From a release on NVidia's web site, that I clicked through from a news page:
"The Tesla GPU features 128 parallel processors, and delivers up to 518 gigaflops of parallel computation."
Unless I'm misreading something, that's half a teraflop on a single chip.
I'm in the middle of Charles Stross' Accelerando. The things he projects about the amount of computing power available seem less far fetched today than they did yesterday somehow.