vision is also compatible with a host of 1080p DLP HDTVs from Mitsubishi. And, if you have real money laying around, it’ll also work an HD 3D projector from a company called LightSpeed Design. I have a hunch that puppy will cost you more than, say, a nice Audi.

The final piece of the 3D Vision puzzle is a PC with a suitable GeForce graphics card. Nvidia has a list of compatible GPUs, most of which are at the higher end of the product spectrum. The oldest graphics card on the list is the GeForce 8800 GTX, and the cheapest is the GeForce 9600 GT. Anything newer or more powerful than those cards, including the GTX 200 series, ought to work. SLI is supported, as well, but only in two-way configurations.

The experience
So… does it work? In a word, yep. Nvidia’s decision to limit 3D Vision to displays with very high refresh rates makes this technology easily superior to most past attempts at stereoscopic 3D on the PC. There’s noticeably less flicker, and the illusion of depth works better as a result.

The biggest catch, at present, is spotty game compatibility. Most games aren’t designed with stereoscopic 3D in mind, and to cope with a variety of potential issues, Nvidia has created a host of game-specific profiles, much like the profiles it uses for SLI. 3D Vision profiles are a little more complicated, though. If SLI doesn’t work, the fall-back behavior is pretty simple: lower performance—frustrating, maybe, but not devastating. If 3D Vision has a compatibility problem, well, all manner of funky things might happen visually, many of which can ruin the sense of depth in the display or send your visual system into a tizzy. What’s more, 3D Vision’s incompatibilities tend to involve certain rendering techniques, so Nvidia will oftentimes ask you to disable some features of a game for the sake of compatibility. In fact, the game profiles will show compatibility information directly onscreen when a game starts up, like so:

Most of the games I tried (all of them relatively new releases) required a few adjustments, many of which meant compromising on visual fidelity somewhat. The most common trouble spot seems to be shadowing algorithms. The profiles frequently recommending dialing back the quality of shadowing in the game’s options, if not disabling shadows entirely.

I tried to get specifics out of Nvidia about what the issues are. Is it one approach, like stencil shadow volumes, that causes problems? But Nvidia has taken the “vague PR blob” approach to answering any and all questions about the technical specifics of GeForce 3D Vision. As a result, we have few tools for handicapping the prospects for future game compatibility with this technology. Instead, Nvidia offers only the reassurance that 3D Vision compatibility is a problem very much like SLI compatibility, and claims that it will take the same approach to surmounting any obstacles: a combination of collaboration with game developers and vigilant profile development. That sounds good, I suppose, but we’re left having to trust that Nvidia will be able to herd cats well enough to make this work.

These issues are relevant because… well, let me tell you about my experiences with using 3D Vision. I first tried Far Cry 2, because that’s a game I’ve spent quite a bit of time playing, so I’m familiar with its (very nice) visuals. When you run the game, the profile overlay pops up recommending that you adjust a couple of image quality options. I had trouble finding these options, until I realized that I had to switch the game into DirectX 9 mode rather than DX10. Once I’d done that, I was able to make the recommended tweaks, including reducing shadow quality.

I also had to turn off the game’s crosshair indicator. This is a common problem with 3D Vision and FPS games: a single crosshair in the middle of the screen will wreak havoc on the 3D mojo. Either the crosshair looks like it’s floating very close to your face, or you’ll see two of ’em. Both are disorienting. In its stead, Nvidia’s drivers have their own crosshair built in, easily enabled with a key combo. The drivers appear to auto-sense the correct depth for the crosshair in response to what’s happening in the game, and it serves its function pretty well, for the most part.

There are spots where this band-aid approach causes problems, though. For instance, some games use the crosshair graphic as an indicator, making it change color or shape depending on what’s happening. The Nvidia crosshair is just an add-on and doesn’t replicate that behavior, so you may lose out by using it. I also had some moments when I was zoomed in, looking at my target either through iron sights or a scope, when the Nvidia crosshair just didn’t seem right. In the grand scheme, those issues aren’t very common, however, and the Nvidia crosshair is generally a decent substitute.

With those changes in place, Far Cry 2 looked pretty good. The illusion of depth was real and obvious enough, and like I said, it’s better than past attempts at stereoscopic 3D, generally speaking.

One of the first things I noticed is that the amount of depth shown, by default, is very much on the low side, something like 16% of the total possible. At this default setting, one can perceive differences in the third dimension, but object themselves look flat—as if one were seeing a cardboard cut-out of a gun placed clearly in front of a cardboard cut-out of a bad guy, with a cardboard cut-out of a tree placed well behind him. I understand the reasons Nvidia chose to make the default setting have a relatively small amount of stereo separation; it’s easier on the eyes and presents fewer problems when the effect isn’t working perfectly. With lots of depth, initially adjusting to 3D Vision after putting on the glasses can take a few seconds to focus—like staring at one of those funky 3D poster things that were popular in in the ’90s. Still, the more I played with 3D Vision and the better the game’s compatibility was, the more separation I found I wanted. At 70-80% of the total possible depth, the cardboard cutout problem is largely banished, and most objects in a game take on a perceptible form. In some cases, I found myself reaching 100% of the available separation and wanting even more. That’s not really practical most of the time, though.

