experchange > slackware

Eef Hartman (11-09-18, 09:02 PM)
Rich <rich> wrote:
> Well, yes, but "super-computer (or a massive parallel one)" today just
> means "GPU rig" (from the same url as above):


No, even in my time "massive parallel" meant some box with at LEAST
64K cpu's (each with their own secondary cache and a bank of local
RAM). By now a massive parallel system could have as much as 1M of
(strictly separate) cpu's.
The speed problem with cpu core's and/or GPU is the access to the shared
memory (I mean RAM, for the program code and optional the dictionary).

> Twenty eight days to brute force an 8 char. crypt() password.


And the other rhing is that nowadays passwords aren't restricted to 8
chars anymore. And the charset might be not pure ASCII (95 printables)
anymore but an ISO-8859 or UTF-8 one, giving many more chars to try.

> And adding more GPU's increases the performance, so the final speed is
> really dependent upon "how much money ya got to spend?"


AND how much time!
Rich (11-10-18, 01:56 AM)
Eef Hartman <E.J.M.Hartman> wrote:
> Rich <rich> wrote:
> No, even in my time "massive parallel" meant some box with at LEAST
> 64K cpu's (each with their own secondary cache and a bank of local
> RAM). By now a massive parallel system could have as much as 1M of
> (strictly separate) cpu's.


Yes, but for password cracking, folks don't buy a traditional
'massively parallel' system, they buy eight Nvida 1080 GPU's.

> The speed problem with cpu core's and/or GPU is the access to the
> shared memory (I mean RAM, for the program code and optional the
> dictionary).


Access speeds to ram onboard video cards for modern GPU setups is
significantly higher than CPU/memory bandwidth pathways. I have no
numbers, but I'd not be surprised if the ram speed for CPU's outpases
even the speed available to the "massive parallel" systems you are
thinking of.

>> Twenty eight days to brute force an 8 char. crypt() password.

> And the other rhing is that nowadays passwords aren't restricted to 8
> chars anymore. And the charset might be not pure ASCII (95
> printables) anymore but an ISO-8859 or UTF-8 one, giving many more
> chars to try.


All true, and beside the point. For a post where one is comparing time
to brute force, one's got to pick *something* as an example. My
'something' was an 8 character password selected from the 95 printable
ASCII characters.

>> And adding more GPU's increases the performance, so the final speed is
>> really dependent upon "how much money ya got to spend?"

> AND how much time!


For most of these rigs, the time involves little more than: "insert
eight Nvidia 1080 video cards into eight PCI-E slots", boot up system,
run hash-cat".

Of course, I'm assuming in my statement above that the individual
running the rig has already: 1) purchased a mother board with eight
pci-e slots, 2) purchased a PSU (or multiple PSU's) of sufficient
wattage to feed the GPUs, 3) already installed an OS on a boot disk
attached to the motherboard. But even adding in all this time, it is
not significant in the grand scheme.
Eli the Bearded (11-10-18, 03:14 AM)
In alt.os.linux.slackware, Rich <rich> wrote:
> For most of these rigs, the time involves little more than: "insert
> eight Nvidia 1080 video cards into eight PCI-E slots", boot up system,
> run hash-cat".
> Of course, I'm assuming in my statement above that the individual
> running the rig has already: 1) purchased a mother board with eight
> pci-e slots, 2) purchased a PSU (or multiple PSU's) of sufficient
> wattage to feed the GPUs, 3) already installed an OS on a boot disk
> attached to the motherboard. But even adding in all this time, it is
> not significant in the grand scheme.


Or the individual creates an AWS instance on someone else's credit
card. p2.8xlarge[*] has 8GPUs for $3.40 / hour. There's also a p2.16xlarge
which is 16GPUs for $6.80 / hour.
[*] "P2 instances are intended for general-purpose GPU compute
applications." It's not the only instance type with a GPU option.
There are G3 "optimized for graphics-intensive" and P3 "latest
generation of general purpose GPU". I'm not an expert by any
means, but I think the P2 is the "Nvidia 1080" one.

Elijah
------
the "someone else's" card bit is optional, of course
Rich (11-10-18, 07:28 AM)
Eli the Bearded <*> wrote:
> In alt.os.linux.slackware, Rich <rich> wrote:
> Or the individual creates an AWS instance on someone else's credit
> card. p2.8xlarge[*] has 8GPUs for .40 / hour. There's also a p2.16xlarge
> which is 16GPUs for .80 / hour.
>[*] "P2 instances are intended for general-purpose GPU compute
> applications." It's not the only instance type with a GPU option.
> There are G3 "optimized for graphics-intensive" and P3 "latest
> generation of general purpose GPU". I'm not an expert by any
> means, but I think the P2 is the "Nvidia 1080" one.


Yup, for those who wish to let someone else pay, this is a way.

Not everyone who builds a gpu password cracking rig is a criminal,
however:

Jimmy Johnson (11-11-18, 07:46 AM)
On 11/08/2018 11:18 AM, root wrote:
[..]
> of things short of re-installing the whole system.
> Any suggestions about what might cause this?
> Thanks.


Okay, so you say you can login as root, well do it and then type #
'passwd root' in the console and enter a root passwd and see if that
solves your problem. Also check your display manager, kdm or what ever
you are using and see if it is set to do auto login.

Similar Threads