aircrack-ng-cuda + WPA(2)-PSK hacking with backtrack 4beta

[fnord0]

my first HOWTO for this site (which this site I so love, thanks for everything backtrack team!! im loving the bt4beta, but sad about switch to ubuntu - make no mistake about it, I think this is a amazing distro! i am running it as my main distro on my linux partition, for all sorts of learning and lab testing! bravo... one question: is there ever going to be a backtrack based off slackware/slax again? x fingers crossed x)

don't be cruel, spelling is not my forte'. *use these directions at yr own risk, only for information, test on yr OWN hardware*

the speed-up afforded by an nvidia GTX 285 compared to a Q6600 overclocked intel proc is quite satisfying to say the least

lets get on with things. download the latest version of aircrack-ng-cuda via SVN/subversion (btw, I love the name of this tool!).
the default BT4beta off the DVD will NOT compile aircrack-ng-cuda properly - you need to make some changes to the system, which I will be detailing here. ::

cd /usr/local/src [[ or whatever temp directory you want to use ]]
svn co svn - Revision 1578: /branch/aircrack-ng-cuda aircrack-ng-cuda
[[ the version I checked out was = resvision 1558 ]]

you now have a new directory called "aircrack-ng-cuda" in your "temp" directory, which we will be working with shortly...

onto CUDA, and all pre-req's. reviewing the README.CUDA file, we find out the following ::

Quick Notes for the CUDA version of aircrack-ng:
-> Only works with wpa cracking
-> Tested on a 1.0 Compute Card using CUDA_SDK 2.2 (LINUX)
-> No SLI Support as yet (I have no way of testing currently)

Preq:
-> CUDA SDK 2.2 Installed with ENV vars (see cuda install guide for more details)
* PATH to contain /usr/local/cuda/bin
* LD_LIBRARY_PATH to contain /usr/local/cuda/cudaprof/bin:/usr/local/cuda/lib
* CUDA_INSTALL_PATH to contain /usr/local/cuda
(or where ever you installed it)

Building CUDA aircrack-ng:
make clean
CUDA=true make

easy enough, just follow the above to the T and building aircrack-ng-cuda will be a breeze (almost... for my computer at least).

here we go - line by line...

-> CUDA SDK 2.2 Installed with ENV vars

get CUDA SDK 2.2 from... who else? nvidia @ Download CUDA Code - complete and free toolkit for creating derivative works

from there CUDA download page, we see the following ::

NVIDIA’s CUDA development tools provide three key components to help you get started: the latest CUDA driver, a complete CUDA Toolkit, and CUDA SDK code samples.

Please download files and install them in this order:

1. CUDA Driver 2. CUDA Toolkit 3. CUDA SDK code samples

so, do as they say (I think #3 can be held off on... I wasn't able to get much in package #3 to compile, but I still downloaded it) :

since this is a BACKTRACK 4beta HOWTO, we are going set the following in the nvidia download page options.
Operating System: -> linux 32bit
Linux Version: -> Ubuntu 8.04

CUDA Driver: NVIDIA Driver for Linux (Ubuntu 8.04) 185.18.08 Beta --> Thank You

• downloads a file called "cudadriver_2.2_linux_32_185.18.08-beta.run" @ 22.6MB
• best to be @ a shell prompt, make sure to kill all xorg sessions (afaik)
• this installs the CUDA-compatible "NVIDIA Accelerated Graphics Driver for Linux-x86 185.18.08". this is the linux nvidia driver/module, this app also can modify the /etc/X11/xorg.conf file (you are NOT required to - if you DO choose to let it modify the xorg.conf file, it will back up the current file, which is helpful)

CUDA Toolkit: CUDA Toolkit 2.2 for Linux (Ubuntu 8.04) --> Thank You

• downloads a file called "cudatoolkit_2.2_linux_32_ubuntu8.04.run" @ 28.2MB
• installs all files by default to /usr/local/cuda, this is what I choose
• make sure your PATH includes /usr/local/cuda/bin
• make sure your LD_LIBRARY_PATH includes /usr/local/cuda/lib
  
  or add /usr/local/cuda/lib to /etc/ld.so.conf and run ldconfig as root(I did this too, just to be sure things did work... you apparently aren't required to do it if you to the implement the LD_LIBRARY_PATH as I explain below)

  Identification: NVIDIA CUDA
  Target directory: pkg
  Uncompressed size: 84004 KB
  Compression: gzip
  Date of packaging: Sun Apr 12 12:51:31 PDT 2009

  OPTIONAL are the Release Notes, Programming Guide, Profiler README, License.

