Amazon EC2 GPU computing

964 views
Skip to first unread message

erhar...@yahoo.com

unread,
Nov 15, 2010, 10:08:30 AM11/15/10
to Pyrit
Today Amazon announced GPU computer cluster. It runs Cent OS 5. Did
anyone tried to install pyrit on an instance? It has an Nvidia Tesla
M2050 GPU. Right now I'm trying to install python 2.6 and I will post
any results(if I succeed).

E

unread,
Nov 15, 2010, 12:15:59 PM11/15/10
to Pyrit
I managed to run install everything I need to run pyrit:Please take in
consideration that I'm noob in linux and it was first time I used
CentOS and rpm packages but I managed to make it all run in 2 hours.
THAT'S CHEAP and easy!

[root@ip-10-17-130-115 ~]# pyrit list_cores
Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3+

The following cores seem available...
#1: 'CUDA-Device #1 'Tesla M2050''
#2: 'CUDA-Device #2 'Tesla M2050''
#3: 'CPU-Core (SSE2)'
#4: 'CPU-Core (SSE2)'
#5: 'CPU-Core (SSE2)'
#6: 'CPU-Core (SSE2)'
#7: 'CPU-Core (SSE2)'
#8: 'CPU-Core (SSE2)'
#9: 'CPU-Core (SSE2)'
#10: 'CPU-Core (SSE2)'
#11: 'CPU-Core (SSE2)'
#12: 'CPU-Core (SSE2)'
#13: 'CPU-Core (SSE2)'
#14: 'CPU-Core (SSE2)'
#15: 'CPU-Core (SSE2)'
#16: 'CPU-Core (SSE2)'


[root@ip-10-17-130-115 ~]# pyrit benchmark_long
Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3+

Running benchmark (48088.8 PMKs/s)... |

Computed 48088.83 PMKs/s total.
#1: 'CUDA-Device #1 'Tesla M2050'': 21224.1 PMKs/s (RTT 3.0)
#2: 'CUDA-Device #2 'Tesla M2050'': 21319.3 PMKs/s (RTT 3.0)
#3: 'CPU-Core (SSE2)': 447.1 PMKs/s (RTT 3.0)
#4: 'CPU-Core (SSE2)': 439.6 PMKs/s (RTT 3.0)
#5: 'CPU-Core (SSE2)': 441.0 PMKs/s (RTT 3.0)
#6: 'CPU-Core (SSE2)': 447.2 PMKs/s (RTT 3.0)
#7: 'CPU-Core (SSE2)': 445.5 PMKs/s (RTT 3.0)
#8: 'CPU-Core (SSE2)': 433.2 PMKs/s (RTT 3.0)
#9: 'CPU-Core (SSE2)': 438.9 PMKs/s (RTT 3.0)
#10: 'CPU-Core (SSE2)': 444.9 PMKs/s (RTT 3.0)
#11: 'CPU-Core (SSE2)': 444.3 PMKs/s (RTT 3.0)
#12: 'CPU-Core (SSE2)': 442.8 PMKs/s (RTT 3.0)
#13: 'CPU-Core (SSE2)': 441.0 PMKs/s (RTT 3.0)
#14: 'CPU-Core (SSE2)': 446.4 PMKs/s (RTT 3.0)
#15: 'CPU-Core (SSE2)': 435.7 PMKs/s (RTT 3.0)
#16: 'CPU-Core (SSE2)': 439.6 PMKs/s (RTT 3.0)


[root@ip-10-17-130-115 ~]# uptime
09:13:08 up 1:44, 1 user, load average: 13.02, 13.61, 8.20


After doing the math you can see that an 1,000,000,000 worldlist and
it will take ~7hours to complete(~40k pmk/s)= $14.7 !(WOW that's
cheap)

Regards,
Erhard Muresan

E

unread,
Nov 15, 2010, 12:23:40 PM11/15/10
to Pyrit
Forgot to mention: You can run 7 instances, it will take you a little
while to configure but it will only take 1hour to complete and it
still cost just $14.7 or you can run 20 instances(max) and it will
take 20 min and cost $42 (and you can crack 3 .cap files because you
still have 40 min left from your instance)

