How does one compare graphics capabilities?

[Comdrpopnfresh]

Since architectures vary, I would suspect core and memory clocks alone are not very helpful. What about the pixel fillrate, texture fillrate, and bandwidth as reported by gpu-z? I noticed these values changed as I overclocked. I've got a 4870 and gpu-z reported 14.4/36.0/121.6 for those three values- does this relate to performance in a way that differing graphics cards can be compared? Are these values 'good?'

 

[blanketyblank]

I'd say the best comparison is game benchmarks followed by benchmarking programs which basically simulate games. It matters little how great your numbers are if none of the games you play work well with it. Also each of those measurements have varying effect on games so it depends what the game uses more heavily.
Since drivers and many factors can alter experience in a particular game I don't think there is much you can actually use to tell performance ahead of time.

 

[happy medium]

Download this and compare your score with others.

http://www.futuremark.com/benchmarks/3dmarkvantage/download/

You should score about 10,000. I did.

 

[Seero]

Since architectures vary, I would suspect core and memory clocks alone are not very helpful. What about the pixel fillrate, texture fillrate, and bandwidth as reported by gpu-z? I noticed these values changed as I overclocked. I've got a 4870 and gpu-z reported 14.4/36.0/121.6 for those three values- does this relate to performance in a way that differing graphics cards can be compared? Are these values 'good?'

Those numbers mean nothing. Start your favorite game on default clock, then start it again with OC. If you see 20% boost, good, otherwise, save electricity.

 

[jvroig]

If you see 20% boost, good, otherwise, save electricity.

Why 20%? Where did you base this figure from? Why won't a 10 or 15% boost matter?

 

[NoQuarter]

Why 20%? Where did you base this figure from? Why won't a 10 or 15% boost matter?

Uhm I think all he was getting at was, if you can notice the difference and it helps your game play smoother then do it, otherwise don't bother.

 

[gorobei]

the only real world measurement is to run a game, benchmark the framerate and assess any image quality differences(msaa, csaa, ssaa, aaf, render errors)

thats the only apple to apple comparison you can make. any shader/bandwidth/fillrate numbers comparison is like those car guys who put a 600 HP v8 in a toyota 510 but dont mod the suspension. the numbers on paper sound good but when the rubber hits the road the car tears itself apart or spins it's wheels cause it cant get any traction.