The Windows NT Magazine Lab tested seven representative systems suitable for CAD, 3D rendor farms, video editing, 2D compositing, or other compute-intensive graphics, using both the BAPCo SYSmark/Windows NT test suite and some scripts of our own making. We also ran the same tests on a baseline system for comparison. The baseline machine is a 100MHz Pentium clone with 32MB of RAM, a 2MB Diamond Stealth 64 PCI video card, a 1.2GB IDE hard disk, a 4X IDE CD-ROM, and a 3Com Etherlink III network card.

We conducted all the tests on NT Workstation 3.51, rather than 4.0, because not all the systems had NT 4.0 drivers when we started our tests; all the systems have NT 4.0 drivers now. Each system review contains configuration details. Most systems ran with at least 64MB of RAM (the HP Vectra XA 6/200 ran with only 32MB of RAM) and screen resolution of 800*600 at 256 colors.

The SYSmark/NT benchmark number is a composite linear score based on automated user routines in popular applications. A system that receives a score of 200 is twice as fast as a system that rates a 100. This benchmark uses both 32-bit and 16-bit applications scripted in Microsoft Test: Microsoft's Word 6.0, Excel 5.0, and 16-bit PowerPoint; Welcom Software's Texim Project; and Orcad MaxEDA. For more information about the SYSmark/NT benchmark suite, visit http://www.bapco.com.

Table A summarizes each system's composite SYSmark/NT score, CPU, memory, graphics adapter, and price (excluding monitor). However, some of these composite scores are misleading: Systems with OpenGL accelerated video cards (e.g., TDZ-410, AlphaStation 500, Durango) often suffer a performance hit on ordinary business-type applications or graphics applications that do not make specific use of OpenGL features.

To help you identify performance strengths and weaknesses, we include system performance graphs for individual applications in the test suite and for other graphics applications. Figure 1 shows the individual application scores in the BAPCo suite. You can see that the Alpha-based systems (the Durango and the AlphaStation 500) perform much better on 32-bit native applications such as Orcad MaxEDA and that the TDZ-410 suffers on most business applications. Figures 2, 3, and 4 show system performance on the Lab's graphics test scripts. Figure 2 illustrates the script runtimes for automated routines in Elastic Reality (using stock images and functions), Figure 3 shows the script runtimes for automated routines in Adobe Photoshop, and Figure 4 depicts scene render times of the Alpha systems running a native version of NewTek's LightWave 3D.

The table and graphs present the systems in a variety of contexts so that you can compare their relative merits. An obvious conclusion is that some systems have excellent price and overall performance, and others perform specialized tasks very well, but at higher prices.