CUDA SDK: CUDA SDK 2.2 code samples for Linux (Ubuntu 8.04) --> Thank You

• downloads a file called "cuda/2_2/sdk/cudasdk_2.2_linux.run" @ 55.9MB
• installs the NVIDIA CUDA SKD v2.2 for linux/ubuntu 8.04
• installs all files by default into ~/NVIDIA_CUDA_SDK
• asks for the install path of CUDA, which if follow my directions above, it gets installed to /usr/local/cuda
• make sure your PATH includes /usr/local/cuda/bin
• make sure your LD_LIBRARY_PATH includes /usr/local/cuda/lib
  

  Identification: NVIDIA CUDA SDK
  Target directory: pkg
  Uncompressed size: 151108 KB
  Compression: gzip
  Date of packaging: Wed Apr 22 13:56:52 PDT 2009

  OPTIONAL are the Release Notes, Getting Started, License

CUDA Debugger: CUDA Debugger 2.2 for Linux --> Thank You

• downloads a file called "cudagdb_2.2_linux_32_rhel5.3.run" @ 1.18MB
• installs all files by default to /usr/local/cuda, this is what I choose
• make sure your PATH includes /usr/local/cuda/bin
• make sure your LD_LIBRARY_PATH includes /usr/local/cuda/lib
  
  or add /usr/local/cuda/lib to /etc/ld.so.conf and run ldconfig as root(I did this too, just to be sure things did work... you apparently aren't required to do it if you to the implement the LD_LIBRARY_PATH as I explain below)

  Identification: NVIDIA CUDA
  arget directory: pkg
  Uncompressed size: 2580 KB
  Compression: gzip
  Date of packaging: Sun Apr 12 10:27:23 PDT 2009

there were a few things above that said we needed to create links/environment variables pointing to the CUDA libraries, here is an exact replica of the top of my /root/.bashrc file (substitue /root with your user account's home dir - ex. "/home/myuser" if you are building/using everything with a normal user)

PATH=/usr/local/bin:/usr/local/sbin:/usr/kde3/bin:/usr/local/cuda/bin:/usr/bin:$PATH
LD_LIBRARY_PATH=/usr/local/cuda/cudaprof/bin:/usr/local/cuda/lib
CUDA_INSTALL_PATH=/usr/local/cuda
SDK_INSTALL_PATH=/root/NVIDIA_CUDA_SDK
export PATH
export LD_LIBRARY_PATH
export CUDA_INSTALL_PATH
export SDK_INSTALL_PATH

logout, and log back in -- or just run "source ~/.bashrc" for all changes to propagate to yr bash session (PS: why no tcsh shell in backtrack by default? i love me some tcsh shell! I know, I know... it's not hard to type "apt-get install tcsh")

also, I did update the file /etc/ld.so.conf and put the following at the bottom of the file, then ran ldconfig ::

include /usr/local/cuda/lib

lastly... with the NVIDIA_CUDA_SDK there is some useful utils/projects, to get them working, you first need this ::

apt-get install freeglut3-dev

then, goto ~/NVIDIA_CUDA_SDK and "make" everything... (I still seemed to get an error with the smokeParticles project... no biggie tho as these tools are not required) ::

# cd ~/NVIDIA_CUDA_SDK
# make
# cp -p 
# cp -p ~/NVIDIA_CUDA_SDK/bin/linux/release/* /usr/bin

that should be it for CUDA setup/install. if you want, you can compile the CUDA SDK projects
now, back to aircrack-cuda-ng.

