I recently bought a gtx460 for the project, however I must say that its performance is rather mediocre. I expected fermi to be faster...

It takes 10-11 hours to complete a WU, with a gpu usage of about 74%
setting swan_sync to 0 gives about 83-84% usage, but still no big jump in performance.

Right now it is on par with my gtx260, is that normal?

Also wikipedia lists 907.2 GFLOPs while boinc sees 363GFLOPs. Did something go wrong?

i just noticed this too:

# Number of cores: 56

that seems a bit small

Are you using all the CPU cores to crunch CPU projects?

If so, try using 99% of the CPUs (basically one less core). This will free up the core and expedite the GPU task. GPU utilization should rise to over 90%

Vista is typically around 11% slower for GPUGrid, with all recommended optimizations applied. 5 or 6 hours would be about right.

56 shaders is not correct, but it is just a reporting error.

Are you using all the CPU cores to crunch CPU projects?

If so, try using 99% of the CPUs (basically one less core). This will free up the core and expedite the GPU task. GPU utilization should rise to over 90%

Vista is typically around 11% slower for GPUGrid, with all recommended optimizations applied. 5 or 6 hours would be about right.

56 shaders is not correct, but it is just a reporting error.

I tried it, it didn't change the gpu usage.

What can I do for the app to use more than 56 cores?

What performance numbers are expected from my card?

The app is already using more than 56 cuda cores, Boinc is just reporting that the card has 56 cores.
You should be able to get at least 25,000 credits per day, if that's what you mean by performance. About 30K is typical.

Just for another point of reference.

Over the last week or so I've gotten anywhere from 5.55 to 8.33 hrs for a ACEMD2 WU using a GTX460.

This is on a Linux system using the 260.19.29 NVidia drivers.

I don't know what the GPU utilization % has been.