Although 3D Vision added tangible depth to Far Cry 2, the experience wasn’t perfect. Certain things, like on-screen indicators (the hand icon that appears before you open a door, for example), particle-based smoke effects, and water reflections just didn’t work right. They’re not aware of 3D Vision and don’t have proper stereo separation, which confused my eyes. I found that in frantic action, in the midst of a firefight, 3D Vision became disorienting, as my visual system worked overtime trying to process what it was seeing. I’d “lose focus” on the stereo image when an icon popped up and struggle momentarily to regain it. More than once, my character died in a routine skirmish while I was disoriented. This is a tough standard to meet, particularly when you’re dealing with games not expressly designed with 3D Vision compatibility in mind, but anything less than perfection can spoil the added value of a 3D display scheme like this one—especially in a fast action game.

Undaunted, I moved on to the game I most looked forward to trying in 3D: Race Driver GRID, a visually stunning title that just cries out for an added dimension. I figured I’d spend hours racing around the track in 3D with this game. I was shocked, though, when I saw Nvidia’s compatibility recommendations: you’re supposed to disable motion blur, which GRID uses to good effect, and to turn off shadowing entirely. On the face of it, the idea of losing shadowing seemed like a bad idea. In reality, it was even worse than I’d thought. Without shadows under the race cars—particularly your own—the game loses its sense of depth, even with a 3D display. Deeply disappointing.

As I was testing GRID, my kids walked into the room, and I decided to have them try out 3D Vision. They were puzzled by the muddy, doubled images they saw onscreen without the glasses. When I asked my nine-year-old son what he saw when he put on the shades, he happily reported that he could see things correctly again, as if he were wearing some kind of secret decoder glasses. Beyond that, I couldn’t quite get him to articulate that he saw depth in the display. When I asked him what he saw that was different, he said it looked like the cars were floating above the track without any wheels—which is precisely how GRID looks without any shadows beneath the cars. Both my seven-year-old daughter and my son took turns trying on the glasses for a while, but neither of them seemed especially wowed by the effect.

I moved on to other recent games with varying degrees of success. Call of Duty: World at War worked about as well as Far Cry 2, with few visual compromises required except for the Nvidia crosshair, but less-than-perfect results. Crysis Warhead looked quite good at first, until I got into the heat of battle, where this game’s intensive particle effects were completely broken. Plumes of smoke seemed to float way out in front of the screen, far from the objects burning. Water reflections didn’t work, sending my eyes and brain into a twisted tug-of-war, for which there could be no winner. And, well, check out the taillights on this jeep. I snapped this picture of the display with a camera, but you can see the problem without the aid of the glasses:

Two jeeps, one set of taillights

The jeep’s taillights are floating out in space, unconnected to either the right- or left-eye image of the jeep itself. With the glasses on, what you see is two sets of taillights, one off to the right and another to the left.

Mirror’s Edge worked better, and I was able to combine 3D Vision with PhysX effects for a perfect storm of Nvidia marketing hype. Moving quickly through this game’s virtual obstacle courses in 3D is a real treat. The high-contrast color palette of this game brought out another quirk, though: ghosting. When looking at a dark skyscraper juxtaposted against the bright sky, I could see a second, faint copy of the building, somewhat offset. Obviously, the left- and right-eye images were bleeding together, as if the glasses shuttering wasn’t quite up to blocking out everything. Once I noticed the ghosting, I later spotted it other games, but nowhere was it quite as obvious, or distracting, as in Mirror’s Edge.

In a little ray of hope, Fallout 3 was almost perfect, with all quality options cranked, marred only by the need for the Nvidia crosshair and the unfortunate fact that the sky textures were often scrambled somehow. I could see the potential for 3D Vision in this game, but I was beginning to get the impression that this tech would never fully realize its potential.

And then I tried Left 4 Dead.

Valve has obviously been working with Nvidia. For one thing, L4D has its own, depth-aware crosshair that behaves perfectly with 3D Vision, adjusting to depth more quickly than the one in Nvidia’s drivers.

Beyond that, though, here’s the thing: everything works just as it should. Nothing shatters the illusion of depth, even in the smallest detail. It just works. Like gangbusters. Well enough, in fact, to change my outlook about 3D Vision. Perhaps some of it is the fact that this game uses an older 3D engine and a darker, more subdued color palette, but killin’ zombies has never been more fun. I could actually see the possibility of people, quite willingly, wearing funny glasses in order to have this experience when gaming.