cd /$TEMP_DIR/aircrack-ng-cuda
[[ $TEMP_DIR being where you downloaded aircrack-ng-cuda to ]]
make clean
make CUDA=true sqlite=true unstable=true install

yr GOOD TO GO! any errors you get here, check your environment variables!

example way to check env variables: "set | grep -i LD_LIBRARY_PATH" or "set | grep -i CUDA_INSTALL_PATH" etc.

test to ensure everything is working, run aircrack on the test wpa-psk capture file, with the included wordlist ::

cd src
./aircrack-ng -p 1 ../test/wpa.cap -w ../test/password.lst

NOTE: Multiple threads will just queue tasks for the gfx card. (-p 1)

[fnord0]

------------------------------------------------------------------------------------------------------------------------------
-- from the README.CUDA file, we have a Performance Review (aircrack-ng GPU vs. CPU)
------------------------------------------------------------------------------------------------------------------------------

GPU:
src # /root/NVIDIA_CUDA_SDK/bin/linux/release/deviceQuery
CUDA Device Query (Runtime API) version (CUDART static linking)
There is 1 device supporting CUDA

Device 0: "GeForce 8800 GTS 512"
CUDA Capability Major revision number: 1
CUDA Capability Minor revision number: 1
Total amount of global memory: 536150016 bytes
Number of multiprocessors: 16
Number of cores: 128
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes
Total number of registers available per block: 8192
Warp size: 32
Maximum number of threads per block: 512
Maximum sizes of each dimension of a block: 512 x 512 x 64
Maximum sizes of each dimension of a grid: 65535 x 65535 x 1
Maximum memory pitch: 262144 bytes
Texture alignment: 256 bytes
Clock rate: 1.62 GHz
Concurrent copy and execution: Yes
Run time limit on kernels: No
Integrated: No
Support host page-locked memory mapping: No
Compute mode: Default (multiple host threads can use this device simultaneously)

Test PASSED

----
GPU Test:

src # time ./aircrack-ng -p 1 ../test/wpa.cap -w ~/words

real 1m12.439s
user 0m29.990s
sys 0m42.370s

keys: 176984 keys

k/p = 2458.11
----


CPU: (DUAL CORE)
src # cat /proc/cpuinfo
processor : 0 & 1
vendor_id : GenuineIntel
cpu family : 15
model : 6
model name : Intel(R) Pentium(R) D CPU 3.00GHz
stepping : 2
cpu MHz : 3014.642
cache size : 2048 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 2
apicid : 0
initial apicid : 0
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 6
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx lm constant_tsc pebs bts pni dtes64 monitor ds_cpl vmx cid cx16 xtpr pdcm lahf_lm tpr_shadow
bogomips : 6029.28
clflush size : 64
power management:

----
CPU (SSE2) Test:

time ./aircrack-ng -p 2 ../test/wpa.cap -w ~/words


real 4m59.881s
user 9m57.630s
sys 0m0.430s

keys: 176984

k/s = 772.85
----

I don't know if the wordilst from these tests is the same that comes with the aircrack-ng-cuda distribution (I suspect NOT, the filenames are different), but here is what it looks like on my EVGA GTX 285 ;;

first, WPA(1) with CUDA

root@bt:~/aircrack-ng-cuda# time src/aircrack-ng -p 1 test/wpa.cap -w test/password.lst

real 0m3.109s
user 0m1.152s
sys 0m1.692s

19200 keys tested (6841.29 k/s)

second, WPA2 with CUDA

root@bt:~/aircrack-ng-cuda# time src/aircrack-ng -p 1 test/wpa2.eapol.cap -w test/password.lst
real 0m3.108s
user 0m1.140s
sys 0m1.700s

19200 keys tested (6837.44 k/s)

using aircrack-ng-cuda with airolib-ng databases to crack WPA(1 and 2) keys (this test is just to see if it works, which it will if everything went smoothly - for some reason it doesn't go that fast, but the key is obtained within 1second!) ::

root@bt:~/aircrack-ng-cuda# aircrack-ng -r test/passphrases.db test/wpa.cap

this is just the tip of the iceberg... there are lots of good tools available, and sure to be released. there is an excellent tutorial by pureh@te that explains HOWTO use pyrit CUDA nvidia tool, which is very nice!

