Date: Tue, 13 Aug 2013 11:33:34 +0530 From: Sayantan Datta <std2048@...il.com> To: john-dev@...ts.openwall.com Subject: Re: key-length for mask mode. On Tue, Aug 13, 2013 at 10:51 AM, Solar Designer <solar@...nwall.com> wrote: > That's good news, but have you reviewed myrice's PG-test and > experimented with it by now (with a variety of loaded hash counts > ranging from one to millions)? If not yet, please stop all other work > and do that now! It is wrong to abandon previous attempts without even > using the opportunity to learn from them (both what was done right and > what was done wrong). And yes, myrice's code achieved a similar speed, > including with more hashes - IIRC, we were testing with 1M and 10M, as > well as with a few thousand, and the speed somehow varied between > invocations, sometimes being above 3000M c/s and sometimes below. (This > is twice slower than hashcat's, though.) > I have tested myrice PG-test branch without any tweaks: OpenCL platform 0: AMD Accelerated Parallel Processing, 2 device(s). Using device 0: Tahiti Local work size (LWS) 128, Global work size (GWS) 65536 Loaded 12906 password hashes with no different salts (Raw MD5 [OpenCL]) end self test guesses: 0 time: 0:00:00:04 0.00% (3) c/s: 11518G trying: ph0370 - 47192769 guesses: 0 time: 0:00:00:08 0.00% (3) c/s: 14075G trying: drap02 - bms100 guesses: 0 time: 0:00:00:10 0.00% (3) c/s: 14586G trying: cenie105 - mcdo1140 guesses: 0 time: 0:00:00:13 0.00% (3) c/s: 13778G trying: shlva - hsets guesses: 0 time: 0:00:00:15 0.00% (3) c/s: 14159G trying: js7712 - rgli guesses: 0 time: 0:00:00:17 0.00% (3) c/s: 14449G trying: bl746t - paybol guesses: 0 time: 0:00:00:19 0.00% (3) c/s: 14679G trying: sith2 - s23h1 guesses: 0 time: 0:00:00:21 0.00% (3) c/s: 14865G trying: sl3adu - bdbmme Myrice's kernel is spilling into scrach registers which might be the reason for lower performance. Most likely you had tested it with a different driver. One thing I noticed in myrice's kernel was that it was using too many kernel arguments. Increasing kernel args increases VGPR usage count which in turn results in register spilling. I am trying to introduce a hash table but that again increases VGPR count. I would try to minimize VGPR usage so that I can introduce a few more kernel arguments. Also can you post a list of 10M hashes? Regards, Sayantan Content of type "text/html" skipped
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.