Cryptohaze.com

I have GTX260...
can anybody tell the best parameter values for this GPU? im interested in those:
--ms --threads --lookup --blocks

and another question:
what for is the parameter --lookup and what is the proper syntax for command parameter -l / --lookup
because program wont run with parameter -l 768

Is that the GTX 260 Core 216 or the GTX 260 "Core 192"
192 / 8 = 24 MPs
216 / 8 = 27 MPs
Blocks should be N * number of MPs.
Threads should be M * 32? >= 128?
I'd try N = 1, 2, 3, 4 with M = 4, 5, 6, 7, 8, 10, 12, 14, 16, 18, 20, 22, 24, maybe higher
And see which one is best. I find that with N = 1 and M = kinda high gets the best results.

-l / --lookup is for large lists of hashes it uses a 512 MiB lookup table. Hmm it looks like the small hash list code was removed (because the large list code is always faster) so it's always a good idea to use "-l". You have 896 MiB of memory so you can actually use this. It is just a flag there is no value for it just "-l" not "-l 768". I made a patch for I think 0.7 that let you select a smaller lookup table for those with less memory like me 256 MiB . So it was "-l 30" (30 bits 2 ^ 30 / 8) for a 128 MiB lookup table.

I've found on Fermis that either --threads 512 or --threads 768 works best. Try 768, if you get an "out of resources" error go down to 512.

I've also found a block count of 240 or 360 seems to lead to very good performance. I won't claim "best best," but I'll say very good.

And as for -m, larger is better. If you have a GUI you need responsive, 20-50 (ms) is reasonable. If you're on a headless server or don't care about the GUI, 200-500 is better. There are no real gains beyond that point.

As Sc00bz said, '-l' is just a flag - no value after it. I intend to write some code that autotunes the bitmap sizes, I just haven't gotten to it yet.

if you are using windows blocks=number of cores that cuda-multiforcer shows/8
was the best for me with my GTX460