- greetz to everyone working hard on the backtrack distro, ya`all are my heroes! peace -
fnord0

[kidFromBigD]

Supporting the nVidia cards with a new version on aircrack-ng is great news. Tell me, how did you find out about this? The main page ( Main [Aircrack-ng] ) contains nothing on CUDA support.

Even so, I'd like to see it go head-to-head with the latest SVN of pyrit proper, just to see who has the upper hand at this point.

pureh@te ? You there? Seemingly no SLI support, but can you give us a comparo?

Thanks for the How-To, definitely bookmarked!

[purehate]

Nice. I wasnt aware of the aircrack-cuda release although I had heard of its development. Just to be clear SLI has nothing to do with CUDA, I'm assuming what you mean is multi card support. I wont know untill I test it if it has multi card support or not. Ill check it out tommorow and post some results. In the mean time we have been working hard on pyrit with multi card support in bt4, cowpatty 4.5 including all the cool s**t, ATI support for pyrit and a md5 /md4 / ntlm CUDA-multiforcer tool. Check out our lastest blog post.

Offensive Security Blog

[fnord0]

Tell me, how did you find out about this? The main page .... contains nothing on CUDA support.

subversion my man, SVN. have a glance @ the README.CUDA file for the aircrack-ng-cuda subversion repository = h**p://trac.aircrack-ng.org/svn/branch/aircrack-ng-cuda/README.CUDA
one other worthy mention, is the latest airoscript which is hella pimp too = h**p://trac.aircrack-ng.org/svn/branch/airoscript

Even so, I'd like to see it go head-to-head with the latest SVN of pyrit proper, just to see who has the upper hand at this point.

I'd like to see other people's results, as I just have an EVGA GTX 285, and I am happy with it's performance boost.

Thanks for the How-To, definitely bookmarked!

glad someone got something out of it, I am happy to be of assistance.

Just to be clear SLI has nothing to do with CUDA, I'm assuming what you mean is multi card support.

ahhhh, the hate machine himself... I am a fan! I think you are referring to the link I posted from the README.CUDA file @ h**p://trac.aircrack-ng.org/svn/branch/aircrack-ng-cuda/README.CUDA ::

-> No SLI Support as yet (I have no way of testing currently)

I took the quote directly from the file - I too suspect the aircrack-ng-cuda team is referring to multiple card decoding support.
of note in the aircrack-ng-cuda documentation ::

./aircrack-ng -p 1 ../test/wpa.cap -w ../test/password.lst

NOTE: Multiple threads will just queue tasks for the gfx card. (-p 1)

and this from the "aircrack-ng --help" output ::

-p <nbcpu> : # of CPU to use (default: all CPUs)

everything points to the "-p 1" switch telling aircrack-ng-cuda to queue tasks directly to the (1?) GPU, I am not so sure how you'd tell it to queue to multiple GPUs (maybe "-p 2" will work for this, but you see later in the README.CUDA documenation that "-p 2" is used for CPU testing, even with aircrack-ng-cuda ::

CPU (SSE2) Test:

time ./aircrack-ng -p 2 ../test/wpa.cap -w ~/words


real 4m59.881s
user 9m57.630s
sys 0m0.430s

keys: 176984

k/s = 772.85

I wont know untill I test it if it has multi card support or not. Ill check it out tommorow and post some results.

woot, glad to hear you have access to multi-GPU, I have a 1x EVGA GTX 285, and 1x EVGA 9800 GTX ... I am looking to try leveraging BOTH for CUDA support, yet I know in windows (shudder) that you get the choice of CUDA support on EITHER/OR, but not both --- is it the same in linux?

In the mean time we have been working hard on pyrit with multi card support in bt4, cowpatty 4.5 including all the cool s**t, ATI support for pyrit and a md5 /md4 / ntlm CUDA-multiforcer tool. Check out our lastest blog post.

