Some flags in Pyrit have changed, the -f (file) flag is replaced by -i (for input) and -o for output.
On page 18 you kinda missed (ignore this if the procedure were different on earlier versions):
to actually start the batch process. Because if you just add the wordlist and then export it with export_cowpatty there will be no lines written. So:
Code:
pyrit -e [bssid] create_essid
pyrit -i [wordlist] import_passwords
pyrit batch
pyrit -e [bssid] -o output.cow export_cowpatty
cowpatty -d output.cow -r wpa.cap -s [bssid]
Now, I was getting nasty buffer overflows with the version of cowpatty shipped with bt4, but there is a new version (4.6) available here: willhackforsushi.com/code/cowpatty/4.6/cowpatty-4.6.tgz
(and that one solved my buffer overflows).
For the adventurous, Pyrit can be checkout from SVN, and it contains everything you need (CUDA support included). More info here: code.google.com/p/pyrit/wiki/Installation
Getting a multi-gpu-setup working was a little bit of a hassle. If you're running several cards and get weird errors, try removing all but one to get it set up, then insert the remaining cards again. I managed to run one card in X, using CUDA. Or several cards in terminal using CUDA, but not in X. But I see no reason to use a GUI for this anyhow, so no loss for me.
And finally, a small tip for those building the cuda SDK from source: If the unpacking seems to freeze, it's actually expecting user input for the path where it's going to install the SDK. Just hit enter and it will place at your current path, continuing the install. I've heard that this was a problem in Fedora8, but I got the same problem running bt4.