How We Tested

To create a comprehensive series of tests to determine the best graphics accelerators, we first identified the most important markets for them. We concentrated on two categories for PCI cards and one for high-end Macintosh graphics adapters, as outlined below.

-- General business: 1024- by 768-pixel resolution with 256 colors; for mainstream Windows users and people who use general-business applications.

-- CAD: 1024- by 768-pixel resolution with 256 colors, and 1600- by 1200-pixel resolution with 256 colors; for engineers, architects, and draftspersons.

-- Macintosh graphics: 1152- by 870-pixel resolution with 16.7 million colors; for desktop publishers and graphic illustrators.p

For testing on the PC side, we required the adapters to support a minimum of 1024 by 768 pix els with 256 colors in noninterlaced mode. They had to have a minimum of 1 MB of memory. We tested the boards in a DECpc XL 590, a 90-MHz Pentium system with 32 MB of RAM. We used a 21-inch Nanao FlexScan F760i-W MultiSync monitor.

Our Macintosh test-bed consisted of an Apple Mac Quadra 840AV with 16 MB of RAM and the same Nanao monitor. We concentrated on 24-bit, true-color adapters capable of displaying 16.7 million colors on monitors with screens measuring up to 21 inches.

Performance was our primary criterion for selecting the winners. After we chose the top performers, we ranked the winners and runners-up by considering cost, support options, usability, and any unique features the boards offered. Because of differences in retail and street prices, we considered a 15 percent cost difference to be insignificant.

In the ease-of-use scores, an excellent ranking was reserved for adapters with exceptionally clear and complete documentation and installation software. Adapters received a goo d rating if an average user could install them without referring to the manual. Boards that were rated fair required our testers to consult the documentation, while poor ease-of-use scores indicate we had to reset some setting or jumper and/or consult the vendor's technical-support personnel.

Although ease of installation was a factor in our usability scores, this judgment was tempered somewhat by the fact that once you get even the most troublesome board up and running, you're likely to be concerned only about performance and compatibility for the rest of the board's life.

Performance

We produce tests whose scores give a meaningful reflection of real-world conditions. Our graphics application tests use images produced from CorelDraw, Excel, PowerPoint, and Word for Windows. This month, we expanded our testing to include AutoCAD, which stresses a different set of graphics operations.

We required each PCI adapter to display a variety of Windows and AutoCAD images, ra nging from straight text to two-dimensional and 3-D bar charts to complex full-color drawings. For the Macintosh tests, we used Corel's export filters to convert drawings into PICT format. We also incorporated images from Mac versions of Aldus Persuasion, Excel, and Microsoft Word. Because many of the test images are platform specific, avoid making direct comparisons between Mac and PC results.

We designed our tests to be resilient to benchmark-optimized drivers (i.e., cheating). And, to increase the accuracy of the tests, we used microsecond timing. This allowed us to accurately measure a single screen paint, and it avoided the necessity of drawing the same screen repeatedly (which is unrealistic and also easy to optimize in the driver).

Our Windows tests drew each of the 15 Windows application screens into both system memory and video memory using four different color modes for more than 120 tests. We also measured the time it took to refresh the screen from an image cached in memory at screen depths of 1, 2, 4, 8, 16, and 32 bits per pixel. Well-written applications cache display images whenever possible to improve response times.

Contributors

Jim Kane, Project Manager/NSTL, has been testing network and PC hardware and peripherals for NSTL for the last five years.

Siva Kumar, Technical Analyst/NSTL, specializes in hardware and network-operating-system evaluations.

John McDonough, Technical Writer/NSTL, has been writing for high-tech publications for the last five years.

---

The Windows Graphics Architecture

illustration_link (12 Kbytes)

A Windows application provides generic requests to the Windows API layer. This layer makes calls to the Windows GDI, which in turn talks to the specific driver for the graphics hardware. For accelerated video, an application can talk directly to the DCI. The DCI driver can then receive motion-video data directly from the application, bypassing the Windows GDI. A multimedia application sends a generic command to the MCI, which relays the request to a multimedia device driver.