P-I-M-P ... u guys are the greatest!
btw - I read yr blog post (it has been subscribed to via RSS... holding my breath for bt4 final) = u my good man have access to some kick-ass hardware (re: dual 295 GTX's)! and yeah, that brute forcing NTLM results are mind warping! thanks for everything u guys do.

-peace-
fnord0

[kidFromBigD]

subversion my man, SVN.

Yes, that's where my digging eventually uncovered the CUDA support nuggets.

... = u my good man have access to some kick-ass hardware (re: dual 295 GTX's)! and yeah, that brute forcing NTLM results are mind warping! thanks for everything u guys do.

+1
pureh@te has the pwn sauce. In fact, he's told us about the h@te machine: Triple-SLI GTX-295, 1200W power supply, web interface with queuing, etc. etc. ... In a word: Incredible.

EDIT:

Ill check it out tommorow and post some results.

Hey, if taking a look at this removes you from delivering BT4final, then leave the busywork to us. BT4final is important enough to get right. Don't loose sleep about this... and Thanks!

[Zermelo]

OK, I've been testing out aircrack-ng with CUDA support. I would be interested to see other people's benchmarks because so far mine have been underwhelming and nothing comparable to the processing speed of pyrit.

Here are some screenshots:





I'm running a single 295 GTX on a Q9650 CPU (not overclocked). As you can see, the results are about 3068 k/s using -p 1 (the GPU) and 3143 using -p 2 (the CPU). Now there are a few caveats with the above benchmarks. For some reason the CUDA 185.18.08 beta driver was causing problems on my system. I was getting I/O errors on /dev/nvidia0 and it had nothing to do with permissions, dmesg revealed:

allocation failed : out of vmalloc space - use vmalloc<size> to increase size
NVRM: RmInitAdapter failed! (0x12:0x2b:1577)
NVRM: rm_init_adapter(0) failed

Tried all the solutions google had to offer, inclreasing vmalloc, etc. to no avail, so I reverted back to the 180.22 drivers, and everything worked and compiled fine. I'm currently on Intrepid, but this seems to me like it's a problem with the beta drivers and the 295 gtx since I have the 185.18.08 driver working fine on my Asus laptop with a Geforce 9300M G - also on Intrepid.

Also it appears that during the attack process, aircrack-ng hangs excessively and takes up considerable resources much more than pyrit in any event. Also if you look at the benchmarks given on the readme.cuda file the benchmarks give are also pretty underwhelming:

k/p = 2458.11

The pyrit performance graph shows for a GeForce 8800 GTS 512 to get somewhere around 3200+ k/s.

Since my performance numbers could strictly be a result of my driver issue, I would be interested in seeing anyone else's.

Sorry for the size of the pics. O_o

[fnord0]

OK, I've been testing out aircrack-ng with CUDA support. I would be interested to see other people's benchmarks because so far mine have been underwhelming and nothing comparable to the processing speed of pyrit.

[SNIP]

Since my performance numbers could strictly be a result of my driver issue, I would be interested in seeing anyone else's.

howdy zermelo,

figured I'd post, as I am curious what others are getting out of this tool.. I will post mine, hopfully others will post here too... I have 1x EVGA GTX 285, and 1x EVGA 9800 GTX ::

root@bt:~# /root/NVIDIA_CUDA_SDK/bin/linux/release/deviceQuery
CUDA Device Query (Runtime API) version (CUDART static linking)
There are 2 devices supporting CUDA

Device 0: "GeForce GTX 285"
CUDA Capability Major revision number: 1
CUDA Capability Minor revision number: 3
Total amount of global memory: 1073414144 bytes
Number of multiprocessors: 30
Number of cores: 240
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes
Total number of registers available per block: 16384
Warp size: 32
Maximum number of threads per block: 512
Maximum sizes of each dimension of a block: 512 x 512 x 64
Maximum sizes of each dimension of a grid: 65535 x 65535 x 1
Maximum memory pitch: 262144 bytes
Texture alignment: 256 bytes
Clock rate: 1.48 GHz
Concurrent copy and execution: Yes
Run time limit on kernels: Yes
Integrated: Yes
Support host page-locked memory mapping: Yes
Compute mode: Default (multiple host threads can use this device simultaneously)

