I'm using GTX480s and I just tried the latest build:
- previous build with K15x16 - 246 KH/s
- current build with F15x16 - ~220 KH/s
- current build with K15x16 - says it requires 3.0 and doesn't validate

other config: -c 0, -i 0, it is second card, so I use the interactive on the other one.

During the last few days I tried using Fxx*xx on both GTX480 and GTX570 and I always get much lower khs compared to using Kxx*xx

I have also noticed a drop in performance since switching to the ulong2 method in fermi. Most noticable in x64. Also, this maybe of no concern of yours as you don't really support cuda 5.5 but, the spinlock kernel crashes the cuda 5.5 cicc compiler when compiling for compute 3.0.

Still crashed cicc when sm_20, went ok with sm_13. Testing against 2013-12-01, 2013-12-07 spinlock is about 5 - 6 khash/s faster.

With fermi kernel 2013-12-01 with the added offset is around 233, 2013-12-07 is about 228. I tested 2013-12-07 with both your precompiled exe and self built from source, results are identical.

Id like to see feedback from more fermi owners. In particular compute 2.0 fermis and even more particular 448 core gtx560ti.

Well i have a pallit one anyway. Cheapest i could find but that shouldn't cause my results to be totally different to yours. I don't think its a problem with os or drivers because i dual boot 32 bit debian and get the same results there.
Although i cant really see it, maybe my weird ddr3 + fsb + pcie 1.1 abomination is causing it.

Anyone using a GTX 760? I've got flags -i 0 -l k96x2 and I've managed to hit 200Khash/s give or take 2. Much better than 100 when I first started a few days ago. Any other configs with better results?

Also, is it normal for the Khash/s rate to vary? I'm hitting 201 while I am using my computer, then when I leave it at night I see dips down to 155Khash/s and everywhere in between. What gives?

Overclock is 1306Mhz core, 3375Mhz memory although I can hit 3402 but I saw no performance increase from the 3402Mhz on memory. Does cudaminer benefit from maintaining a specific ratio like the AMD cards do?

I am running cudaminer on Arch linux from the AUR repository (cudaminer-git) and it works just fine.

There is a dependency issue with the latest update, though. The newest version apparently depends on CUDA 5.0 instead of 5.5 and 5.0 depends on an older version of gcc than the one used in the Arch repositories. I should note that the package is not broken, just outdated. Hopefully it will be fixed soon.

nice, i was at 1280 mhz core 3005 mem with config k94x2 on 182 k/hs. With the new build core 1280 kh/s and mem 3150 i now get 190 kh/s.
Cant seem to push it more, the driver crashes. Only at 69% TDP though and just 60 degrees temp.
Got a MSI GTX-760 TF. System watt uses under full gpu load is only 250w, so while not an AMD card in quite in the green.

No nightly drops btw.

I ran into a weird issue trying the Titan kernel on my GTX670. It was reporting almost 400khash, but every share was being invalidated by the CPU, besides otherwise acting fairly normally. Maybe there's an improvement to be made somewhere that could let Kepler achieve these kinds of khash rates.

For anyone looking for a config for their GTX670, I'm at 180 khash with these settings. If someone has something better, I'm all ears.
-i 1 -l K98x2 -C 2 -m 1

Skimming this thread earlier, there's a lot of mentions of
-l F
-l K
for autotune with a selected kernel, leaving out the specifics (not sure what the #x# represents). Maybe you need a capital F. If it doesn't work, just try -l auto and it'll choose a kernel for you. It says which one it chose in the command prompt as a full descriptor like
K98x2
. You might be trying to use a kernel that's not supported on your card, and using autotune will give you a starting point.

Are you starting from a config that autotune is detecting? Try whatever it finds, then work from that. If just the -l F#x# that autotune chooses doesn't work (and you're using a lowercase L, and you have an updated cudaminer version), you'll need help from cbuchner.

Seems like using just a K for me shows "Given launch config 'K' does not validate".

Interesting aside, it seems the 98x2 I've been using doesn't show up in the debug for autotune, despite being the best config I've found so far. This debug panel gives a few more leads to check out, though.

EDIT: Looking through the results of a couple tries, it seems the results of autotune are pretty inconsistent. Is each configuration being tested only once? Maybe an average more tests per config would work better as an option for people who are going to use the results to find an ideal configuration to set themselves instead of as a final configuration.