Fermi specs for CUDA

I’ve tried to find this information elsewhere but failed.

How many threads can a multiprocessor on Fermi run at the same time?

How many blocks can a multiprocessor on Fermi run at the same time?

What is the maximal number of threads per block for Fermi?

Can the grid on Fermi have 3 dimensions? (GTX 285 can have 3D blocks but only a 2D grid)

It is all in Appendix G of the 3.0 release programming guide.

Thanks, I’ve must have misssed that.