I have seen some implementations of "true" random number generators. However they are all producing a gauss distribution. I have asked myself, if it would be possible to run these numbers through some sort of algorithm, so that one can get an uniform distribution in the end?
I guess it is impossible, can someone explain to me, why?
The comment by Clement C. is all you need. Let $F(\cdot)$ be the cdf of a r.v. $X$ that is normally distributed. Then the random variable $Y =F(X)$ is uniformly distributed. So if you know how to sample $x$ from $X$, transform the result to $y=F(x)$ and this is equivalent to sampling from a uniform distribution.
Most implementations work the other way around. They sample from a uniform $X$ and use the transform to generate another distribution. I suspect this is also what happens behind the scenes for the "gauss distributions" you refer to.