On Nov 15, 7:15 pm, E <erhard_...@yahoo.com> wrote:
> I managed to run install everything I need to run pyrit:Please take in
> consideration that I'm noob in linux and it was first time I used
> CentOS and rpm packages but I managed to make it all run in 2 hours.
> THAT'S CHEAP and easy!
>
> [root@ip-10-17-130-115 ~]# pyrit list_cores
> Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueghttp://pyrit.googlecode.com
> This code is distributed under the GNU General Public License v3+
>
> The following cores seem available...
> #1:  'CUDA-Device #1 'Tesla M2050''
> #2:  'CUDA-Device #2 'Tesla M2050''
> #3:  'CPU-Core (SSE2)'
> #4:  'CPU-Core (SSE2)'
> #5:  'CPU-Core (SSE2)'
> #6:  'CPU-Core (SSE2)'
> #7:  'CPU-Core (SSE2)'
> #8:  'CPU-Core (SSE2)'
> #9:  'CPU-Core (SSE2)'
> #10:  'CPU-Core (SSE2)'
> #11:  'CPU-Core (SSE2)'
> #12:  'CPU-Core (SSE2)'
> #13:  'CPU-Core (SSE2)'
> #14:  'CPU-Core (SSE2)'
> #15:  'CPU-Core (SSE2)'
> #16:  'CPU-Core (SSE2)'
>
> [root@ip-10-17-130-115 ~]# pyrit benchmark_long
> Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueghttp://pyrit.googlecode.com
Message has been deleted

Lukas Lueg

unread,
Nov 16, 2010, 3:02:56 PM11/16/10
to Pyrit
Thanks for the information, very interesting!

Lukas Lueg

unread,
Nov 16, 2010, 5:21:44 PM11/16/10
to Pyrit
E, how are you billed when using an EC2 instance? Per CPU hour? Then
it would be wise to disable all those (underperforming) CPU cores...

On 15 Nov., 18:15, E <erhard_...@yahoo.com> wrote:
> I managed to run install everything I need to run pyrit:Please take in
> consideration that I'm noob in linux and it was first time I used
> CentOS and rpm packages but I managed to make it all run in 2 hours.
> THAT'S CHEAP and easy!
>
> [root@ip-10-17-130-115 ~]# pyrit list_cores
> Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueghttp://pyrit.googlecode.com
> This code is distributed under the GNU General Public License v3+
>
> The following cores seem available...
> #1:  'CUDA-Device #1 'Tesla M2050''
> #2:  'CUDA-Device #2 'Tesla M2050''
> #3:  'CPU-Core (SSE2)'
> #4:  'CPU-Core (SSE2)'
> #5:  'CPU-Core (SSE2)'
> #6:  'CPU-Core (SSE2)'
> #7:  'CPU-Core (SSE2)'
> #8:  'CPU-Core (SSE2)'
> #9:  'CPU-Core (SSE2)'
> #10:  'CPU-Core (SSE2)'
> #11:  'CPU-Core (SSE2)'
> #12:  'CPU-Core (SSE2)'
> #13:  'CPU-Core (SSE2)'
> #14:  'CPU-Core (SSE2)'
> #15:  'CPU-Core (SSE2)'
> #16:  'CPU-Core (SSE2)'
>
> [root@ip-10-17-130-115 ~]# pyrit benchmark_long
> Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueghttp://pyrit.googlecode.com

Dave

unread,
Nov 17, 2010, 7:07:33 AM11/17/10
to Pyrit
It's per instance I believe.

The cg1.4xlarge which has the GPUs are $2.10 per hour per instance.

Lukas Lueg

unread,
Nov 17, 2010, 7:39:03 AM11/17/10
to py...@googlegroups.com
But what is 'an hour' ? CPU-time or wall-clock time?

