Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20121208193046.GB31553@openwall.com>
Date: Sat, 8 Dec 2012 23:30:46 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: GPU based cracking, AMD or NVIDIA

On Fri, Dec 07, 2012 at 08:16:23AM +0100, hm0@....net wrote:
> I'm thinking of trying john (and maybe some other progs) on GPU. I wonder if better go for gtx 660 or AMD 7870 (well I dont want to spend $300-400+ for a card...).

Of NVIDIA cards, if you consider them at all, go for GTX 5xx series.
GTX 6xx are generally a lot slower at password cracking.

Here are some reasonable options:

ATI/AMD:
5770 / 6770 (same thing)
5850
5870
5970 (dual GPU)
6950 (some are unlockable to 6970 with BIOS re-flash)
6970
6990 (dual GPU)
7850
7870
7950 (and its revisions - default clock rates vary)
7970 (and its revisions - default clock rates vary)
7970 X2 (dual GPU, unofficial)

NVIDIA:
GTX 560 336 SPs (GF114)
GTX 560 Ti 384 SPs (GF114)
GTX 560 Ti 448 SPs (GF110)
GTX 570
GTX 580

> I tried to find out if amd or nvidia is in general better (faster / maybe openCL vs CUDA?) and/or better supported (maybe again opencl vs cuda?) for gpu based cracking (eg hash, aes, wpa ...).

AMD/ATI is generally faster for password cracking (often 3x faster at
the same price point), but NVIDIA gives you both OpenCL and CUDA - so
lets you run CUDA-only tools as well.  There are some occasional
exceptions - e.g. JtR's sha512crypt-opencl is slightly faster on
GTX 570 and GTX 580 than it is on HD 7970 - but these are just that,
exceptions.

NVIDIA drivers tend to be more reliable than AMD's.

For a mid-range GPU card, your suggested HD 7870 makes sense.

> btw: does gpu based cracking benefit from more RAM on the gfx card?

With current versions of JtR, no, but some future algorithms might
(attack implementations and new hash types).  Of currently available
software, I guess oclHashcat-plus will let you use larger wordlists (and
apply mangling rules on GPU) with more memory on the GPU card.
Similarly, oclHashcat-plus, Cryptohaze Multiforcer, and hashkill load
hashes to crack onto the GPU card, so more memory on the GPU card will
let you crack more hashes at once with those tools.  (JtR does not do it
yet, but it will for so-called fast hashes when we implement that.)
Also, Litecoin miners do benefit from more RAM on GPU card - it can
translate to higher speed there.

That said, it is unlikely that GPU card RAM size will matter for you, as
long as it's sane (currently 1 GB or more).

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.