I made this discovery about Left 4 Dead‘s 3D excellence shortly before my buddy Andy Brown, of old-school TR fame, arrived in Damage Labs to serve as one of our 3D Vision test subjects. I resisted the urge to share my thoughts with Andy initially. Instead, I played things close to the vest in the hopes of getting an honest reaction out of him that was totally his own. And I started him out with Left 4 Dead, to see how 3D Vision in its purest form would appeal to him. I should say here that Andy is a smart, open-minded guy, but he’s a prototypical hard-core gamer who doesn’t tolerate gimmicks that get in the way of gameplay. I suspected the sheer novelty of 3D glasses wouldn’t count for much in his book.

Somewhat to my surprise, Andy’s initial reaction was pretty darned positive, thanks to the magic of Left 4 Dead. He didn’t laugh, didn’t call the whole scheme cheesy, and seemed to see the appeal of it quite clearly. Andy spent a fair amount of time running around in L4D with the glasses on and came away fairly impressed. His positive reaction to the whole scheme faded as I walked him through other games, though, winding up in bemusement over the state of particle effects and water reflections in Crysis Warhead.

My next victim was my buddy Mike, who joined Andy and me for a little Left 4 Dead co-op action. We took other machines and let Mike sit at the 3D Vision system. Mike was immediately impressed with the 3D effect, and he wound up playing through the entire first co-op campaign in Left 4 Dead while wearing the glasses. Every so often, between levels or during breaks, he’d stop and say, “You know, it really does give you a sense of depth,” or something like that, in slight wonderment. Eventually, he gave that up and just started hitting on Zoey, who we all agreed looks unexpectedly hot in true 3D. Although that might have been the Leapin’ Leprechaun speaking.

After more than, heck, at least 90 minutes of solid gaming with the 3D Vision glasses on (and a few pints of the Leprechaun), Mike didn’t have any complaints about headaches, irritated eyes, or anything of the sort. I expect that next time we all get together for some L4D co-op, he’ll be asking for the 3D Vision system again.

About that performance hit
If you follow these things at all, you probably know that it doesn’t take a terribly expensive video card to drive a 1680×1050 display in the latest games. Of course, stereoscopic 3D will necessarily involve some sort of performance hit, because you’ve basically got to render the each frame twice, once for each eye, in order to achieve a given frame rate. Handicapping the magnitude of this performance hit is difficult. Nvidia claims it does some nifty things in its drivers, including “smart culling,” in order to keep performance up, but it’s very light on the details. Going on what little information I had, I decided to play it what I thought was fairly safe and test 3D Vision with a GeForce GTX 260 graphics card (the version with 216 SPs). The GTX 260 is pretty fast, after all, and not a bad value at present.

Well, that didn’t work out too well. Some games felt sluggish with stereoscopic 3D enabled, even at 1680×1050, and even with features like high-quality shadows sometimes disabled. I really didn’t want slow frame rates to spoil the effect for my test subjects, so I tried swapping in a GeForce GTX 285 instead. When that wasn’t enough, I just went whole hog and plugged in a second GTX 285. That did the trick, but it ought to have—we’re talking about a pair of $350 graphics cards.

Since this is hardware review site, I’m required by OSHA and the FDA to supply you with some benchmark numbers to prove my point. (Well, OK, the numbers don’t have to prove a point, but they’re required anyhow.) I tested on the same basic system config documented here, using a single GeForce GTX 285 graphics card, both with and without stereoscopic 3D enabled. I then tested with two GTX 285s in SLI and 3D Vision enabled, as well. Here’s what I found:

As you can see, the performance hit is sizeable—maybe even bigger than the hit Michael Phelps took off of that bong. In Left 4 Dead, the extra work required for stereoscopic 3D doesn’t present much of a problem for the GTX 285; it still averages nearly 60 FPS. The performance drag is considerable, though: one GTX 285 without 3D Vision is faster than two GTX 285s with it.

Fallout 3 is similar in this respect. A single GTX 285 without stereoscopy is quite a bit faster than two GTX 285 cards with it. Even more unfortunately, SLI is no help at all in Crysis Warhead. Once you turn on 3D Vision, frame rates take a nosedive, regardless.

Of course, the performance hit will vary from one game to the next, and Nvidia claims it’s working on refining its 3D Vision profiles for improved performance as well as better compatibility. Still, right now, the stakes are pretty easy to see: if you want stereoscopic 3D, you’re going to have to fork out for a pretty beefy graphics subsystem, as well. This isn’t an issue one can ignore, because smooth frame rates are an incredibly vital component of perceived image quality in a game. As fundamental as depth is to our visual systems, the illusion of motion is even more crucial.