2010/11/17 Dave <cbo...@rarr.org.uk>:

> --
> You received this message because you are subscribed to the Google Groups "Pyrit" group.
> To post to this group, send email to py...@googlegroups.com.
> To unsubscribe from this group, send email to pyrit+un...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/pyrit?hl=en.
>
>

E M

unread,
Nov 17, 2010, 9:53:15 AM11/17/10
to py...@googlegroups.com
wall-clock time;).
Btw, I didn't pay anything because of the new user "Free Tier" amazon offer:
http://aws.amazon.com/ec2/pricing/

--- On Wed, 11/17/10, Lukas Lueg <lukas...@googlemail.com> wrote:
> To unsubscribe from this group, send email to pyrit+unsub...@googlegroups.com.

> For more options, visit this group at http://groups.google.com/group/pyrit?hl=en.
>
>

--
You received this message because you are subscribed to the Google Groups "Pyrit" group.
To post to this group, send email to py...@googlegroups.com.
To unsubscribe from this group, send email to pyrit+unsub...@googlegroups.com.

cisco guy

unread,
Nov 18, 2010, 10:47:00 AM11/18/10
to Pyrit
Hey Lucas,

First I would like to thank you for Pyrit. It’s a master piece :)

I just read the post on running Pyrit on Amazon’s cloud and wanted to
ask which is more efficient from your perspective:

A) Run the brute force attack on the fly by feeding Pyrit with output
from Crunch directly on the cloud. (Key span: 9 – 12, Alfa Numeric)

B) Generate the word list dictionary off the cloud, upload it,
calculate PMKs using the cloud for certain network, and then dump the
db back on a regular computer to run the attack.

Thank you,

CG

oxident

unread,
Nov 21, 2010, 3:06:21 AM11/21/10
to Pyrit
It's a pity that those expensive Tesla cards don't reach half of the
speed a single ATI 5970 does :-(


On 15 Nov., 18:15, E <erhard_...@yahoo.com> wrote:
> I managed to run install everything I need to run pyrit:Please take in
> consideration that I'm noob in linux and it was first time I used
> CentOS and rpm packages but I managed to make it all run in 2 hours.
> THAT'S CHEAP and easy!
>
> [root@ip-10-17-130-115~]# pyrit list_cores
> Pyrit 0.4.0-dev (svn r288) (C)2008-2010Lukas Lueghttp://pyrit.googlecode.com
> This code is distributed under the GNU General Public License v3+
>
> The following cores seem available...
> #1:  'CUDA-Device #1 'Tesla M2050''
> #2:  'CUDA-Device #2 'Tesla M2050''
> #3:  'CPU-Core (SSE2)'
> #4:  'CPU-Core (SSE2)'
> #5:  'CPU-Core (SSE2)'
> #6:  'CPU-Core (SSE2)'
> #7:  'CPU-Core (SSE2)'
> #8:  'CPU-Core (SSE2)'
> #9:  'CPU-Core (SSE2)'
> #10:  'CPU-Core (SSE2)'
> #11:  'CPU-Core (SSE2)'
> #12:  'CPU-Core (SSE2)'
> #13:  'CPU-Core (SSE2)'
> #14:  'CPU-Core (SSE2)'
> #15:  'CPU-Core (SSE2)'
> #16:  'CPU-Core (SSE2)'
>
> [root@ip-10-17-130-115~]# pyrit benchmark_long
> Pyrit 0.4.0-dev (svn r288) (C)2008-2010Lukas Lueghttp://pyrit.googlecode.com
> This code is distributed under the GNU General Public License v3+
>
> Running benchmark (48088.8 PMKs/s)... |
>
> Computed48088.83PMKs/s total.
> #1: 'CUDA-Device #1 'Tesla M2050'': 21224.1 PMKs/s (RTT 3.0)
> #2: 'CUDA-Device #2 'Tesla M2050'': 21319.3 PMKs/s (RTT 3.0)
> #3: 'CPU-Core (SSE2)': 447.1 PMKs/s (RTT 3.0)
> #4: 'CPU-Core (SSE2)': 439.6 PMKs/s (RTT 3.0)
> #5: 'CPU-Core (SSE2)': 441.0 PMKs/s (RTT 3.0)
> #6: 'CPU-Core (SSE2)': 447.2 PMKs/s (RTT 3.0)
> #7: 'CPU-Core (SSE2)': 445.5 PMKs/s (RTT 3.0)
> #8: 'CPU-Core (SSE2)': 433.2 PMKs/s (RTT 3.0)
> #9: 'CPU-Core (SSE2)': 438.9 PMKs/s (RTT 3.0)
> #10: 'CPU-Core (SSE2)': 444.9 PMKs/s (RTT 3.0)
> #11: 'CPU-Core (SSE2)': 444.3 PMKs/s (RTT 3.0)
> #12: 'CPU-Core (SSE2)': 442.8 PMKs/s (RTT 3.0)
> #13: 'CPU-Core (SSE2)': 441.0 PMKs/s (RTT 3.0)
> #14: 'CPU-Core (SSE2)': 446.4 PMKs/s (RTT 3.0)
> #15: 'CPU-Core (SSE2)': 435.7 PMKs/s (RTT 3.0)
> #16: 'CPU-Core (SSE2)': 439.6 PMKs/s (RTT 3.0)
>
> [root@ip-10-17-130-115~]# uptime

