Today, most workstations have some 3D graphics and image processing support. The speed, functionality, and quality of this support is primarily dependent on the cost of the machine (particularly, the graphics hardware subsystem). In general, the more you pay the more you get. Using the Silicon Graphics product line as an example, an Indy desktop workstation [11] configured with good low-end 3D capabilities costs around $10,000, while an Onyx multi-processor deskside machine with RealityEngine graphics [2] costs $100,000 or more. This order of magnitude difference in system cost represents a substantial difference in interactive 3D graphics quality, capability, and performance demonstrated by these ``ballpark'' comparisons of the RealityEngine to the Indy's standard graphics subsystem:
Even while new low-end and mid-range graphics subsystems promise improved performance and functionality, the rate of improvement is likely to be matched or exceeded by improvements in high-end rendering systems. Some high-end features like multisampling are unlikely to be available in low-end hardware in the foreseeable future.
Applications like visual simulation, large-scale computer aided design (CAD), digital film editing, scientific visualization, virtual reality, and medial imaging use high-end graphics systems to meet their demanding requirements for speed, features, and high-quality rendering. In particular, high frame rates, high CPU-to-graphics bandwidth and low update latency are vital to these applications. These demands are summarized by the word immediacy. Typically, applications that demand immediacy tend to monopolize the entire machine.
To justify such hardware expense, these applications must have a very high return on investment. The advantage of open heart surgery rehearsal or training commercial and fighter jet pilots via computer simulation justifies the expense, but this model for utilizing high-end graphics hardware limits the applications and markets for such hardware.
The monopolistic model of graphics hardware allocation is too expensive for many application domains. Where immediacy is not a key requirement, a network-extensible render service can permit multiple applications to concurrently share a high-end graphics subsystem. Conceptually, such a service lets clients render sophisticated 3D graphics and retrieve the resulting image for display on a low-end workstation or other manipulation. In the same way networked laser printers permit hardware cost to be amortized across multiple users and applications, a networked render service amortizes the cost of high-end graphics hardware and offers higher levels of hardware utilization.