You are making benchmarking WAAAY more complicated than it has to be. We're talking about some dude considering a switch to Linux, but isn't sure the performance is on par. Just load up your game and hit the benchmark button. No sane and rational person is going to be clowning around with driver revisions or regedit, because those types of people think that is more fun than playing the games.
>nothing is running that will taint the results
No, running background crap IS the result, because that's real world conditions, and not some artificial lab condition.
>You need to know which to tweak, and which to leave alone.
That one is easy. You leave all of them alone. Windows tweakers do more harm than good. Besides, replicating benchmark results is impossible after you do brain surgery on the OS.
>You need to decide if you're benchmarking more GPU-heavy or CPU-heavy[...]
You benchmark the games you play. Benchmarking anything else would be completely pointless.
>Only a handful actually do it right.
Rumors say that Hattori Hanzo used to work for AnandTech. I wonder what he's up to these days.
> You are making benchmarking WAAAY more complicated than it has to be. We're talking about some dude considering a switch to Linux, but isn't sure the performance is on par. Just load up your game and hit the benchmark button. No sane and rational person is going to be clowning around with driver revisions or regedit, because those types of people think that is more fun than playing the games.
Benchmarking is uncomplicated in the sense that you can press a button and watch the pretty things on-screen and get it to spit out a number; but is your room a little hotter than usual today? Was something downloading in the background? Did you have a transient network issue that caused some process to stall and eat some CPU time? Is one of your fans running a little slower than usual? Did you wait for the precomputed shaders to fully compile? What about the ones Steam supplies?
It's not about fun, it's tedious work. But without proper controls in place, data is just noise.