The key feature of Groestlcoin is its good performance on a wide range of platforms and reduced ASIC advantages the algo contains. This is in part dure to the Grøstl algo's resistance to parallelization.
*cough* Groestl is a direct AES derivative, using the same S-box but a somewhat bigger state matrix. Perfectly suited for doing an FPGA or ASIC implementation.
Note: We're currently achieving 7.8 MHash/s with an nVidia 750Ti card and 11 MHash/s with a 780Ti.
Be afraid. nVidia is going to take the cake. Also: stock up on 750 Ti's