Hmm it seems that in the latest build is the X-Kernel not anymore working on my Fermi.
Getting this error messages repeating:

[2014-01-20 09:06:53] GPU #0: cudaError 4 (unspecified launch failure) calling '
cudaMemcpyAsync(X, context_odata[stream][thr_id], mem_size, cudaMemcpyDeviceToHo
st, context_streams[stream][thr_id])' (C:/Users/Patrick/Desktop/cudaminer/cudami
ner/salsa_kernel.cu line 889)

[2014-01-20 09:06:53] GPU #0: cudaError 4 (unspecified launch failure) calling '
cudaStreamQuery(context_streams[stream][thr_id])' (C:/Users/Patrick/Desktop/cuda
miner/cudaminer/salsa_kernel.cu line 863)

[2014-01-20 09:06:53] GPU #0: cudaError 4 (unspecified launch failure) calling '
cudaMemcpyAsync(context_idata[stream][thr_id], X, mem_size, cudaMemcpyHostToDevi
ce, context_streams[stream][thr_id])' (C:/Users/Patrick/Desktop/cudaminer/cudami
ner/salsa_kernel.cu line 843)

[2014-01-20 09:06:53] GPU #0: cudaError 4 (unspecified launch failure) calling '
cudaStreamWaitEvent(context_streams[stream][thr_id], context_serialize[(stream+1
)&1][thr_id], 0)' (C:/Users/Patrick/Desktop/cudaminer/cudaminer/salsa_kernel.cu
line 851)

[2014-01-20 09:06:53] GPU #0: cudaError 4 (unspecified launch failure) calling '
cudaEventRecord(context_serialize[stream][thr_id], context_streams[stream][thr_i
d])' (C:/Users/Patrick/Desktop/cudaminer/cudaminer/salsa_kernel.cu line 857)

My Start line:
D:\bitcoin\yacoin\test>cudaminer.exe -a scrypt-jane -i 1 -l X13x1 -o http://yac.
coinmine.pl:8882 -O pato.2:password -C 1 -b 4096

It works when using the Fermi kernel, but sadly that means lower hash rate (from 0.64 to 0.41). Back to using older build.

Yac takes a while to fully confirm on http://yac.coinmine.pl/ It will slowly change over. So dont worry, you'll get paid.

I'm finding YAC to be falling at the moment, it might be better to look at other coins.

Am I missing something obvious here ? If I let it autotune, it's best is around K59x2 (GTX 660).

 If I try to specify that manually :

cudaminer.exe  --algo=scrypt-jane -d 1 -l K59x2  -H 0 -o stratum+tcp://yac.coinmine.pl:9088 -O user:pwd

[2014-01-20 15:05:35] GPU #1: cudaError 4 (unspecified launch failure) calling 'cudaStreamWaitEvent(context_streams[stream][thr_id], context_serialize[(stream+1)&1][thr_id], 0)' (C:/__test/CudaMiner-master/salsa_kernel.cu line 814)
[2014-01-20 15:05:35] GPU #1: cudaError 4 (unspecified launch failure) calling 'cudaEventRecord(context_serialize[stream][thr_id], context_streams[stream][thr_id])' (C:/__test/CudaMiner-master/salsa_kernel.cu line 820)
[2014-01-20 15:05:35] GPU #1: cudaError 4 (unspecified launch failure) calling 'cudaMemcpyAsync(X, context_odata[stream][thr_id], mem_size, cudaMemcpyDeviceToHost, context_streams[stream][thr_id])' (C:/__test/CudaMiner-master/salsa_kernel.cu line 852)
[2014-01-20 15:05:35] GPU #1: cudaError 4 (unspecified launch failure) calling 'cudaStreamQuery(context_streams[stream][thr_id])' (C:/__test/CudaMiner-master/salsa_kernel.cu line 826)
[2014-01-20 15:05:35] GPU #1: cudaError 4 (unspecified launch failure) calling 'cudaStreamSynchronize(context_streams[0][thr_id])' (C:/__test/CudaMiner-master/salsa_kernel.cu line 163)
[2014-01-20 15:05:35] GPU #1: cudaError 4 (unspecified launch failure) calling 'cudaStreamSynchronize(context_streams[1][thr_id])' (C:/__test/CudaMiner-master/salsa_kernel.cu line 164)


 I've used the latest code , building in Release x64.

@cbuchner
Big thanks, new version works.
Even better, hash rate went from Thursdays build to current build from 0.63 kh/s to a whopping 0.93 kh/s! (with -L 2 and -l X54x2 and -i 1)
Will test some more -L variants

How can I compile a x64 version?
When I select Release x64 and try to run it, I get an application error: 0xc000007b...

I read it makes a big difference, so I want to try...

That makes sense, silly me :p works now

I used the dlls from the x64 folder of the 18-12 release.
Also, autotune crashed for me: -b 2048 -L 3 -i 1 --algo=scrypt-jane
Edit: It doesn't crash with -L 4, getting me ~4.25 khash/s

I get 230-250 khash/s on my standard 2GB GTX670. Does this sound about average  for a 670?

I have gpu core clocked +100MHz and memory at 1550MHz.

My settings are -i 1 -l K7x32  -C 1 -H 1

K14x16 works well but K7x32 seems to be more consistently high.

I use -i 1 because 0 makes my PC stutter too much. I need that extra power for general usage.

Using -auto always results in a statement saying the results would yield something like 300khash+ but when I run the settings it suggests, I just get 100khash or so.

Also, does memory amount on gpu matter? Cudaminer is only using half of the 2GB available. How do I adjust the memory usage? I thought more memory helps mining.

I think no one has tested this yet.

But the lookup gap makes things much less critical now on cards with lesser memory.

Christian