a mok

unread,
Nov 21, 2010, 5:09:46 AM11/21/10
to Pyrit
So ATI hardware still does *that* much better than NV hardware at
integer arithmethic?

Even though the Tesla M2050 is based on Fermi arch which has much
improved integer handling, or at least should have.

oxident

unread,
Nov 22, 2010, 5:12:11 PM11/22/10
to Pyrit
Well, I can only compare the OP's results (42k PMKs/sec with CUDA)
with my own setup (140k PMK/s with CAL on a 5970). Maybe the M2050
outperforms the Cypress chips in many other tasks but not in this
case.

Dave

unread,
Nov 25, 2010, 2:57:46 AM11/25/10
to Pyrit
That's 3.5x difference.. that's huge!

a mok

unread,
Nov 25, 2010, 6:54:18 AM11/25/10
to Pyrit
Yes, that's a huge difference indeed.

The only thing I could find in a quick google was that the new ATI
cards support bit rotation operations natively, and Fermi only does
bitshift. Then again, I don't believe that can cause such a big
difference.

Lukas Lueg

unread,
Nov 25, 2010, 7:06:00 AM11/25/10
to py...@googlegroups.com
There are a *lot* of bitwise rotations required in SHA1. Each has to
be done with two shifts and one logical or-combinations on nvidia.

In the end however, the ATI hardware seems to have much better
instruction scheduling on the GPU, getting more performance out of the
code. The CUDA-kernel is not optimized in any way and may deliver much
better performance on the hardware side if generated from better code.

2010/11/25 a mok <witchs...@gmail.com>:

> --
> You received this message because you are subscribed to the Google Groups "Pyrit" group.
> To post to this group, send email to py...@googlegroups.com.

> To unsubscribe from this group, send email to pyrit+un...@googlegroups.com.

Erdem Umut Altinyurt

unread,
Dec 3, 2010, 9:12:18 AM12/3/10
to Pyrit
AMD HD6850 could give you 44K too.

death@triQuad:~/Desktop> pyrit benchmark
Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3+

Running benchmark (42646.0 PMKs/s)... /

Computed 42646.02 PMKs/s total.
#1: 'CAL++ Device #1 'ATI'': 43775.1 PMKs/s (RTT 2.7)
#2: 'CPU-Core (SSE2)': 547.9 PMKs/s (RTT 3.0)
#3: 'CPU-Core (SSE2)': 548.5 PMKs/s (RTT 3.0)
#4: 'CPU-Core (SSE2)': 551.8 PMKs/s (RTT 3.0)
#5: 'Network-Clients': 0.0 PMKs/s (RTT 0.0)
death@triQuad:~/Desktop>

james0p...@googlemail.com

unread,
Dec 4, 2010, 5:26:09 AM12/4/10
to Pyrit
Note that using the Free tier, you could use 100 Micro instances for
7.5 hours using server, to do a 80,000k/s. Using pyrit serve.

oxident

unread,
Dec 6, 2010, 2:52:04 PM12/6/10
to Pyrit
You are sure, free tier is also available for GPU instances? I've
tried it and now Amazon has charged me 2.50$ ;-)

On 4 Dez., 11:26, "james0p0corb...@googlemail.com"

Gurgalof

unread,
Dec 7, 2010, 12:15:19 AM12/7/10
to Pyrit
Free Tier is only for micro instances, not GPU...

