The last couple of "tech calls" I've taken revolved around what type of graphics card the customer should select for some specific applications. When I answer these questions, I tend to take a "Switzerland" approach, and answer in very general terms and performance standards. The bottom line is that each graphics card manufacturer offers a unique spin on the same concept: delivering pictures from your PC to a display.
There is one other surprising commonality between the graphics card manufacturers: none of them build a card specifically to support the Rental and Staging industries. Typically, we utilize cards that are maximized for use on a "Gamer PC" or for non-linear video editing. After purchasing a card, we invariably end up diving into the world of special drivers and the need to tweak our applications, in our attempt to make the card deliver the required imagery. In the end, it's best to have (a) some experience with building a PC from the ground up, (b) a keen understanding of how RAM, video RAM and processor speeds interact, and (c), a lot of patience.
Multiple "Stretched" Displays
If you're running a "blended" wide screen application in which multiple projectors are stitched together to form a single image, your total pixel screen count can reach three or four thousand horizontal pixels by one to two thousand vertical pixels -- or bigger. To address a pixel screen count of say, 3600 x 1080 on a pixel for pixel basis, you'll need to output that resolution from your source PC. The reason to do this is to create the best possible image on the large format screen. The complexity arises because a single "standard" graphics card will not generate an image this large.
Content can be created at very large pixel counts within a source program, such as Adobe PhotoShop. Unfortunately, when the content is displayed through a standard card, it is choked down to the resolution at which the card is set (e.g., 1400 x 1050). Standard cards will not output an image over 2048 x 1200 (approximately).
The only way to fully display an extremely large image is to divide the content over multiple outputs. In our sample screen above, the 3600 x 1080 image can be handled by using two output channels, each of which would be set at 1920 x 1080. The overlap of the projectors will take care of the difference between the 3840 total pixels of our two graphics card outputs and the 3600 horizontal pixels of our screen.
Multi-Head Graphics Cards
Multiple output cards (a.k.a. Multi-Head Cards) are prevalent at your local computer store. Within the professional realm, the rub when using them is that we tend to see image "stutter" when displaying animations or video files over the multiple outputs. The reason for this is fairly basic. The graphics card's onboard video RAM is being overloaded as it writes constantly changing content over more than one output. The cards are very happy when displaying static images, like spread sheets or GUIs on one or more screens, but motion video can cause stutter. The only way around this is to maintain as much video RAM per output as possible, and this means installing cards with at least 512K of available video RAM. Of course, more is better. At 512K, you can run two outputs at essentially 256K each, but below this threshold, you'll be entering stutter city even with a simple animation.
More Than One Way to Skin a Screen
The other way to get a pixel-for-pixel, perfect image on a blended screen is to dice the source content over multiple PCs -- each with a single output card. Each PC is then only responsible for a specific portion of the screen, and it can concentrate its processing power on a "standard" size portion of the total screen. Animations and other "motion" effects tend to work well with this setup, and a number of systems are now available in the marketplace that support this functionality -- including many media servers originally designed for the lighting world.
Users should note that issues could arise when combining this style of technology with seamless switchers, specifically when live windowing is added over background images from the PC. The addition of the required buffers in the seamless switcher can cause a tear to appear in the seams between the projector images.
While the content starts at the same frame on each PC, the graphic cards themselves are not locked together, and can be as much as a half a frame out of synchronization with respect to each other. While that doesn't seem to be a great deal of time, the interval can actually increase by as much as a full frame when the signal passes through the seamless switcher. The switcher channels are locked together, and they output signals to the projectors simultaneously, but any PC content that arrives after the first frame of video is sent to the projectors must wait for the second frame.
To explain this issue, I use the analogy of a train station. The Switcher is the train terminal, and every 60th of a second, a train leaves the terminal. The content from the various source PCs arrives at the station -- and they're all trying to get on board the same train. But if one of the pieces is just a few lines late, it misses the first train and has to hop on board the second. When the pieces are put together on the other end, the missing piece causes a tear to appear.
This issue only appears on fast moving animations or videos, but with static images, you'll never see the tearing. The only way to fix this issue (with separated PCs) is to genlock the entire system together. We don't see this from a multi-head card because all the outputs are locked to each other. Remember that your choice of which way to go should be based on the type of images you are displaying. Static images work well over multiple output cards, while animations work better over a distributed PC system.
A Matter of Scale
These methods of delivering "background" content to a blended screen are required when a "pixel-perfect" image is required. Typically, however, the requirements of the application only require pretty pictures or some nebulous, abstract background imagery -- or even the ubiquitous low-contrast monochrome shot of the corporate headquarters.
In those cases, the best approach is to create your content to fit the screen's aspect ratio, and then to use the system's scalers to stretch the content over the entire screen. This method opens up the type of source machine to just about any PC or video playback device. When working in this method, I tend to recommend a one-for-two ratio. Specifically, for any one input pixel, it should not address more than two screen pixels. As you move beyond this ratio, the content starts to look pixilated on screen. You can see this phenomenon if you ever take a standard definition DVD player and stretch the image over two or three projector blends. It may look awful, but if that's what the producer wants... who are we to stand in the way?