Speed isn’t Everything !

[RealEnder]

Hash-IT, I thought myself about better dictionaries in the world of GPGPU based rule engines and wrote tool to collect words from various sources: wlc The last version can parse wikipedia xml.bz2 dumps, but it is a bit long process due the size. You can download some preprocessed dicts from DWPA site. Keep in mind, that all dicts there are unique as a whole.

[scriptx]

Wordlists are great, but not guaranteed. Bruteforce is guaranteed (ignoring max pass length issue with hashcat), but obviously very time consuming.

correct, but only theoretically. to crack a long password it will take so much time (millions of years) that you can not say its guaranteed to be cracked.

I love maskprocessor because it allows you to be creative, however if you're piping output into hashcat or something like pyrit, you may be at a loss for an ETA or current progress. Knowing the math ahead of time is a huge help.

Let's take all possible lowercase possibilities that are 8 characters in length:

./mp64.bin ?l?l?l?l?l?l?l?l

The output here, as previously discussed, is 26^8 = 208,827,064,576 words -- ouch! Even at my measly 27900 PMK/s (pyrit) or 19980 c/s (hashcat), that will take well over 12 weeks!

you can still calculate it easily: http://hashcat.net/wiki/doku.php?id=comb...nt_formula

(Not sure why pyrit outperforms hc for me, but it does).

The only card where this can happen is ATI/AMD of the 4xxx series. the 4xxx series did not support access to fast memory for the OpenCL interface, only for CAL. Since pyrit uses CAL it is faster on 4xxx cards. That changed with 5xxx and higher.

So, being dedicated, and having exhausted perhaps any other rule-based strategies or dictionary attacks, you can break maskprocessor output into chunks in different ways:

./mp64.bin a?l?l?l?l?l?l?l = 26^7 = 8,031,810,176. yes, you'd have to keep track of, and increment the first letter, but at the same 27900 PMK/s, I am looking at 3.3 days, after which I can review and move on. This affords your PC bathroom breaks, and makes it easier to not lose your place in the process if a hardware or power failure occurs. This can also be helpful for distributing the same effort across multiple machines.

exactly. also see here: http://hashcat.net/wiki/doku.php?id=dist...oclhashcat

Additionally, the -s and -l flags can be used to stop and start at certain words, or "checkpoints", if you wish. Set yourself a pace, knowing your tries/sec.

yeah its an unique feature of maskprocessor

[scriptx]

The only card where this can happen is ATI/AMD of the 4xxx series. the 4xxx series did not support access to fast memory for the OpenCL interface, only for CAL. Since pyrit uses CAL it is faster on 4xxx cards. That changed with 5xxx and higher.

Not to hijack a thread but I'd love to then figure out what's limiting my ability!..

cudaHashcat picks up both my geforce cards:

Device #1: GeForce GTX 560 Ti, 1023MB, 1670Mhz, 8MCU
Device #2: GeForce GT 430, 1023MB, 1400Mhz, 2MCU

Using the -d flag, I can successfully isolate the process to use just the 560Ti (the better of the two). I have also played with -n flag, however any value above 8 does not yield an increase in computations per second.

./cudaHashcat-plus64.bin -o result -d 1 -n 8 -m 2500 belkin54g-handshake.hccap cow.txt
cudaHashcat-plus v0.06 by atom starting...

Hashes: 1
Unique salts: 1
Unique digests: 1
Bitmaps: 8 bits, 256 entries, 0x000000ff mask, 1024 bytes
Rules: 1
GPU-Loops: 128
GPU-Accel: 8
Password lengths range: 8 - 15
Platform: NVidia compatible platform found
Watchdog: Temperature limit set to 90c
Device #1: GeForce GTX 560 Ti, 1023MB, 1670Mhz, 8MCU
Device #2: skipped by user
Device #1: Allocating 9MB host-memory
Device #1: Kernel ./kernels/4318/m2500.sm_21.64.cubin

Scanning dictionary cow.txt: 1047582 bytes (10.56%)
Scanned dictionary cow.txt: 9924323 bytes, 995755 words, starting attack...


Status.......: Exhausted
Input.Mode...: File (cow.txt)
Hash.Target..: belkin54g
Hash.Type....: WPA/WPA2
Time.Running.: 45 secs
Time.Left....: 0 secs
Time.Util....: 45584.7ms/240.0ms Real/CPU, 0.5% idle
Speed........: 21766 c/s Real, 20086 c/s GPU
Recovered....: 0/1 Digests, 0/1 Salts
Progress.....: 995755/995755 (100.00%)
Rejected.....: 3553/995755 (0.36%)
HW.Monitor.#1: 0% GPU, 62c Temp

Started: Thu Oct 20 11:34:24 2011
Stopped: Thu Oct 20 11:35:10 2011

If I do not specify -d nor -n, I'll get:
Speed........: 19276 c/s Real, 18214 c/s GPU

[scriptx]

I was thinking perhaps because I was using cudaHashCat vs ocl, though when trying to run oclHashCat, "ERROR: No AMD compatible platform found", the resolution of which is to use cudaHashcat obviously

So is the GeForce GTX 560 Ti just not the powerhouse I was hoping it was?

[atom]

absolutly not. the 460/560 is a low budget card which is also sm_21 which requires vector datatypes to run efficient. bad choice for hashcracking.

[scriptx]

absolutly not. the 460/560 is a low budget card which is also sm_21 which requires vector datatypes to run efficient. bad choice for hashcracking.

Thanks.. had an enlightening conversation in #hashcat yesterday about it. Off to ebay it goes!

For anyone else interested, this is a decent matrix for helping you decide a card intended for hash cracking:

http://golubev.com/gpuest.htm

[atom]

ivans table is a good one, but gateways is also good since it shows "Pwr eff,MH/W":

http://www.gat3way.eu/est.php

if website is down, use google cache:

http://webcache.googleusercontent.com/se...%2Fest.php

[atom]

absolutly not. the 460/560 is a low budget card which is also sm_21 which requires vector datatypes to run efficient. bad choice for hashcracking.

new oclHashcat-lite will have special support for sm_21