A benchmark is a program that lets you measure how fast some part of your computer is. The simplest ones test one particular component, but more complicated tests can strain every part of your system.
The recognized industry standard for an entire system benchmark is the Ziff-Davis Winstone. Their test runs a suite of programs, covering a variety of application categories. The current versions include Winstone 96, with 16 bit applications that run under Windows 3.1 or Windows 95, and Winstone 32, with 32 bit applications for Windows 95 or Windows NT. The test is freely available; you can get a copy from Ziff-Davis, download a copy from their WWW site, or get one on the PC Magazine CD (available at many computer stores). While it's not good for everything, the Winstone 96 and Winstone 32 measurements are by far the best overall system benchmark.
If you want a big-picture feel for how your system rates, you need to install one or both of these benchmarks and get them running correctly on your system. The problem is that they are designed to be run on a clean system. If you've got tons of software installed on your PC, not only will they give inaccurate results, the program may not be able to run at all. For example, I have had Winstone 96 stop working correctly after I installed extensions to Microsoft Word. With the extensions installed, menu items changed around enough that the test script didn't work correctly any more.
Peformance Fact: Serious performance testing requires that you blow away your OS and reinstall with a clean configuration.
There are some other things to watch out for, but the Winstone programs will tell you about most of them. You shouldn't have drivers installed for a sound card, because that will cause the system beep (which does happen during the testing) to take longer, using a digitized .WAV file instead of the standard sound. Often, video drivers have processing running for things like display panning; these all need to get disabled for proper results. Under Windows 95, you can see all the processes running by hitting Control-Alt-Delete to pull up the task list. You shouldn't have anything but the Explorer process going when you start any of the benchmarks.
If I only had one number to judge the performance of a PC by, it's Winstone score would have to be it. That's the purpose behind the test, and I'd say they have been successful. But while it's nice to have a high level benchmark, it won't help you with improving your system. Just because your score says your computer is slow, that doesn't tell you why. For that, you need to get a benchmark that concentrates on one part of your system.
Unfortunately, these tests have many of the same problems. First off, they reduce lots of factors into one number. You don't get a feel for the interaction between the components in your system. For example, if you get a faster CPU, you will most probably end up with faster video. Accordingly, a Graphics Winmark is useless for comparison purposes unless you know every detail about the system it was generated in. There are just too many things involved to say that it is an independent measurement of video performance for graphics cards only.
Performance fact: It's impossible to totally isolate the components in your system. Every piece of your PC interacts with the others to some extent. Understanding those interactions is half the battle for speed.
There really are no benchmarks that let you really know how fast one particular piece is. What component benchmarks are highly useful for is comparing things. You can compare your system to itself after you've changed something, and you can compare your system with ones similar to it to see how it rates. If you want to tweak your PC to improve it's speed, either via hardware or via software, there's a couple of things to look for in a benchmark:
Most of the other benchmarks people like to brag about, like the Norton SI, Landmark, or a host of others, fail for not being specific and comprehensive enough (yes, as odd as it seems, they fail at both those criteria at once). No, there's only one benchmark utility I've ever gotten real results from: Wintune 95, from Windows Magazine. It does a great job of highlighting some of the factors that get ignored by every other benchmark I know of. Specifically, it gives you a spotlight into how your computer deals with memory. That memory information is central to the system tuning we'll get into in future columns; memory access is the ignored factor in PC performance, and it's arguably the most important thing to concern yourself with.
Here's what you need to do before you go any further with the material I'll present. Get a copy of Wintune 95, and get your system running Windows 95 or Windows NT so you can use it (these are the operating systems I'm going to be primarily discussing in the future). It's available on the Windows magazine CD, which shows up in many computer and book stores along with the magazine itself. You can download a copy from their WWW site. It's not very big, but make sure you grab the NT addendum if you're using that OS (you'll need to download this from the home page even if you get the CD version). Try the program out, compare your system with some of the ones in the database, and you'll be ready to start making your PC fast.