Its simply one of many tools to measure performance. To get a better idea of real world performance, you'll want to run a benchmark that will stress your cards as much as the games you play would.

Graphics cards don't always perform the same in all situations. For example, the gtx 690 will outperform a 780 ti in 3dmark vantage, but when you run 3dmark firestrike, the gtx 690's lower vram starts to show its age and falls behind in score.

>>55439381
FLOPS means nothing from one architecture to another.
Its a theoretical figure derived from this equations
number of ALUs x clock speed x operations per clock
IE the Radeon 7750 has 512 ALUs, it is clocked at 800mhz, and it can perform 2 ops per clock. Hence we get 819,200 FLOPS or 819.2GFLOPs.

The caveat here is that it tells you nothing about how the GPU actually performs. It doesn't take into account how those ALUs are fed instructions, geometry performance, scalar throughout, pixel throughput of the ROPs, texturing performance, how the GPU handles memory intensive operations, etc.

Its basically used as a marketing bullet point, and can only roughly compare two GPUs in the same family. Beyond that its useless.

>>55439477
>>55439445
So how can we really compare cross platform GPUs?

I'm curious just how far behind phones are compared to laptops and desktops for graphics performance

>>55439538
You'd have to write an OS agnostic benchmark suite to test all areas of performance like ComputeBench.