Device 1: "GeForce 9800 GTX/9800 GTX+"
CUDA Capability Major revision number: 1
CUDA Capability Minor revision number: 1
Total amount of global memory: 536608768 bytes
Number of multiprocessors: 16
Number of cores: 128
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes
Total number of registers available per block: 8192
Warp size: 32
Maximum number of threads per block: 512
Maximum sizes of each dimension of a block: 512 x 512 x 64
Maximum sizes of each dimension of a grid: 65535 x 65535 x 1
Maximum memory pitch: 262144 bytes
Texture alignment: 256 bytes
Clock rate: 1.67 GHz
Concurrent copy and execution: Yes
Run time limit on kernels: No
Integrated: Yes
Support host page-locked memory mapping: Yes
Compute mode: Default (multiple host threads can use this device simultaneously)

root@bt:~/aircrack-ng-cuda/src# ./aircrack-ng -p 1 ../test/wpa.cap -w ../test/password.lst

Aircrack-ng 1.0 rc3 r1555
[00:00:02] 19200 keys tested (6784.01 k/s)

KEY FOUND! [ biscotte ]

Master Key : CD D7 9A 5A CF B0 70 C7 E9 D1 02 3B 87 02 85 D6
39 E4 30 B3 2F 31 AA 37 AC 82 5A 55 B5 55 24 EE

Transient Key : 33 55 0B FC 4F 24 84 F4 9A 38 B3 D0 89 83 D2 49
73 F9 DE 89 67 A6 6D 2B 8E 46 2C 07 47 6A CE 08
AD FB 65 D6 13 A9 9F 2C 65 E4 A6 08 F2 5A 67 97
D9 6F 76 5B 8C D3 DF 13 2F BC DA 6A 6E D9 62 CD

EAPOL HMAC : 28 A8 C8 95 B7 17 E5 72 27 B6 A7 EE E3 E5 34 45

sooo, not too bad... ~6700+k/s

let's now try with the "-p 2" flag, per the README.CUDA file, this will send it to the CPU....

root@bt:~/aircrack-ng-cuda/src# ./aircrack-ng -p 2 ../test/wpa.cap -w ../test/password.lst

Aircrack-ng 1.0 rc3 r1555
[00:00:05] 38400 keys tested (6765.60 k/s)

KEY FOUND! [ biscotte ]

Master Key : 59 69 14 95 B1 DA 4A 48 C8 3D F8 F8 77 39 24 4C
F0 FE 90 3E 99 91 34 3D EA A2 EB 31 67 C8 E6 09

Transient Key : A7 38 41 57 5D 4D B1 43 4B E6 80 6C 5A E1 86 B4
59 BC 69 57 78 3A CB 4A 0D C2 9D 3D C7 BC 15 13
FE 0C D1 DB F3 C8 E0 AC 05 41 06 74 97 2C 47 B8
BB 81 A2 4A FB A0 D1 AD 75 DC D9 35 82 24 69 A6

EAPOL HMAC : 43 BE F4 6B 1D 7D 38 F2 58 BF A5 AA B1 82 D2 1A

hmmmmm, that is REALLY surprising, as I only have a Q6600 OC'd to 3.22GHz, but whatever (??)

here's my CPU info to confirm this is the CPU i have (btw I compiled my own i686 kernel for backtrack 4beta) ;;

root@bt:~/aircrack-ng-cuda/src# cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 15
model name : Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
stepping : 11
cpu MHz : 3222.033
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 0
cpu cores : 4
apicid : 0
initial apicid : 0

[SNIP]

bogomips : 6444.06
clflush size : 64

processor : 1
vendor_id : GenuineIntel
cpu family : 6
model : 15
model name : Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
stepping : 11
cpu MHz : 3222.033
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 1
cpu cores : 4
apicid : 1
initial apicid : 1

[SNIP]

bogomips : 6444.26
clflush size : 64

processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 15
model name : Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
stepping : 11
cpu MHz : 3222.033
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 3
cpu cores : 4
apicid : 3
initial apicid : 3

