system performance. Finally, we used Intel's MMX benchmark. We ran all tests under Windows 95.
The Word test performs a mix of operations that simulate intensive word processing. The File I/O test stresses a system's hard drive subsystem. The Insert, Paste, Select, and Format tests work the video. The Calculation, Compare, and Sort tests primarily measure the efficiency of the processor and memory.
The Excel test emphasizes floating-point calculation as well as housekeeping operations. The Clear, Insert, Delete, Paste, and Screen tests stress a system's video subsystem. The Calculation test measures the processor and hard drive efficiency, while the File I/O test works a system's hard drive subsystem.
The Microsoft FoxPro 2.6 for Windows benchmark uses five database tables: link, book, authors, order, and entry. The test runs Extended Edition FoxPro, which uses the system's available extended memory for caching. The FoxPro benchmark stresses a system's CPU pr
ocessing, the hard drive subsystem, and the memory subsystem.
NSTL's InterMark benchmark tool profiles applications by capturing the system calls and actions. InterMark then replays these actions to test a system's components without the need for third-party applications. A precision event timer measures the response time for each task. NSTL used the video, hard drive, CPU, and CD-ROM components of the InterMark suite to test these MMX PCs.
We divided the video tests into three categories: Windows draw, bit-mapped images, and non-bit-mapped images. The Windows draw tests use the Windows API calls to draw images, including text, a picture in the Windows metafile format, horizontal and vertical lines, rectangles, and ellipses. Images are then stretched to one and one-half and then twice their original size.
Hard drive tests measure the speed of the hard drive for reading and writing. They simulate sequential, constant-rate sequential, random, localized-random, and segmented activity in varying
block sizes. The result is a measurement of the average response time, the sustained rate, the burst rate, the random-access time, and CPU use of the hard drive.
CPU tests measure processor performance for both integer and floating-point calculations. These tests provide an index of the computing power of the processor. CD-ROM tests examine several aspects of CD-ROM performance: random service time at 500 and 1000 milliseconds, cached service time at 13.3 milliseconds, and CPU use at 550 KBps.
To measure MMX performance, NSTL uses an automated test provided by Intel that puts the systems through a gauntlet of multimedia tasks: home photo editing with Adobe PhotoDeluxe, playing an Intel Indeo multimedia video clip, playing an MPEG-1 video file, a Direct 3-D game, and business photo manipulation with Adobe Photoshop. These are all common tasks that the MMX technology is designed to exploit.
We also evaluated the systems for usability. Our checklist includes a large number of items, ranging fr
om how easy it is to open the case to whether setup and configuration utilities are stored in CMOS or on disk. We like to see ports labeled on the rear panel, zero insertion force (ZIF) sockets for CPUs, and a healthy dose of clear, helpful documentation.
illustration_link (13 Kbytes)
At $1750, Polywell Computers' Poly 5200Tx offers the most bang for the buck.
Test Specs
