|
|
||||||
|
Graphics Scalability Comes of Age (Continued)
This same data explosion is all too familiar, however, even at home. There was a time, not very long ago, when an 8GB hard drive seemed fairly spacious; now it's enough for a set of 5MPixel holiday photos and a couple of video clips. But there's far more to come...today's 5Mpixel cameras work at much lower precision and resolution than cinematic film. Increasing resolutions and precision to beyond 35mm film to enable high fidelity 7"x10" prints will push data sizes towards 100MB, compared to a typical 2MB today, enough to ensure that flash memories need Moore's law for many years to come. Putting this into high fidelity video means that filming that all important school play requires some 5.4GB/s, or 19TB for 1 hour; data-loads that demand scalability in the system. Today ViewSonic has a 9Mpixel display that is routinely used with entry-level Silicon Graphics Onyx4 systems. Today's ultimate power-user tool, is tomorrow's home commodity; thin, light and with eye-limiting resolution (which means you need a magnifying glass to see the dots). But, increasing resolution on the screen increases the load on the GPU in the system.
The Power of Scalability A common example of effective scaling can be found in any supermarket. In order to serve more shoppers, the supermarket opens more cashiers. If each cashier takes a line of carts, then increasing the number of cashiers automatically increases the number of carts per minute. This is an example of successful scaling. If, however, each cashier were to take only a specific product in their line, say one for dairy, one for bakery, one for fruits etc., then each shopper would have to travel from cashier to cashier in order to pay for the contents of a single cart. Each shopper spends a great deal of time moving from cashier to cashier and adding cashiers doesn't shorten a shopper's time through the line. This is an example of very unsuccessful scaling. In visualization, we often see this inefficiency in visualization clusters, where adding resource doesn't always increase performance. Multiple CPUs and GPUs makes programming more complicated for software developers. As in the supermarket example, we have to pick the correct way to distribute work among GPUs and CPUs in order to achieve successful scaling. This complex problem is made trivial by graphics APIs such as OpenGL Performer™, and OpenGL Multipipe™, which allow programmers to focus on their application goal, not on how to achieve scalability. With the coming increase in data and screen resolution, home PCs are beginning to add multiple CPUs and multiple GPUs to provide higher performance.
Scalability -- The Key to Innovation without Limits The convergence of mainstream graphics scalability and the scalable Linux and Itanium® 2 processor based Silicon Graphics architectures creates an exciting new option for users. When PC performance is not sufficient, these same new applications will be able to leverage the unique Silicon Graphics architecture to deliver significantly higher levels of performance. This is made possible through its award winning ccNUMA architecture which ensures that all system capabilities, such as memory size and system bandwidth scale to extraordinary levels as the system grows. As a result, applications scale seamlessly, allowing users to effortlessly break their barriers to interactivity and deliver innovation without limits. Related article: Scalable Graphics: Coming to a PC Near You (PDF ) 
Images credits:Image courtesy of Fakespace, Inc. and Oak Ridge National Laboratory | ||||||
