>It isn't complicated
Getting reliable, consistent, meaningful performance numbers is in fact, extremely complicated:
* You need a consistent way to reproduce the exact same outputs - accounting for things like the game's RNG. You can't just walk around and snap the FPS counter in the corner of the screen and call that good.
* For Windows (and occasionally Linux) you need to ensure nothing is running that will taint the results (updates, AV scans, etc)
* Sometimes individual driver versions work very poorly with a specific game. Just because it ran badly doesn't mean you got good data, it may just be a bug in that specific driver version
* You can't just run the benchmark once. You need to run it many times, establishing run-to-run variance
* There are often a good dozen-to-hundred individual OS settings which can impact performance, and in some cases run-to-run variance. You need to know which to tweak, and which to leave alone.
* Sometimes the result of individual in-game settings differs between driver versions. Just because setting X had a big impact once, doesn't mean it always did
* FPS is not a great metric - it's an average. You need to check and see if there are huge frametime spikes. If there are, the game will have a 'good' FPS but feel horrible to play due to stuttering.
* You need to decide if you're benchmarking more GPU-heavy or CPU-heavy - those types of benchmarks require drastically different settings. If you run a CPU-like benchmark you may see a wildly different gap in framerate compared to a GPU-heavy one for the same game.
Benchmarking properly means accounting for thousands of tiny variables. Only a handful actually do it right.