Hey, when I checked my mobile phone this morning I noticed that an nVidia engineer has submitted an optimized kernel for Kepler devices. Apparently they are aware of the whole AMD vs nVidia  mining discrepancy and want to help me put nVidia into a better position.

I will review their code submission and integrate it if it's better than my code (which is likely, considering they designed this silicon). They don't include scrypt-jane yet, so I will have to do that part myself.

Christian

Doesn't matter

@ 15 cents per kwh you are spending 3.6$/day on electricity
@ 30 kh/s you are earning ~30 cents a day.
Net loss 3.3$/day

Can you explain what you did, I have the same problem and which ddl's were responsible ?
Thanks

You might want to check if you're using x64 dll's in the executable folder

X64 versions neither work for me, but I'm happy with the x32 one.

The lookup gap has turned my 10 kHash/s 450 Watts Yacoin mining rig into a devilish 14 kHash/s 666 Watts mining rig. Not quite as high as I had hoped for, but the new Wattage is nice.

I run GTX 780 with -L 6 -l 12x32    up to 3.65 kHash/s
and GTX 780Ti with -L 6 -l 15x32   up to 4.7 kHash/s

still quite an easy to remember formula with a decent performance. There may be better values but that is what I found within an hour of tinkering.

Christian

I managed to compile mine on x86 and ran it on my two GTX660 using :

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

but I am getting only :

[2014-01-19 17:07:10] GPU #1:    2.14 khash/s with configuration K19x3
[2014-01-19 17:07:10] GPU #1: using launch configuration K19x3
[2014-01-19 17:07:11] GPU #1: GeForce GTX 660, 0.80 khash/s
[2014-01-19 17:07:22] GPU #1: GeForce GTX 660, 1.53 khash/s
[2014-01-19 17:07:22] accepted: 1/1 (100.00%), 1.53 khash/s (yay!!!)
[2014-01-19 17:07:58] Stratum detected new block
[2014-01-19 17:07:59] GPU #1: GeForce GTX 660, 1.59 khash/s

and

[2014-01-19 17:08:14] GPU #0: GeForce GTX 660, 1.74 khash/s
[2014-01-19 17:08:14] accepted: 6/6 (100.00%), 1.74 khash/s (yay!!!)
[2014-01-19 17:08:18] GPU #0: GeForce GTX 660, 1.71 khash/s
[2014-01-19 17:08:18] accepted: 7/7 (100.00%), 1.71 khash/s (yay!!!)
[2014-01-19 17:08:22] GPU #0: GeForce GTX 660, 1.69 khash/s
[2014-01-19 17:08:22] accepted: 8/8 (100.00%), 1.69 khash/s (yay!!!)
[2014-01-19 17:08:28] GPU #0: GeForce GTX 660, 1.72 khash/s
[2014-01-19 17:08:28] accepted: 9/9 (100.00%), 1.72 khash/s (yay!!!)
[2014-01-19 17:08:35] GPU #0: GeForce GTX 660, 1.75 khash/s
[2014-01-19 17:08:35] accepted: 10/10 (100.00%), 1.75 khash/s (yay!!!)

Any recommended -l setting for GTX 660 ?

Thank you

I also see the Windows version crashing during autotune now. I will investigate.

Christian

You lost me at "Select which version you want to compile, and compile it." What "version" do I select where? And by compile one usually means hitting that green debug arrow right? =)
Actually I think I might have lost it at the extract part already: What should the final main folder look like? .\ with 6 folders ( CudaMiner-master, curl-7.29.0 ..., pthreads)? And if you open the miner-master folder you end up in the folder with code and compat?
Little more dummie style explanation would be appreciated =)

if you get a higher kHash/s then yes...

GTX 780 is a Compute 3.5 part. I haven't finished the lookup-gap for that kernel yet.

I expect the higher end devices like 660Ti, 760, 770, 780, 780Ti to benefit from the lookup gap.

Also the lower end cards with 1GB (e.g. GT 640 GK208 with 1 GB DDR5 memory)

depends entirely on the card. cannot generalize here, sorry.

I will install Cuda 5.5 when I get home and test again, but running individually didn't work ok for me so far.

Any tutorial on how to compile the latest source code from github ? I haven't done any C programming in ages, I am very rusty.

Quite easy:
Get Visual Studio Pro 2010 (Trial if you must)
Get Cuda SDK 5.5
Get the source code from github (google cudaminer)
Exctract source code in it's own folder
Get libraries from first post by Christian (almost 50mb)
Extract libraries to the same folder that the source code folder is in, so not to the source code folder but one up.
Open the .sln file in the source code with Visual Studio.
Select which version you want to compile, and compile it.
If it errors:

There was a breaking change today regarding the format of launch configs for David Andersen's kernels.

This has advantages because of more fine grained control and memory allocation. It is however a nightmare
for maintainers of the Google spreadsheets


this affects K,T,X kernel configs only (these are derived from David's code) - and only when you run a
github version from today or later.

or you can simply autotune again to find a good config, saving you the hassle of converting...

The main advantage to this is that users of 1GB cards can now use up to 10% more memory
than before (memory is now allocated in increments of 32MB for Yacoin - previously it was 128MB).

Edit: The first scrypt ASIC's are out:
(360kH/s for ~500 USD?)
Go scrypt-jane!

Go to your Microsoft Visual Studio 10.0\VC\bin folder and rename or relocate cvtres.exe and do the same with another cvtres.exe under Microsoft Visual Studio 10.0\VC\bin\amd64, then run Visual Studio 2010 and try compiling.

If it doesn't work, just change everything back.

The best case is that it achieves exactly half the hash rate as with N=14. Why? because for N=15 it's exactly twice the amount of work. This will affect CPUs likewise.

The usual case for most GPU models will be that the performance degrades by more than half, because the occupancy of the CUDA cores goes down (too many cores for fewer hashes to be computed simultaneously given the given available memory). Cards with 1GB and 2GB will be hit the hardest...

Cards with 4GB should barely see an impact for N = 14 --> 15

For cards with lots spare GPU cores (say, a GTX 780 or better) we can cut memory requirements in half and increase compute requirements instead (LOOKUP_GAP). It's on my TODO list.