The 12-18 version is having problems with my Tesla C2075's.   I have two C2075s on a server. I do not use #0  for mining.  I run cudaminer on #1. Before the 12-18 version came out, everything was fine. I use -d 1 -i 0 -C 2 -l F14x16 and get about 178kh/s. The driver was 311.35 (win8 64bit, cuda 5.0).

After the 12-18 came out the first thing I found out was that cudaminer couldn't find the gpu card. So I upgraded to the latest driver 331.82. Funny things began. I couldn't benchmark (-D) #0 because if I used -i 1 I get a lof garbage on the screen; if I use -i 0 I get a lot of cup-doesn't-validate errors. If I bench mark #1 card, I get low performance at ~130kh/s but the most strange thing is that the nvidia-smi program reports that the #0 card is using power and getting hot when I run cudaminer on #1 (-d 1) -- on the cudaminer console it does say "#1" as shown below. So somehow the miner is using the wrong card.
Then I figure maybe I should upgrade cuda toolkit to cuda 5.5. After I did, the driver version went down to 320.57, but cudaminer can still see the cards. However not only the 12-18 cudaminer uses the wrong card, even the 12-10 version and older start to use the wrong card, too!

This is the output of 11-20 version. Its performance is top knotch compared with all versions

Please let me know what you can make out of it. since I want to use only #1 for mining,.