pyrit CUDA nvidia Tutorial + Nvidia overclock instructions

[atat@]

Hi

I have Geforce GTX 295 and i7

and I have problem when I update the BackTrack4.

---

Before update:

root@ّ~:~# pyrit benchmark
Pyrit 0.2.3 2008, 2009 Lukas Lueg h++p://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3

Running benchmark for at least 60 seconds...

#1: 'CUDA-Device #1 'GT200'': 8950.3 PMKs/s (Occ. 99.2%; RTT 2.9)
#2: 'CUDA-Device #2 'GT200'': 8958.3 PMKs/s (Occ. 99.1%; RTT 2.6)
#3: 'CPU-Core (SSE2)': 159.7 PMKs/s (Occ. 96.5%; RTT 2.9)
#4: 'CPU-Core (SSE2)': 148.0 PMKs/s (Occ. 97.0%; RTT 2.9)
#5: 'CPU-Core (SSE2)': 126.5 PMKs/s (Occ. 96.4%; RTT 2.9)
#6: 'CPU-Core (SSE2)':158.4 PMKs/s (Occ. 96.1%; RTT 3.0)
#7: 'CPU-Core (SSE2)': 164.1 PMKs/s (Occ. 96.2%; RTT 3.0)
#8: 'CPU-Core (SSE2)': 144.1 PMKs/s (Occ. 96.4%; RTT 2.9)

It is working normal without any problem and I can use my computer when I create the database.

---

After update:

root@~:~# pyrit benchmark
Pyrit 0.2.3   2008, 2009 Lukas Lueg h++p://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3

Running benchmark for at least 60 seconds...

#1: 'CUDA-Device #1 'GT200'': 10922.3 PMKs/s (Occ. 99.2%; RTT 2.9)
#2: 'CUDA-Device #2 'GT200'': 10781.3 PMKs/s (Occ. 99.1%; RTT 2.6)
#3: 'CPU-Core (SSE2)': 383.7 PMKs/s (Occ. 96.5%; RTT 2.9)
#4: 'CPU-Core (SSE2)': 397.0 PMKs/s (Occ. 97.0%; RTT 2.9)
#5: 'CPU-Core (SSE2)': 355.5 PMKs/s (Occ. 96.4%; RTT 2.9)
#6: 'CPU-Core (SSE2)': 351.4 PMKs/s (Occ. 96.1%; RTT 3.0)
#7: 'CPU-Core (SSE2)': 347.1 PMKs/s (Occ. 96.2%; RTT 3.0)
#8: 'CPU-Core (SSE2)': 367.1 PMKs/s (Occ. 96.4%; RTT 2.9)

I can't use my computer when I create the database also it's hang and gave me error like this:

root@~:~# pyrit batch
Pyrit 0.2.3   2008, 2009 Lukas Lueg h++p://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3

Working on ESSID 'NETGEAR'
Segmentation faultPMKs so far; 23491 PMKs per second; 125189 passwords buffered.

I saw different in the CPU-Core speed between before and after update also maybe it is the main reason for hanging.


notes:

1- I use NVIDIA-Linux-x86-180.22-pkg1 driver before update and
I use NVIDIA-Linux-x86-190.22-pkg1 driver after update but the pyrit doesn't compatible with it, after that I returned back the driver to NVIDIA-Linux-x86-180.22-pkg1.

2- The speed before update was around 18000 PMKs/s and after update was around 23500 PMKs/s.

Is there any one has a solution to fix my problem in the hang and interrupt batch??

Thank you and sorry for my language

[imported_vvpalin]

Passthrough, creating databases is only quicker if you have caps from multiple ap's with the same SSID



PureH@ates' airolib creation benchmark was achieved using one 8800gt GPU, I'm pretty sure you could buy more than one of them with £1000. You'd get roughly the same speed with passthrough method



GPU's are getting faster and they seem to be doing about 12 PPS per GFlop. Given the GT300's are meant to hit 3TFlops, they could be doing as many as 36,000 PPS, maybe more with MIMD??

Also, there's that scientist at NVIDIA who's predicting 20GFlops per GPU by 2015! That'll do your 20Billion passwords in about 1 day on a single GPU, knowing NVIDIA, they probably stick more than one on a card

Then there's the untapped potential of gaming consoles, PS3's have been used in supercomputer clusters for a while now, some people are saying they could do about 2TFlop each? Pyrit can use OpenCL and aren't PS3's able to use OpenCL? anyone benched a PS3?

edit: Apparently PS3's are only capable of about 160Gflops, there goes that idea

shhh your getting me weak in the knees ..

Its always so funny when i hear people say something like "it will take 300 years to crack that given so and so's method using over a hundred computers"

Everyone always seems to forget moors law.

[benjsh]

Hi
Anyone know performance of the ATI Radeon HD 5970 ?

[Neo013]

thank for the script!!

i will try it....

[snowmanwithahat]

Just wanted to say thanks for the great tutorial. I had to do a bit of troubleshooting to get it to work because of differences in the newest pyrit version and the one you used... but it's pretty manageable.

I really appreciate the tutorial.

Also, for comparison, in case anyone was really wondering whether it was 'worth it' or not.... It certainly is

473986 passphrases tested in 2.68 seconds:  177180.04 passphrases/second

That's cowpatty output with the pyrit hash-table.

...and without it...

1956 passphrases tested in 13.78 seconds:  141.97 passphrases/second

Also, for those of you interested in how the latest version of pyrit is performing in comparison to cpus

root@bt:~# pyrit benchmark
Pyrit 0.2.5-dev (C) 2008, 2009 Lukas Lueg hxxp://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3

Connecting to storage... connected

Running benchmark for about 2 seconds... /

Computed 17212.70 PMKs/s total.
#1: 'CUDA-Device #1 'GeForce GTX 285'': 15265.2 PMKs/s (Occ. 95.8%; RTT 2.5)
#2: 'CPU-Core (SSE2)': 884.4 PMKs/s (Occ. 100.0%; RTT 3.0)
#3: 'CPU-Core (SSE2)': 893.0 PMKs/s (Occ. 95.5%; RTT 3.0)

That's with a 3.4ghz Q6600 and a GTX 285. Truly amazing results. I can't wait to put this to test a bit more '
Once again i really appreciate the write-up. It's helped a lot.

[snowmanwithahat]

Okay.. this takes a little longer than the cowpatty tables due to sql stuff but the advantage to airolib-ng is it can use HUGE files while I think cowpatty fails with hash files over 2 gigs.

I've found that this happens when it's compiled with a certain flag set..... To fix it go into the makefile and make the suggested change



Makefile

LDLIBS		= -lpcap
CFLAGS		= -pipe -Wall -DOPENSSL 
#CFLAGS		+= -O2    This needs to be commented out
LDLIBS		+= -lcrypto
CFLAGS		+= -g3 -ggdb
#CFLAGS		+= -static
PROGOBJ		= md5.o sha1.o utils.o cowpatty.o genpmk.o
PROG		= cowpatty genpmk
BINDIR		= /usr/local/bin

That should solve the buffer overflow problem... sorry if its already been posted but it's been driving me insane... it literally took 3 hours ouputting in an aircrack format and then I found how to fix cowpatty to take larger files... It's much faster and definetly worth it. Just download the tarball and un-tar, edit the makefile, remake, and then copy it into /pentest/wireless/cowpatty/ hope this helped