n the processed foods industry, manufacturers delight in releasing recipes that require their product. The maker of Ritz crackers, for example, has a “Mock Apple Pie” recipe (see http://www.nabisco.com) that adorned its boxes for years—a recipe that replaces apples with lemon-juice-soaked Ritz crackers. It uses an awful lot of crackers. Very interesting when you consider that apples are easily available in the targeted markets, regardless of season; that the crackers cost as much or more than the required apples; and that few people would prefer the taste of damp, lemon-soaked crackers to real apples. The same behavior occurs in the computer industry. The biggest proponents of computer graphics research these days aren’t the graphics companies, but companies who have invested heavily in the development of general-purpose hardware, software, and operating systems and need a way to sell their “crackers.” The computer technology of five years ago performed quite adequately for most household tasks (word processing and tax preparation), so manufacturers need a way to sell in a saturated market. Collaborative virtual reality research is driven similarly. The companies making the greatest investment, both through in-house development and funding of external research programs, are those with similar depth of investment in network infrastructure. They need a way to sell their bandwidth. Are next-generation computer graphics and networked VEs just modern computing’s version of mock apple pie? Diminishing returns Special-purpose graphics hardware technology has improved steadily, and graphics software has leveraged off of spectacular developments in general-purpose hardware. General frameworks and interfaces for graphics production have undergone the scrutiny of the entire community, resulting in a suite of effective standards. Striving towards solving the rendering problem has served computer graphics researchers well for years. Likely another 20 years or more of research remains. The speed at which we’re drawing pictures (or warping depth images, or computing radiosity solutions) will double, and double again, and then again. Drawing today is a million times faster than it was for Sketchpad in 1963, with moderately improved realism. But at some point, when the rendering quality and speed have doubled yet again, no one will notice. The CPU industry is rapidly approaching that point: the processing speed of desktop computers has doubled eight times in the last 15 years. Most users who purchased first-generation Pentium II’s don’t need an upgrade now, even though available speed has doubled since. The latest attraction is the speed of the bus that conveys information to the processor, rather than the processor itself.