What is power for a graphics card measured in?

[Hipshot]

Lets say that speed and power for a HDD is measured in MB/s
For a CPU, the amounths of calculations per sec.
But what 'term' would be correct for a graphics card, ofc a gfx also uses MB/s and calculates data, but more specific for the gfx-card. FPS is no good, since it of course depends on the whole system and varies from game to game.

[Grudge]

HDD speed is not limited to transfer speed only, there are lots of variables; average write speed, average read speed, average seek time, latency and so on. It's the same with CPUs, there are lots of different "calculations", there's not one single metric that by itself is representative of the "power" of a CPU.

Hence, the answer to your question: there is no meaningful "term" to use for graphics cards, as little as there is one for HDDs and CPUs.

[Hipshot]

I know there's more to a HDD than just MB/s, but it's still a good value to measure the performance in, values other than the RPG on a disk has more to do with the seektime and accesstime, not the real speed when a file already is beeing accessed.

[shadd_]

a variety of benchmarks with a variety of resolutions and AA+AF.

of course the benches need to be done with the fastest cpu on the market to dismiss any cpu bottlenecks(wich there are some anyways).

disregard benchmarks from biased(on the payroll)review sites as well as any reviews done with anything but recent drivers.

[primaltheory]

graphics cards are pretty much computer processors with different instructions and different architechure, they are mesured in raw power (mhz/ghz), memmory size/speed (mb/mhz/ghz), then the technical stuff (tris/second and such, stuff I don't know)

they also use FPS...

[^misantropia^]

- clock speed
- memory speed
- read size in bits[1]
- pixel pipelines[2]

[1] 32/64/128/256 bits. Always a power of two.
[2] For every extra pipeline, the card can process another pixel per clock cycle.

[shadd_]

pipeline were a fair way to judge older cards. both nvida and ati have redefined the traditional pipeline with nv g70(7800) and ati r520(1800).

ROPs(raster operation units) are 16 on both cards. ROPS were part of the traditional pipeline but have been seperated in the new generation.

so untill the ROPS are bumped up to match, a 24 pipe 7800 is really only a 16 pipe card on crack and a 48 pipe x1900 is also more or less a 16 pipe card on crack.

eg.

Radeon X1900 with 48 pixel shader processors, 16 TMUs and 16 ROPs (not a true in the traditional sense 48 pipe card)

GeForce 7900 with 32 pixel processors, 32 TMUs and 16 ROPs.(also not a true 32 pipe card in the traditional sense)