Batchprocessing to stdout – Update

Someone pointed out that cowpatty 4.3 indeed does not accept hashfiles to be read via stdin by default. This is caused by a small bug in cowpatty. I fixed that for my version of cowpatty but didn’t tell anyone about this fact :-)

If you get stuck with cowpatty not accepting stdin as hashfile input try this small patch:


More performance data

Added some performance numbers for the GeForce 8800 GTS 512 and the GeForce 260 GTX. New blinkenlights here.

Elcomsoft again

Today a story about elcomsoft’s “patent-pending” GPGPU-WPA-Cracker is taking it’s rounds through the blogs. Here are my 2 cents about this:

  • In what respect is all this “new” ?
  • Who cares for all this buzz-talk ?
  • Why is this patent-pending? Pyrit implements all this in under 2.000 lines of code…
  • Who the fuck is David Hobson and why is everyone listening to him?

[Update] Someone mailed me to point out that the original story on is indeed posted under the “Advertisment” section…

The twilight of Wi-Fi Protected Access

More background on Pyrit – See here

Slow exporting to airolib-ng format

People have been asking why exporting to an airolib-ng database is so slow. Don’t expect things to change dramatically on that side…

SQLite databases provide ACID-compliance on the row-level; this creates a lot of overhead which causes lots of disk-seeks. In turn sqlite’s performance for this specific use is quite bad. Pyrit’s own datastore (the ‘blobspace’) provides no atomicity, *some* consistency, *some* isolation and no durability. We can easily live with that as the cost of providing all those idioms is much higher than the expected rate of failure mulitplied by the cost to re-compute some workunits.

Also be aware that airolib-ng databases create a huge overhead in terms of storage and is easily three times the size of the original blobspace.

Even more bang

I just updated Pyrit to rev53 and the way Pyrit handles workunits while batchprocessing should now almost perfectly scale to faster cards when working with small workunits. Also some regressions introduced in the merger of rev52 were fixed. Everyone should update.

More bang for the buck

I’veĀ  updated svn to rev52. The blobspace classes are somewhat more pythonic now and the damn Pyrit-class had to go. The merger probably introduced some bugs with missing attributes/functions…

The more imported thing is that reading and writing of result files and reading of password files is now done asynchronously. This means that there is practically no idle time between switching workunits and in result much higher efficiency of the CUDA core.

  • RSS Unknown Feed

    • An error has occurred; the feed is probably down. Try again later.