[SNIP]

bogomips : 6444.34
clflush size : 64

processor : 3
vendor_id : GenuineIntel
cpu family : 6
model : 15
model name : Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
stepping : 11
cpu MHz : 3222.033
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 2
cpu cores : 4
apicid : 2
initial apicid : 2

[SNIP]

bogomips : 6444.28
clflush size : 64

so here's the kicker! of note/interest below here is the = "-p 4" using that I am getting ~11,000k/s! I thought the "-p 1" switch meant to use the GPU to process, and "-p 2" or anything higher then 2 would tell it to use CPU to process??... but looking above in the CUDA deviceQuery output, see the part "Compute mode: Default (multiple host threads can use this device simultaneously)" for each of my GPU's... wonder whats up with that??
btw, I am running CUDA 2.2, with nvidia linux driver v180.60
so what'cha think is going on??

root@bt:~/aircrack-ng-cuda/src# ./aircrack-ng -p 4 ../test/wpa.cap -w ../test/password.lst

Aircrack-ng 1.0 rc3 r1555
[00:00:11] 76800 keys tested (11197.03 k/s)

KEY FOUND! [ biscotte ]

Master Key : 59 69 14 95 B1 DA 4A 48 C8 3D F8 F8 77 39 24 4C
F0 FE 90 3E 99 91 34 3D EA A2 EB 31 67 C8 E6 09

Transient Key : A7 38 41 57 5D 4D B1 43 4B E6 80 6C 5A E1 86 B4
59 BC 69 57 78 3A CB 4A 0D C2 9D 3D C7 BC 15 13
FE 0C D1 DB F3 C8 E0 AC 05 41 06 74 97 2C 47 B8
BB 81 A2 4A FB A0 D1 AD 75 DC D9 35 82 24 69 A6

EAPOL HMAC : 43 BE F4 6B 1D 7D 38 F2 58 BF A5 AA B1 82 D2 1A

[Zermelo]

Thanks for the info fnord0. I'm going to chalk my crappy performance to the use of the old drivers. I'm going to try upgrading the driver again to see if I can resolve the errors.

On a separate note, Edgan has made a patch for aircrack-ng which allows it to read the passthough hash output from pyrit as stdout.

You can get the patch or modified suite here:

Index of /~edgan/hash/hash

Note: compile it without sqlite support, as there is a bug in it that will cause it to fail if compiled with sqlite=true. The patch works, the only problem there isn't any performance figure output, the only thing that is displayed is whether the key is found or the "passphrase is not in the list". Here is a pic of the figures on my box. Left screen is the output using the passthough, the right is a normal dictionary attack:




Doing the math (1113548/116.124) = 9589.3 k/s. Based on this figure, I'm assuming that both cores of the 295 gtx are not being utilized (or could still be my drivers), but still not bad figures. When both the cuda and hash code is optimized, we can expect to see some very nice performance figures. Note in the picture you'll notice that I used "aircrack-hash" rather than "aircrack-ng", this is my convention, to separate the "hash" supported aircrack, from the "cuda" supported aircrack from the standard aircrack-ng, after compiling expect to use the standard "aircrack-ng" as the name.

EDIT: Did a complete reinstallation of 64bit Intrepid, installed latest cuda drivers without any problems, installed pyrit, and test benchmarks, ~19500 hashs/s. Installed aircrack-ng with cuda support and I still get the same performance figures. So, it was not a driver issue. For whatever reason with the 295 GTX, there still needs to be a lot of optimization.

[imported_gr8gorilla]

Ok, I asked a question and then found my answer, apparently both CUDA and Firestream are supported by pyrit. Now has anyone actually used a machine with either a Tesla or an ATI firestream GPU for this, I am building my new machine with the ability to upgrade in the future to a dual gpu setup, that is not counting the graphics cards I am putting in it. I have found a pair of tesla's on ebay for 999$ dollars each, thats a nice discount. Also the new Firestreams from AMD/ATI are about 1200 I think, they have those on New Egg.

That is my question, what kind of speeds would we reach with a basically quad gpu system with a quad core processor? Two video cards and then either two stream processors from AMD or two Tesla's from nVidia?