Discussion:
GPU-side mask vs. tens of millions of hashes
(too old to reply)
magnum
2015-10-01 20:58:38 UTC
Permalink
Sayantan,

I got the idea to test a really large number of hashes with GPU-side
mask (primarily to stress-test the new FMT_REMOVE code), and it bailed
ungracefully:

$ ../run/john 25m_md5.in -form:raw-md5-opencl -mask:?a -min-len=3 -max-len=6
Device 0: GeForce GTX TITAN X
Loaded 24917564 password hashes with no different salts (Raw-MD5-opencl
[MD5 OpenCL])
128 bit add overflow.

Is this an expected outcome, or is 25 million hashes supposed to work
fine? If it's not supposed to work, what is the design maximum? Perhaps
we can bail out with a bit more finesse, eg. telling the user a max.?

oclHashcat supports "up to 100 million hashes". I'm not sure we need to
support that, but it wouldn't hurt.

magnum
Sayantan Datta
2015-10-02 03:26:13 UTC
Permalink
Post by magnum
Sayantan,
I got the idea to test a really large number of hashes with GPU-side mask
(primarily to stress-test the new FMT_REMOVE code), and it bailed
$ ../run/john 25m_md5.in -form:raw-md5-opencl -mask:?a -min-len=3 -max-len=6
Device 0: GeForce GTX TITAN X
Loaded 24917564 password hashes with no different salts (Raw-MD5-opencl
[MD5 OpenCL])
128 bit add overflow.
Is this an expected outcome, or is 25 million hashes supposed to work
fine? If it's not supposed to work, what is the design maximum? Perhaps we
can bail out with a bit more finesse, eg. telling the user a max.?
oclHashcat supports "up to 100 million hashes". I'm not sure we need to
support that, but it wouldn't hurt.
magnum
This is an issue with table building. I'll fix. 25 million shouldn't be
much of an issue. In fact, I had 100-120 million hashes in mind when
writing tables algorithm, and I tested it for 110 million 128bit hashes and
upto 60 million 192 bit hashes. However, as solar said, some adjustments
are required regarding initial table size.

Regards,
Sayantan

Loading...