Anyway to benchmark with specific parameters?
#1
Information 
I know that benchmark let's you select the algorithm but I was wondering if there was any way to do it with different parameters because the default benchmark mode for scrypt is (as far as I know) 2^10,1,1 whereas the minimum recommended is 2^14,8,1. With bcrypt it was simple to extrapolate as it just increases cputime with strength but with the memory differences I can' find a way to specify it w/o just putting a huge hash at it and trying to see how long it takes to guess it.

If that's the sole way to do it, does it list the guesses that were required?


P.S. I have an old laptop gpu but should be enough for some rough guesstimates(it's a GTX 485M 2GiB)
#2
Eh? Benchmark scores are measured in hashes per second, not time to exhaust keyspace... Just start cracking your hash and the "Speed.Dev.#1...: 10275.6 kH/s (0.00ms)" figure is your benchmark value. Am I missing something?
#3
(03-09-2017, 06:11 PM)rico Wrote: Eh? Benchmark scores are measured in hashes per second, not time to exhaust keyspace... Just start cracking your hash and the "Speed.Dev.#1...: 10275.6 kH/s (0.00ms)" figure is your benchmark value. Am I missing something?

I was talking about doing a benchmark with specfic parameters for scrypt. I know hat the benchmark scores aren't based upon time to crack a specific hash I am wanting to do something similar with benchmark scores but instead of the deafault parameters for scrypt  of 1024,1,1 have it be 16384,8,1 and see how fast it is. So then you're saying it's impossible to do so?
#4
I don't think you can alter the benchmark parameters or attack mode (default is bruteforce -a 3 I believe) but why extrapolate when you've already identified a solution? Start cracking a hash configured as you've outlined and the Speed.Dev result is your answer...
#5
because the site _doesn't_ explain how it wants the hashes for the various formats I tried doing N:r:p:hash but that isn't working.

and even dongi the following in a file gives me the same error.
Hashfile 'a' on line 1 (SCRYPT:1024:1:1:MDIwMzMwNTQwNDQyNQ==:5FW+zWivLxgCWj7qLiQbeC8zaNQ+qdO0NUinvqyFcfo=): Line-length exception


Edit: I figured it out, I have to specify the mode(why when it's stated at the front of the hash?)
#6
Ah yes, -m 8900. Not all hashes identify themselves so conveniently you must appreciate. Specifying the hash mode is required (unless MD5, the default, -m 0).
#7
(03-09-2017, 07:03 PM)rico Wrote: Ah yes, -m 8900. Not all hashes identify themselves so conveniently you must appreciate. Specifying the hash mode is required (unless MD5, the default, -m 0).

well I can'tseem to get it work and for some reason hashcat isn't seeing the available memory as I see it.

OpenCL Platform #1: NVIDIA Corporation
======================================
* Device #1: GeForce GTX 485M, 496/1984 MB allocatable, 8MCU



I'm trying to do a test hash of the following one.

SCRYPT:16384:8:1Confusedalt:dFcxr0SE8yOWiWntoomu7gBbWQOsVh5kpayhIXl793M=

the password is password so _why_ it'd not be able to use the ram is beyond me and if my math is right each pass of scrypt should only require 20MiB of ram but it sets the thing to --scrypt-tmto 5 and say the following. Where's the remaining 1.5GiB of ram gone? It of course dies because doing 6x the cputime is probably going to cause it to eat dirt on my old laptop but I'm wondering moreso why it can't do even one single pass.

Not enough single-block device memory allocatable to use --scrypt-tmto 2, increasing...
Not enough single-block device memory allocatable to use --scrypt-tmto 3, increasing...
Not enough single-block device memory allocatable to use --scrypt-tmto 4, increasing...
Not enough single-block device memory allocatable to use --scrypt-tmto 5, increasing...



nvidia settings says the following about it
Used Dedicated Memory:77 MB (4%)

so where is the remaining 1.5GiB of memory going? I'm doing nothing except a web browser with a few tabs open and no video playing at the time of trying to run it.
#8
It's not hashcats fault, because the memory allocatable is limited by the OpenCL runtime. We've reported this to nvidia, here's a writeup: https://devtalk.nvidia.com/default/topic/992502
#9
(03-10-2017, 11:52 AM)atom Wrote: It's not hashcats fault, because the memory allocatable is limited by the OpenCL runtime. We've reported this to nvidia, here's a writeup: https://devtalk.nvidia.com/default/topic/992502

So then why doesn't hashcat use cuda to get around this weirdness? I read the thread and it's supposed to be a single allocation maximum, I can't seem to do any realistic values for scrypt on the gpu as it wants me to go up to 6 for the --scrypt parameter thing.

The binaries provided are opencl only from what I saw when I downloaded it, to get cuda I'd have to find some really old version of hashcat I guess then right?
#10
Because CUDA doesn't run on AMD GPU, but OpenCL does