Desktop Visualization: A New Reality

Every day, researchers at the University of Illinois' National Center for Supercomputer Applications remind themselves of one all-important rule: be famous, not popular. They simply can't collaborate with every researcher who wants to turn data into stunning images. So to keep the nation's scientists from descending on the midwestern computer wonderland, several NCSA projects are trying to take the supercomputer mountain to Mohammed. They aim to someday put the power of "visualization" on the desk of any scientist who owns nothing more sophisticated than a PC and a phone. It's an ambitious goal, however, and roadblocks loom at every turn. Advanced graphics present some of the greatest challenges to today's computer computation and memory capacity, for example. To attain even a moderate level of realism and detail, each frame of an animation sequence must typically be drawn at a resolution of 1024 by 1024 "pixels," or dots on the screen. In full color, each pixel may consume between three and eight bytes of memory. Multiply it out, and you've got over eight megabytes of memory for a frame that may only be displayed for 1/30th of a second as part of a three-minute animation. Storing, transfering, and displaying such massive amounts of data push even modern supercomputers to the limit. So how will advanced graphics become a standard tool for a scientist who only owns a PC? One potential solution is for researchers to run the computationally intensive aspects of their numerical simulations remotely, on high-performance machines such as those operated by the five National Science Foundation supercomputer centers. The scientist could then read the computed data over a high-speed network such as the newly established NSFNet, and analyze it at leisure using special visualization software running on an inexpensive PC or workstation. If it sounds too good to be true, that's because, for the moment, it is. The seminal NSFNet is still limited to a half-dozen "nodes," or entry points, and was only recently upgraded to 1.3 million-bits-per-second capacity, not nearly enough for the rapid transfer of graphics. But there is considerable support for developing greater capabilities. Sen. Albert Gore (D-Tenn.), a long-time advocate for a new high-speed national network, introduced legislation late last year to expand NSFNet to three gigabaud (more than 2,000 times its current capacity) using existing and planned fiber-optic telephone lines. Gore's proposal, which will be reintroduced this spring, would give NSF $400 million through 1995 to establish the new network, and $1.5 billion over five years for other high-performance software and hardware technology. The other half of the equation - visualization software for PCs - is also in its early stages. One promising package (but still restricted to two-dimensional graphics) is called NCSA Image (formerly Imagetool), and runs on the Macintosh and the Sun 3 workstation. It's an example of what its creators at NCSA call "shareware" - software distributed free of charge to the scientific community. Thousands of scientists across the country are already using it to examine the results of fields such as computational fluid dynamics. A planned 3-D version for the color Macintosh II will come closer to achieving high-quality visualization, although it may not be available until next year. For scientists with more money to spend, a new breed of machines called "desktop supercomputers" can already offer instant in-house visualization. The computers, led by the pioneering Ardent Titan, are capable of up to 64 million mathematical calculations per second (about half the performance of a Cray 1), and include special graphics hardware that can compute and draw hundreds of thousands of lines and shaded polygons per second. Yet with a price between $65,000 and $150,000, these single-user supercomputers are within the reach of many labs and university departments. In the last year, desktop supercomputers in the same price range have also been introduced by Stellar (the GS 1000), Apollo (Series 10000), Silicon Graphics (IRIS POWER Series), and AT&T (Pixel Machine). Like NCSA, many of these companies (most notably Ardent and Stellar) are also performing much of the basic research in the fast-growing visualization field. Chemists and biologists have been among the first to capitalize on the new technology. While some design and test theoretical molecules, others use the powerful graphics machines to visualize real-world data. Syracuse University chemist George Levy is using an Ardent Titan to process NMR data from his studies of small proteins and nucleic acids. By transforming the traditionally 2-D data into a revealing three dimensions on the Titan, "we're doing in a minute what used to take over an hour on a Vax [mainframe]," he says. So far, choosing between a PC with a network, and a desktop supercomputer has involved a tradeoff in expenses - "You either do the graphics remotely and pay for the network time to transfer them to your PC, or send just the data and pay for a powerful machine to visualize it with," says Maxine Brown, associate director of the Electronic Visualization Lab at the University of Illinois, Chicago. But it needn't always be that way. Her lab is working on an economical compromise - a local area network of PCs connected to a desktop supercomputer. If the lab's research pays off, scientists may enjoy the best of both worlds: transfer compact data on the cheap, and then visualize it to their hearts content on an in-house supercomputer from their own PC. --G.C.A.

Desktop Visualization: A New Reality

Interested in reading more?

Become a Member of