Tools and Methodology
The first task is to identify the performance of the various subsystems, including the graphics, I/O and memory subsystems, as well as the processor itself. There are a number of benchmark programs designed to test these components, each with it’s own issues and limitations.
It is important to note that the relative performance between two systems may vary greatly depending upon how that system is used. This includes not only the specific applications, but also whether the applications are run stand-alone or concurrently, as well as the operating system they run on.
Though various flavors of the Windows operating system are by far the most prevalent on the PC platform, benchmarks are less reliable when run on this OS because of the additional layer between the software and hardware. For this reason, DOS based benchmarks (or those using their own OS) are preferred for testing the actual capabilities of the hardware. Unfortunately, the lack of good video and IDE drivers prevents some of the more advanced features of today’s systems, such as write-combining and busmastering, from being utilized.
The benefit of benchmarks which run under Windows is that the same drivers used for real applications are used by the benchmarks, providing a better ‘real world’ scenario. Also, it is conceptually possible to have concurrent tests running to measure the actual bandwidth capabilities (vs. the theoretical bandwidth), however at this time there do not appear to be any readily available benchmarks that take advantage of this capability. In the end, I decided to use ZDBop’s Winbench 99 and Passmark Software’s Performance Test for most of the subsystem performance testing.
One interesting benchmark that runs under DOS is Membench, by Intelligent Firmware. This benchmark is extremely consistent in it’s results, because it does not have Windows resource management to contend with, which oft-times will steal time from the running task to do ‘system management’ operations. This means that the results are actually more reliable where driver support is not required. The downside of this is that only CPU and memory performance can be measured without the use of special non-standard, DOS based graphics and/or IDE drivers.
After analyzing the performance of the various subsystems on each platform, it is then a relatively simple matter analyze the results of various application benchmarks and understand what causes any differences that may exist between platforms. With this information, it is a bit easier to be able to decide which platform is best for a specific environment. Note that no game benchmarks are included, as the focus is upon business and professional usage of the system. For comparisions of gaming performance, it would be best to refer to one of the numerous publications that are dedicated to gaming enthusiasts, and are much more knowledgeable of gaming issues than I am.
Be the first to discuss this article!