I tried that build, it wouldn't run on my machine.  I was probably missing a dependency or something, LOL.

I just tried the cudaminer build provided here and I'm closer to 360 khash/s now, which is a 1-1.5% improvement over K kernel.  I have no idea why my build is so different (yet works)!  I noticed that the build I linked was built for x86, I might try doing that on my machine and see how it turns out.

EDIT:  Just tried it with your settings, wow... that's quite the improvement!  Up to about 380 khash/s now.  Thanks for the suggestions!  Specifying -C 2 was the culprit--that option shaves off about 25-30 khash/s on my rig.

Thanks for the suggestions!  Running -m 1 and -C 0 and -l auto ended up with these results:


Not sure why auto-tune is reaching so hard on the launch configuration calculations... last time it was 64x16, this time it's a whopping 372x3!  And it's still selecting the K kernel instead of the Y kernel, but I'm guessing that's just autotune's process thinking the K kernel would be a better choice.

-----

I decided to specify the Y kernel.  I tried several configurations, including Y64x16, Y64x8, Y32x32, Y16x32 and Y32x16... all of them crashed, then finally I specified Y8x32 and got this:


Setting -C 2 and omitting the -m 1 option bumps the numbers up very slightly (to around 122 khash/s).  Still can't get up over 300 khash/s for some reason.  Any additional thoughts would be appreciated, I'm glad to test any option(s) you can think of or recompile in some other way.  Heck I'll even dive through some of the code if you have a logical way for me to debug it!

-----

I'm all for experimentation to find the optimal configuration, but I'm not sure there is an optimal configuration for the code/build I have right now that's superior to the 12-18 release from last month.  The only other alternative I can come up with is I built the wrong executable (Release|x64) or I copied the wrong executable/DLLs in when performing my test(s).

-----

EDIT:  I just noticed you made a commit about 30 minutes ago, so I pulled that down and compiled it as well, just in case there would be any change.  Sadly the numbers above still stand even with the newest build.

I'm having some difficulty pulling off improved numbers after building from the repository.  I pulled the repository about 6 hours ago so I'm pretty sure I have the newest code to build from.

For reference, I run an i5-3570K @ 4.2GHz and a GTX 680 (2GB).

-----

With the 12-18 build I get around 350 khash/s, and my command line is: cudaminer -r 10 -R 30 -T 30 -d 0 -H 1 -i 0 -C 2 -l K8x32

A quick sample of the output I receive with this config...

[2014-01-31 23:00:40] 1 miner threads started, using 'scrypt' algorithm.
[2014-01-31 23:00:40] Starting Stratum on stratum+tcp://[address-omitted.com]
[2014-01-31 23:00:41] GPU #0: GeForce GTX 680 with compute capability 3.0
[2014-01-31 23:00:41] GPU #0: the 'K' kernel ignores the texture cache argument
[2014-01-31 23:00:41] GPU #0: the 'K' kernel requires single memory allocation
[2014-01-31 23:00:41] GPU #0: interactive: 0, tex-cache: 0 , single-alloc: 1
[2014-01-31 23:00:41] GPU #0: using launch configuration K8x32
[2014-01-31 23:00:41] GPU #0: GeForce GTX 680, 8192 hashes, 58.93 khash/s
[2014-01-31 23:00:41] GPU #0: GeForce GTX 680, 98304 hashes, 301.53 khash/s
[2014-01-31 23:00:41] accepted: 1/1 (100.00%), 301.53 khash/s (yay!!!)
[2014-01-31 23:00:45] Stratum detected new block
[2014-01-31 23:00:45] GPU #0: GeForce GTX 680, 1343488 hashes, 345.08 khash/s
[2014-01-31 23:01:43] GPU #0: GeForce GTX 680, 20709376 hashes, 355.88 khash/s
[2014-01-31 23:01:52] Stratum detected new block
[2014-01-31 23:01:52] GPU #0: GeForce GTX 680, 3137536 hashes, 356.36 khash/s
[2014-01-31 23:02:00] GPU #0: GeForce GTX 680, 2818048 hashes, 355.66 khash/s
[2014-01-31 23:02:00] accepted: 2/2 (100.00%), 355.66 khash/s (yay!!!)
[2014-01-31 23:02:26] GPU #0: GeForce GTX 680, 9281536 hashes, 355.39 khash/s
[2014-01-31 23:02:26] accepted: 3/3 (100.00%), 355.39 khash/s (yay!!!)

-----

With the 01-20 build (I created on my machine using VS2012) I get about 120 khash/s, and this iteration I ran this command line: cudaminer -d 0 -H 1 -i 0 -C 2 -m 1 -l K32x32

[2014-01-31 22:40:15] 1 miner threads started, using 'scrypt' algorithm.
[2014-01-31 22:40:15] Starting Stratum on stratum+tcp://[address-omitted.com]
[2014-01-31 22:40:15] Stratum detected new block
[2014-01-31 22:40:16] GPU #0: GeForce GTX 680 with compute capability 3.0
[2014-01-31 22:40:16] GPU #0: interactive: 0, tex-cache: 2D, single-alloc: 1
[2014-01-31 22:40:16] GPU #0: 8 hashes / 1.0 MB per warp.
[2014-01-31 22:40:16] GPU #0: using launch configuration K32x32
[2014-01-31 22:40:16] GPU #0: GeForce GTX 680, 32.90 khash/s
[2014-01-31 22:40:31] GPU #0: GeForce GTX 680, 129.99 khash/s
[2014-01-31 22:41:25] GPU #0: GeForce GTX 680, 130.92 khash/s
[2014-01-31 22:41:25] accepted: 1/1 (100.00%), 130.92 khash/s (yay!!!)
[2014-01-31 22:41:44] Stratum detected new block
[2014-01-31 22:41:44] GPU #0: GeForce GTX 680, 125.98 khash/s
[2014-01-31 22:41:54] GPU #0: GeForce GTX 680, 123.38 khash/s
[2014-01-31 22:41:54] accepted: 2/2 (100.00%), 123.38 khash/s (yay!!!)
[2014-01-31 22:42:17] Stratum detected new block
[2014-01-31 22:42:17] GPU #0: GeForce GTX 680, 127.27 khash/s
[2014-01-31 22:42:31] Stratum detected new block

-----

GPU monitoring during both of these sessions can be found here:  Images  -- what's odd is the spikiness of the GPU when running the beta build.

Any thoughts on options I should try?  I've attempted to force the texture cache off (-C 0) with no improvement.  I've tweaked my launch configuration (-l auto suggested K64x16, which seems odd but I went with it initially, no difference in output) without success.  I tried both omitting and providing -m 1 but there was no significant difference in performance either.