whose floating float caculation is faster, cpu or gpu

someone told me that GPU’s float compute ability is stronger than cpu.
I doubt.

can someone tell me if that is true?

It is, a 3GHz intel cpu has about 6 GFlops of computing power.
A GF6800 GPU has about 80 GFlops from what i have read.
So a GPU does have a significantly higher floating point computing power, but it is used differently, a GPU has many small subprocessors and excells at gather ops but has a hard time doing scatter operations, while CPUs have a more general way of doing things and can do scatter and gather equally well.

It is like comparing a car and a plane :
car (CPU) : versatile, precise, very efficient for all tasks in general
plane (GPU) : specialized, needs infrastructure, imprecise, excels at huge tasks.

A big problem with a GPU is to get back result in an efficient manner.