Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 4 Sep 2020 00:51:24 +0200
From: Axymeus <axymeus@...il.com>
To: john-users@...ts.openwall.com
Subject: Re: rar-opencl performance

As you can see the performance is only marginally better in phase 3

Device 1: GeForce RTX 2070 SUPER
Using default input encoding: UTF-8
Loaded 1 password hash (rar-opencl, RAR3 [SHA1 OpenCL AES])
Will run 12 OpenMP threads
Proceeding with single, rules:Single
Press 'q' or Ctrl-C to abort, almost any other key for status
0g 0:00:00:21 33.72% 1/3 (ETA: 00:35:02) 0g/s 973.5p/s 973.5c/s
973.5C/s Archive.rarmp4j..zarchive41
Almost done: Processing the remaining buffered candidate passwords, if any.
Warning: Only 15837 candidates buffered for the current salt, minimum
20480 needed for performance.
Proceeding with wordlist:/run/password.lst, rules:Wordlist
0g 0:00:03:07 25.16% 2/3 (ETA: 00:46:24) 0g/s 1042p/s 1042c/s 1042C/s
laropmet..bailey9
0g 0:00:03:41 65.46% 2/3 (ETA: 00:39:38) 0g/s 1159p/s 1159c/s 1159C/s
2zacefron..River7
Warning: Only 13486 candidates left, minimum 20480 needed for performance.
Proceeding with incremental:ASCII
0g 0:00:04:48  3/3 0g/s 1220p/s 1220c/s 1220C/s jaith..murtos
0g 0:00:05:32  3/3 0g/s 1183p/s 1183c/s 1183C/s bakku..mysey
0g 0:00:06:24  3/3 0g/s 1183p/s 1183c/s 1183C/s saluke..mynen1
0g 0:00:08:06  3/3 0g/s 1145p/s 1145c/s 1145C/s bordaku..miamm1
Session aborted

However running rar-opencl with -test gives... 46704 c/s real, 46811
c/s virtual (Benchmarking: rar-opencl, RAR3 (length 5) [SHA1 OpenCL
AES]... (12xOMP))

On Fri, Sep 4, 2020 at 12:28 AM Solar Designer <solar@...nwall.com> wrote:
>
> On Thu, Sep 03, 2020 at 11:47:50PM +0200, Axymeus wrote:
> > However while john is now running, I'm not entirely convinced it's
> > making any use of the GPU. I'm not seeing any GPU utilization from
> > john (really 0%), and performance is only slightly better than running
> > with the CPU (~800 p/s versus 250 p/s, using an i7-8700, a quick test
> > with cRARk tells me my GPU should be able to reach ~18,000 p/s. Of
> > course we're looking at different softwares but still). John also uses
> > the same number of OpenMP threads (12) on both runs
>
> The 0% utilization is weird, and 800 p/s is low.  How long have you been
> running it for?  Can you show the status line?  My guess is it's still
> in "single crack" mode (pass "1/3") where it isn't giving the GPU enough
> work yet.  If the RAR file's name isn't a good hint for use in candidate
> passwords, you may skip that and use other cracking modes right away
> (you'll need to request those explicitly).
>
> As to comparison with cRARk, as I recall it benchmarks for password
> length 5, and we do the same in our "--test" for consistency, but
> performance is lower for higher lengths (with any tool):
>
> [solar@...er run]$ ./john -test -form=rar-opencl -dev=5
> Will run 32 OpenMP threads
> Device 5: GeForce GTX TITAN X
> Benchmarking: rar-opencl, RAR3 (length 5) [SHA1 OpenCL AES]... (32xOMP) LWS=32 GWS=6144 (192 blocks) DONE
> Raw:    31879 c/s real, 30306 c/s virtual, Dev#5 util: 100%
>
> [solar@...er run]$ ./john -test -form=rar-opencl -dev=5 -mask='?a?a?a?a?a?a'
> Will run 32 OpenMP threads
> Device 5: GeForce GTX TITAN X
> Benchmarking: rar-opencl, RAR3 (length 6) [SHA1 OpenCL AES]... (32xOMP) LWS=32 GWS=6144 (192 blocks) DONE
> Raw:    29257 c/s real, 27551 c/s virtual, Dev#5 util: 100%
>
> [solar@...er run]$ ./john -test -form=rar-opencl -dev=5 -mask='?a?a?a?a?a?a?a'
> Will run 32 OpenMP threads
> Device 5: GeForce GTX TITAN X
> Benchmarking: rar-opencl, RAR3 (length 7) [SHA1 OpenCL AES]... (32xOMP) LWS=32 GWS=24576 (768 blocks) DONE
> Raw:    27306 c/s real, 26520 c/s virtual, Dev#5 util: 100%
>
> [solar@...er run]$ ./john -test -form=rar-opencl -dev=5 -mask='?a?a?a?a?a?a?a?a'
> Will run 32 OpenMP threads
> Device 5: GeForce GTX TITAN X
> Benchmarking: rar-opencl, RAR3 (length 8) [SHA1 OpenCL AES]... (32xOMP) LWS=32 GWS=6144 (192 blocks) DONE
> Raw:    25134 c/s real, 24041 c/s virtual, Dev#5 util: 100%
>
> I don't suggest you actually use mask mode when cracking, although you
> may (good idea if you recall part of the password) - rather, this is how
> we specify a length for the benchmarks above.
>
> Alexander

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.