Relation between benchmarks and games

I know that a P5000 GPU will perform much better than a P2000 GPU but what about a smaller difference? For example, take the 6970m CF and the 7970m. People are getting....what.....P6000+ scores on their 7970m (on the M17x)? This gives 3D Mark 11 scores of 5600 for the CF 6970. Now if we compare fps, Metro 2033 has on an average 30% higher fps. On the other hand, Crysis 2 gets about 25% better fps on the 7970m.

This is only an example I used. There are many such cases with many other GPUs. So my questions are -

1) Well firstly, is notebookcheck THAT unreliable with their fps values? I would doubt that but if that is the case, then this whole thing goes away

2) If the answer to 1 is no, then why do we have benchmarks at all? I have noticed that GPGPU performance relates much much closer to the benchmarks the games. So other than to someone who is offloading his heavy computations on the GPU (which I believe on a laptop would form the minority of users) why do we have benchmarks at all? If we purchase the GPU for gaming, then if benchmarks vary this much, we might as well not consider them

3) Is this dependent on the game THAT heavily? I'm assuming the guys that developed the Cryengines worked with nvidia and, for example, Alan Wake guys worked with AMD. So we have differing performance.

I'm just trying to understand how these GPUs really work and what the 3D Mark benchmarks do that is different from games. I'm assuming the benchmarks also try to render a very fancy image with lots of edges and what not.

 

i find that notebookcheck and 3dmark to be a good indicator of how your card will fair.

Notebookcheck doesn't optimize any game...they just run their bench and post...so it can be a good baseline.

3dmark is a good indicator of how your card will do. It is a good tool to compare your card with another. But as you recognized, some games prefer nvidia and some AMD. So that has to be factored into the equation. but that said, two cards that score the same in 3dmark11 will perform similarly (differences in games and game preferences to a certain video card vendor not withstanding).

Your question about dual card vs single card brings in all kinds of additional factors...like game/driver support, GPU scaling etc. Some games scale very well with dual cards while some don't. So in your example, it could be that Metro is very optimized for Xfire. Your Crysis2 example brings in another interesting variable into the equation...DX11 features perform better on 7970m than they did with 6970. So for example, the 7970m does very well in tessellation...much better than the 6970's ever did (even 2x6970's can't match). So game settings will factor into the equation too

As someone who has had Crossfire and SLI rigs all his life, my preference is always to have a single more powerful card versus 2 cards in SLI or Xfire because a single does not rely on game/driver support and is more trouble free...however, my achilles heel is power so my desktops usually run dual cards...currently have SLI 670's which for the most part has been trouble free

 

if you're concerned about performance in a particular game, just try to find a benchmark that covers the systems you are comparing on that game. They are generally abundant.

If you're concerned with overall performance, do a metaanalysis of available game-benchmarks, or use 3dmark scores to approximate.

 

Thank you both.