Here's something that baffles me:
I used to mine just with my GTX 660 Ti, then I remembered I still had a GTX 275 lying around, so I built that into my rig as a 2nd card to see what kind of performance I would get. In theory the GTX 275 actually isn't
that bad despite its age, here are some numbers from Wikipedia, quickly edited together:
http://i.imgur.com/10cpcJb.pngThe GTX 275 has 41% of the compute power and 88% of the memory bandwidth of the 660 Ti.
I used to get around 200 khash/s with the 660 Ti even with version 2013-12-10 (now improved to ~280

). So even judging from the old 200 khash/s, I should be getting at least 80 khash/s with the GTX 275, right? Nooope...
50 khash/s.

I'm using K7x32 (thanks Christian!) for the 660 Ti and L60x3 for the GTX 275, which is the setting consistently picked by auto tuning. The other settings are "-C 0 -m 1 -H 2 -i 0", all tested and proven to deliver the best performance I can get on both cards.
Now to get to the point: Why is the GTX 275
that slow? Is it just because it only supports CUDA 1.3? Any tips on how I might get some more performance out of that old piece of crap?
The GTX 275 uses the much older architecture and iirc doesn't have the half-clocked stream processors that came in the GTX 400 series (with Cuda 2.0). Essentially, all bets are off with how the new code interacts with the oldest Cuda capable cards, as the hardware was much different.