For scrypt on GTX TITAN i think this is little bit better:
-d 0 -i 0 -C 2 -m 1 -H 1 -l T14x24 (605khash/s @ 928Mhz)
I'll have to try this out, I'm fairly sure that i'm getting very similar results with my config.
No matter what combination I use, the best I can get with T12x24 for my 780 is less than 1/3 of T12x20, ~170-185kh/s depending on settings.
Confirm some things for me:
- Make sure you are using the latest Nvidia Drivers, 332.21 for full Compute 5.5 support.
- Make sure you are unzipping a fresh copy of the files, not overwriting your current folder. I had a small issue where unpacking the latest releases and overwriting the files did not work correctly and cause funky hash rates. Not sure what went wrong but worth noting.
- Let us know the model of your card.
I'll try out the 14x20 for myself after work tonight and see if maybe it's not a bad secondary option. I'm open to any ideas

At the least we might be able to see if we can identify why this oddness is occurring and maybe make a separate recommendation for those that don't fall under the SMX x 24 bucket as intended. At worst it might be an unknown bug, but no telling until we run everything down.
Using 332.21:Yes. Win 8.1 x64 enterprise, 6gb ram, core i7 920 @ 4ghz
Unzipping to new folder: No difference. x86/x64 version, no difference
Card model: Gigabyte GTX 780 WF V2.0
Memory and GPU usage etcetc is the same, except that the power draw sits at around 45-50% instead of 85%+ with x20
I have the same card and am experiencing the same issue. This is the card;
I'm running the latest build (w/ hotfix). I also compiled it myself, but it has the same issue.
GeForce 334.67 beta driver, Windows 8.1 x64 Pro, 16gb ram, Core i7-3770k @ 3.5 ghz