blazer wrote:During the chain generation is it possible to implement reversing and not carry out the full rounds, then reverse the skipped steps when searching for the hash. Will this screw up the distribution of the hashes and will the gains if any be worth it?
Using the GRT reduction function you can get away with these reversals.
For MD5 you can reverse:
- 11 out of 64 steps for 8-8 or less
- 3 out of 64 steps for 12-12 or less
- 2 out of 64 steps for 16-16 or less
For NTLM you can reverse:
- 10 out of 48 steps for 6-6 or less
- 6 out of 48 steps for 8-8 or less
- 5 out of 48 steps for 12-12 or less
- 4 out of 48 steps for 14-14 or less
- 2 out of 48 steps for 16-16 or less
For MD4 you can reverse:
- 10 out of 48 steps for 8-8 or less
- 9 out of 48 steps for 12-12 or less
- 4 out of 48 steps for 16-16 or less
GTX295 gets 705 Mmd5s/sec per core on BarsWF (estimating about 528.75 M full md5s/sec)
GTX295 gets 436 Mlinks/sec per core on GRT (don't know if this is for 6-6, 7-7 or other)
BarsWF 1.8913 ns/full md5 (1 / 0.705 * 64 / 48)
GRT 2.2936 ns/link (0.4023 ns for the reduction function [apples painted orange - oranges = usable info])
So for MD5 8-8 or less you'll get 508 Mlinks/sec on GRT. So my guess is 16.5% faster (note that 20.8% is the very max (64/53-1)).
blazer wrote:Oh also, how many bits am i suppose to specify, seems to give me an error when i do it i'm just picking random numbers, though seems they still can crack without the index.
This will get you an average of 16 to 32 chains per index. Max indexedBits is 32 bits.
indexedBits = ceiling(log2(chainCount / 32))
-------
I used reversing for MD5 when I wrote my reduction function and got 87.6 Mlinks/sec on a Core2Quad 2.5 GHz in 32 bit mode.
http://www.freerainbowtables.com/phpBB3 ... html#p5758haha "Good news, the rt calculator is almost done." -me posted 13 Aug 2008
I over complicated that thing and never finished it.
Core2Quad 2.5 GHz
BarsWF 166 Mmd5s/sec (6.9026 ns/adjusted md5 [1 / 0.166 * 55 / 48])
My rt reduction function 87.6 Mlinks/sec (11.4155 ns/link)
39.5% of the time is spent on my reduction function.
17.5% of the time is spent on GRT's reduction function.
As you can see here GRT's reduction function is faster granted you shouldn't really compare these two situations as they are so different, but it gets the expected answer.