Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Sat, 21 Apr 2012 18:47:42 +0200
From: magnum <>
Subject: New RAR OpenCL kernel

I have now committed a new version of the RAR kernel that does not use
byte addressable store, so should work on any GPU card. Hopefully it is
also faster, I do not know yet :)

Please test and report. I actually haven't tested it on any GPU yet.

You may want to try different worksize values. You can get a raw speed
test using KPC=0 with a given LWS like this (it tries higher values
until one call exceeds 5 seconds):

LWS=4 KPC=0 ../run/john -test -fo:rar
OpenCL platform 1: AMD Accelerated Parallel Processing, 1 device(s).
Using device 0: Intel(R) Core(TM)2 Duo CPU     P8600  @ 2.40GHz

Calculating best keys per crypt for LWS 4, this will take a while
kpc    4	  36 c/s       3391668 sha1/s   0.109 sec per crypt_all()+
kpc    8	  37 c/s       3485881 sha1/s   0.216 sec per crypt_all()+
kpc   16	  48 c/s       4522224 sha1/s   0.328 sec per crypt_all()+
kpc   32	  57 c/s       5370141 sha1/s   0.556 sec per crypt_all()+
kpc   64	  63 c/s       5935419 sha1/s   1.011 sec per crypt_all()+
kpc  128	  66 c/s       6218058 sha1/s   1.912 sec per crypt_all()+
kpc  256	  68 c/s       6406484 sha1/s   3.736 sec per crypt_all()+
kpc  512	  68 c/s       6406484 sha1/s   7.506 sec per crypt_all()
Optimal keys per crypt 256
(to store this, put "rar_KPC = 256" in john.conf, section [Options:OpenCL])

It then exits, so for actually running RAR you'd set KPC=256 and re-run.


Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.