37 Which video benchmark is the best?
Description
This article is from the comp.sys.ibm.pc.hardware.video Frequently Asked Questions, by Michael Scott with numerous contributions by others. (v1.0).
37 Which video benchmark is the best?
I won't stand at the pulpit and get carried away, but here are some things to consider when looking at benchmark figures.
[From: Dylan Rhodes (Formerly of Hercules)] "Any benchmark program is separated from the real world to some degree. The fastest benchmark score on the planet means little to the user if their applications crash, or if they can't get help when they need it."
[Michael Scott (scott@bme.ri.ccf.org)] 1. The first thing to remember is that a benchmark measures the speed of certain specific operations that the computer is performing. You have to decide if a given benchmark is measuring anything that is meaningful to _you_. This isn't always easy, because often benchmark authors don't provide details on exactly what operations their test suite is performing.
2. Results from one benchmark program can not be extrapolated to other applications or benchmarks. In particular, VGA (DOS) benchmarks may be completely unrelated to GUI (i.e. Windows 3.1, OS/2, etc) benchmarks. This is because the VGA circuits on many video cards are completely separate from the graphics accelerator (Matrox is an example).
3. Comparisons of the same benchmark on different systems may, or may _not_ be meaningful. For example:
Most so-called 'video benchmarks' rely heavily on the CPU, and may not be good indicators of the speed of the video card itself. This is not necessarily a fault of the benchmark author. For example, the majority of VGA operations are performed in the CPU, then the raw pixels are dumped down the bus. This implies that _all_ programs which measure the speed of VGA operations are highly dependent on CPU speed.
One particularly popular graphics benchmark is 3DBench. This is a VGA-based benchmark that will _not_ take advantage of any acceleration capabilities of your video card. It strictly measures DOS VGA speed which is highly CPU dependent. As a result, it is _not_ a good measure of video card speed, but rather measures combined CPU _and_ video card _and_ bus speed. In fact, I believe it was written before VLB even existed, so I doubt it takes advantage of that, either. It is very difficult (impossible?) to measure the pure VGA speed of a card because of this CPU and bus dependency.
GUI-based benchmarks consist of WinMarks, WinStones, WITS, Xstones, etc. Again, most of these are highly CPU dependent, but the advantage of these benchmarks is that when used with the appropriate driver for your video card (i.e. _not_ the VGA/SVGA drivers that come with Windows 3.1 or XFree86) they can take advantage of your card's acceleration capabilities. In particular, WITS and WinStone measures time real-world applications, so they are a closer indicator of how much of a speed increase you should see on a day-to-day basis.
4. Don't expect a new video card to make your whole system scream. No matter how fast a video card is, it's only responsible for a portion of the overall system speed. You won't get Lamborghini performance out of a Lada, even if you put a V8 in it. :-)
Continue to:
My Books
