Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <413c2ae532a525aba6c60de955e42996@smtp.hushmail.com>
Date: Fri, 24 Apr 2015 19:56:12 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Jumbo AVX2 vs. Hashcat

I thought I'd run sanity-checks of Jumbo against Hashcat v0.49 in order 
to see any bad figures we should improve on. I expected HC to show 
really challenging speeds (admittedly just based on oclHashcat 
experience), but here are the results:

Running AVX2 with 1 or 4 threads (i7-4790, no HT):

                            real runs (mask mode)
Hash       HCx1    HCx4    JtRx1   JtRx4   fork4
------------------------------------------------
MD4      23.93M  75.12M    69.02M  72.13M   250M
MD5      19.48M  60.95M    59.25M  53.29M*  215M (* look into this)
NT       20.55M  63.02M    66.04M  75.37M   250M
SHA1     16.83M  42.17M    36.71M  44.09M   135M
SHA256    9.22M  27.71M    19.36M  29.21M    68M
SHA512    4.11M  11.38M     8.15M  24.37M    30M
WPAPSK    1.54k   4.31k     3.25k   6.37k
DEScrypt 600.4k   2.29M       N/A

Err, what the heck is this!? I haven't looked much at Hashcat ever but I 
did not expect it to be this much slower than JtR. We match/beat Atom's 
4x speeds using just a single thread. This can't be true, can it?

After some head scratching I thought I'd bench AVX vs. AVX too and 
there's the answer - something is fundamentally wrong just with 
Hashcat's AVX2. Here's a battle using just AVX:

                             real runs
Hash       HCx1    HCx4     JtRx1   JtRx4
-----------------------------------------
MD4      22.90M  76.76M    35.45M  56.26M
MD5      18.92M  71.96M    27.38M  44.99M
NT       21.52M  64.87M    36.72M  59.53M
SHA1     17.14M  53.92M    22.63M  33.54M
SHA256    9.23M  35.22M     9.66M  28.57M
SHA512    4.12M  11.37M     3.89M*  7.59M
WPAPSK    1.53k   4.32k     1.71k   4.92k
DEScrypt 600.4k   1.82M     5.95M  19.61M (bitslice FTW!)

We're still generally faster than Hashcat on single thread (or fork), 
but not with silly amounts. Perhaps HC doesn't feature interleaving. 
Anyway we should be able to improve on SHA512 with at least 5%, which is 
what I looked for in the first place %-)

magnum

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.