Nothing is repeatable.. This whole argument is bunk.. every benchmark you do is gonna score slightly different.. for instance, 3dmark.. etc.
The PC will always be in a different state every time you test it. It could be .00000003 degrees warmer/colder on one test less than an hour apart.. ANYTHING could change.
With this argument, NOTHING is a benchmark.
Dyno'ing my car wouldn't be a benchmark.. because it'll never dyno exactly the same.. perhaps within 5 horsepower.. but even if it dynos 400 twice, it still didn't dyno 400.xx, or whatever the case may be.
In other words.. no 2 results will ever be exactly the same.. whether it be hundreds or thousand of decimals back.. there is still margin of error. Super-Pi is relatively consistant.
Josh
You still have a margin of error, and when the margin of error is 300% in between runs, you know the "value" of it as a benchmark is nill... There are quite a few legitimate benches out there. Superpi just isnt one of them. If it could actually calculate pi in a realistic time frame, and produce reliable reproducable results consistantly across testbeds, then it would be OK to xcall it a benchmark, but in its current form it cant do that... At best it is a poorly written pi calculator and nothing more.