Starting point/hash

[kevinkk5]

is there a possibility to start cracking a hash from a point that i can choose?
i think if i crack a hash the multiforcer starts each time from the same password and goes on right?
so for example he starts everytime with aaaa then aaab or something like that.

so is it possible to chosse for example zzbb as the beginning point?

that would be great because my multiforcer stops very often during a cracksession with lentgh8 and charsetupper-lower-numeric-symbol with a hashlist of 20k hashes.

[Bitweasil]

Which version are you using?

Currently, there is not a way to do this. What error are you getting on failure, what GPU, and what thread/block count?

[kevinkk5]

I´m using version 0.72 and win7 64-bit.

Phenom II X6 1055T with 3.1 GHz
8GB DDR3-Ram
MSI Geforce 460 GTX Hawk

I´m running the multiforcer always with threads=768 and blocks=7

Acutally i don´t know how the errormessage was and it´s not easy to reproduce..
that was when i cracked 19k MD5 hashes with charset lower-upper-numeric-symbol and lenght 9 i think.
after a longer time my screen got black and a few seconds afterwards i could see everything again and
there was an error.. win7 told me that the nvidia driver broke down and that it restarted the driver.
but my GPU isn´t OC.... it has 780MHz standard and that was never a problem. at the moment i´m cracking
19k hashes with charset numeric+space and length 13. it takes hours but no problems.. heat is also unimportant
because my system has always the following temps: cpu: max 46° , motherboard: max 41°, GPU: max 61°

so it should be a problem with either my software or the program when the cracking is more complicated.
I´m sorry that i can´t reproduce the error at the moment. when i achieved that i will tell you.


so does the program run everytime the same order of symbols? so does ist start everytime with aaa or is that
kind of random?

[Bitweasil]

I´m using version 0.72 and win7 64-bit.

Hm. I really need to get the new versions out for Windows.

I´m running the multiforcer always with threads=768 and blocks=7

You'll probably find better performance with 120 or 240 blocks. The 460 just isn't a good compute card, though. For those hash lists, you'll want to use the '-l' switch as well.

after a longer time my screen got black and a few seconds afterwards i could see everything again and
there was an error.. win7 told me that the nvidia driver broke down and that it restarted the driver.

That sounds like a kernel timeout. I'm aware of some glitches that can cause this in my newer code with some screwball conditions, but in 0.72... that's likely a hardware glitch of some variety. That code has been quite stable for me, though I run it on headless Linux boxes.

so does the program run everytime the same order of symbols? so does ist start everytime with aaa or is that
kind of random?

It runs the same sequence every time. There's currently no way to "start in the middle." I'm working on that for the new codebase, but it's still not Windows friendly yet.

[kevinkk5]

a new version for windows would be great because i´m doing all my work on windows and cracking at the same time.. of course i have linux as well and like it but i´m doing almost nothing in linux.

ok could be possible that it´s not a problem with your software.. shouldn´t be a criticism.

why is the 460 so bad in computing?

and actually i get in none configuration more hashes/second than with 768 threads and 7 blocks..
my speed with that hashlist is with charset loweruppernumeric and length 9 about 470M/s and with 256 threads and 120 blocks only 380M/s !
i tried a lot but i couldn´t find a faster solution than the one i use already..
and i´m using --lookup of course.. i just forgot to write that. without it i would have onl 95M/s^^ a bit low..
so according to my experience and several tries to optimize the speed i can say that with my system there is no faster option than 768 with 7 blocks.
how much cpu usage takes that software with your system? and how fast is it? i took the uncracked md5 hashes from freerainbowtables.com

[Bitweasil]

Huh. Ok, I've always had more luck with higher block counts. Guess your system prefers fewer.

It's certainly possible it's a bug in my code. What you described is typically caused by the code requesting more work per kernel instantiation than the GPU can do in ~5s. Given that it normally runs 50-100ms kernels, it would take a serious change to trigger that type of bug. The code self-adapts to load, so even if the GPU starts getting used for gaming, it should auto-adapt it's work to keep things sane. I've seen that type of error in development code, but that's due to some other changes I'm making that make my existing adapting code less useful.

The 460 just has an odd architecture. It's a very video-focused card, not a compute-focused card. Look at the Anandtech article on the GF104 for why.

The CPU isn't used much. It just preps work for the GPU. It may be defaulting to spin locks (100% CPU usage on one core), but I've changed that in newer versions.

[kevinkk5]

ah ok.. i also played around a bit with -m but it had no effect at all. only a few M/s more.. with m=5000 i got a driver break without error message from the program.
and now i see that the 460 is not the best for computation.. i can even watch a video while craking and the hashrate doesn´t drop much. in GPU-Z there is the Video Engine Load that is of course totally unused during cracking and although the whole system grafic is a bit slow a video works perfect^^
with my system the program needs 18% of my cpu.. don´t know whether this is normal.

did you think about a skype chatroom in addition to this board or don´t you like that?
when do you plan to bring out (even if it would be a bit limited) the new version for win7?

ah and i wrote something wrong above. i wasn´t cracking 19k MD5 hashes it was 190k hashes!

[Bitweasil]

ah ok.. i also played around a bit with -m but it had no effect at all. only a few M/s more.. with m=5000 i got a driver break without error message from the program.

To be expected. Windows terminates programs that violate the GPU timer with extreme prejudice. It doesn't give them a chance to error out nicely.

Also, the Vista/Win7 driver model kind of sucks for GPU execution of code. I wouldn't be surprised if you saw a several percent or more gain under Linux.

did you think about a skype chatroom in addition to this board or don´t you like that?
when do you plan to bring out (even if it would be a bit limited) the new version for win7?

ah and i wrote something wrong above. i wasn´t cracking 19k MD5 hashes it was 190k hashes!

There's an IRC channel. #cryptohaze on freenode. That's as "interactive" as I intend to get.

The new version uses pthreads & boost, both of which are just a pain in the ass to compile for Windows. And 190k hashes isn't that huge. I believe I've thrown 5M+ at it before.

Donations to fund additional development hardware are always welcome if you find the tools useful. Part of the delay on the Windows development is that I have to reboot my primary dev box into Windows if I want to do anything - it would be much nicer to have two systems.

[kevinkk5]

yes i understand you totally. programming for windows is horrible in that case.

where should i get a 5M hashlist?

maybe i´ll donate at some time when i earn enough money...

[Bitweasil]

where should i get a 5M hashlist?

Well, I got mine from a PHP script I wrote to stress test my tools I don't actually do hash cracking, I just enjoy writing tools.