Benchmarks

[grandslammer]

Hi again...

Can anyone tell me what they feel is the most reliable benchmark testing software is? Presently, I'm using WinTun97. I can run that thing 4 times in a row, and never come out with the same readings. Kinda depressing, since I'm trying to eak out all the performance I can by changing various settings, jumpers, etc. and I need some reliable feedback.

Thanks Again for your input


Mike P.

[Joel Kleppinger]

The ZDBOP benchmarks are pretty good/reliable benchmarks for 2D testing. Their 3D testing lacks, however.
www.zdbop.com (http://www.zdbop.com)

You might have noticed we use the Winstone99 and WinBench99 benchmarks in our reviews. You can download most of the WinBench99 benchmarks, but it is quite a hefty download.

Another pretty good synthetic benhmark is 3DMark at www.3dmark.com (http://www.3dmark.com) . It measures subsystems for their 3D performance.

I usually shy away from Wintune and Sandra because I don't trust their ability to compare across system classes. It's kind of like using RC5 to compare the power of each CPU. (ok, not quite http://www.sysopt.com/forum/smile.gif)

[BBA]

I like the actual ones that use real apps. Quake2 is a great tuning tool, its fps are usually within 1-2 fps per dozen runs, so it's great for tweaking the sys.

The bad thing is there is really currently no way to cross platform test in game apps because there are too many variables, like resolution, refresh rates, color depths, driver dependancies and etc.

In an Ideal world, the benchmark should only concentrate on the specific part of the system it's testing. In the real world, it's more a matter of how long does it take for the spreadsheet to run when I'm listening to WinAmp and Outlook is running along with some DDE server feeding from a database connection somewhere.

These times are when the Celeron really fades into the poor performance class and the P2/3 really comes on strong, while a benchmark program will devote 100% CPU for it's one dedicated thread, which reduces cache dependancy and gives a skewed result for actual performance.


BBA