Support of LM ?

[Rolf]

Bad news.

Whoa, I'd like some more of those bad news
It's higher than I estimated for my gtx 480, which are clocked at 1.64Ghz shader, so it's gonna be even higher!

What's the trick ?
S-box stuff in shared mem ?

P.S. You pwnt all known CUDA LM crackers - edpr(bitsliced lol) and EGB.

[Bitweasil]

What's the trick ?
S-box stuff in shared mem ?

Yes, pretty much. The code will be open shortly when I clean stuff up, but I'm using Jan's DES implementation from the FRT code, and improved it by moving things from constant memory into shared memory. 4k of shared is used for structures that speed sbox and key generation schedules. There are a few other optimizations I would like to make with regards to the initial permutation or plaintext gen - right now I've got a number of rather inefficient shifts and masks happening that I can mostly eliminate.

[pyr]

Nice job !

On my GTX260 :
1 hash => 84 M/s
13'000 hashes => 74 M/s