Anyways, it's still cheap!
I have successfully created some Cowpatty hashtables with some common
ESSID's here in Sweden.

Maybe I will create a step by step tutorial someday...

vidkun

unread,
Dec 8, 2010, 4:43:31 PM12/8/10
to Pyrit
Just for comparison, I was getting around 43,000 - 48,000 PMK/s on the
Amazon GPU Cluster. On my machine at home with a Core i7-930, a GTX
480, and a GTX470, I can reach 50,000-58,000 PMK/s.



On Nov 15, 11:15 am, E <erhard_...@yahoo.com> wrote:
> I managed to run install everything I need to run pyrit:Please take in
> consideration that I'm noob in linux and it was first time I used
> CentOS and rpm packages but I managed to make it all run in 2 hours.
> THAT'S CHEAP and easy!
>
> [root@ip-10-17-130-115 ~]# pyrit list_cores
> Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueghttp://pyrit.googlecode.com
> This code is distributed under the GNU General Public License v3+
>
> The following cores seem available...
> #1:  'CUDA-Device #1 'Tesla M2050''
> #2:  'CUDA-Device #2 'Tesla M2050''
> #3:  'CPU-Core (SSE2)'
> #4:  'CPU-Core (SSE2)'
> #5:  'CPU-Core (SSE2)'
> #6:  'CPU-Core (SSE2)'
> #7:  'CPU-Core (SSE2)'
> #8:  'CPU-Core (SSE2)'
> #9:  'CPU-Core (SSE2)'
> #10:  'CPU-Core (SSE2)'
> #11:  'CPU-Core (SSE2)'
> #12:  'CPU-Core (SSE2)'
> #13:  'CPU-Core (SSE2)'
> #14:  'CPU-Core (SSE2)'
> #15:  'CPU-Core (SSE2)'
> #16:  'CPU-Core (SSE2)'
>
> [root@ip-10-17-130-115 ~]# pyrit benchmark_long
> Pyrit 0.4.0-dev (svn r288) (C) 2008-2010 Lukas Lueghttp://pyrit.googlecode.com

Gurgalof

unread,
Dec 9, 2010, 8:04:06 AM12/9/10
to Pyrit
And for those like me that only owns a laptop, the Amazon GPU cluster
is fine.
I know that you could easily build a machine that is better, but I
cannot afford it.

And I don't need THAT much power that it will be cheaper with a
machine of my own.
Message has been deleted

jang thinh

unread,
Jan 10, 2011, 5:45:18 PM1/10/11
to Pyrit
Did anyone tried pyrit in virtualised environment - I mean VMware,
Oracle VirtualBox, XEN etc... Amazon Cloud is also virtualised...

Jimmy

unread,
Feb 24, 2011, 5:02:11 PM2/24/11
to Pyrit
In virtual environment cuda or opencl doesn't support.

E M

unread,
Feb 25, 2011, 7:13:29 AM2/25/11
to py...@googlegroups.com
AFAIK, in Xen virtualization you can pass the PCI Express Video card in domU. Amazon uses Xen virtualization to make the instances.
Source: http://trac.nchc.org.tw/grid/wiki/Xen_GPU_cluster

--- On Fri, 2/25/11, Jimmy <ando...@gmail.com> wrote:
--
You received this message because you are subscribed to the Google Groups "Pyrit" group.
To post to this group, send email to py...@googlegroups.com.
To unsubscribe from this group, send email to pyrit+unsub...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages