Surely you mean the application? The client just runs to get the application going and to handle the commo with the server. Rick Alther handles boarding the applications for the grid, but they are owned by University of Washington (Rosetta) and TSRI (AutoDock 4.0).

Hello Vicmot,
I doubt that the project applications have been particularly optimized for any architecture, other than those optimizations included in the selected math library. Some math libraries have very optimized subroutines for different chips. The problem is that this works best if the original university programmers creating the program structure their code to make use of very high level math subroutines, such as BLAS (Basic Linear Algebra Subroutine library).

I have talked about this before (such as http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=3810#29826 ) and Rick Alther has never chimed in with any information. To my dark, suspicious mind this silence on his part indicates that it's all spaghetti code before it makes it out of the university and reaches us. Well, most cutting-edge research programs are by the time they make it out the lab door. So my suspicion is that the only optimizations are those that an architecture-aware compiler produces in conjunction with the math library.

The very best optimizations are those produced in the design of the algorithm itself. We are giving AutoDock 4.0 its first big run over a number of different families of compounds. The results from what Dr. Olson is calling Phase 1 (2000 compounds and 200 HIV varieties) should point out all sorts of areas where the algorithm and code needs improving. So I expect that 2006 is going to see Dr. Garrett Morris and his merry crew of programmers staring into their displays at all hours.

All this is just my personal opinion. Nobody is saying anything to me about this, which in itself I consider to be significant information.
mycrofth