Is there any high-speed beta code for nt-opencl, netntlmv2-opencel and mssql05-opencl?

[Richard Miles]

Hi guys,

I love john the ripper rules and in my opinion they are better in
comparison with hashcat rules, and I would like to take benefit from my GPU
card (opencl), however I noticied that the speed of nt-opencl is the same
of one single processor of my box, which is pretty slow. Also, I was unable
to find opencl modes for netntlmv2 and mssql05 algorithm. Is there any
working in progress? Or any beta code that I could try?

Thanks.

[magnum]

Post by Richard Miles
I love john the ripper rules and in my opinion they are better in
comparison with hashcat rules, and I would like to take benefit from my GPU
card (opencl), however I noticied that the speed of nt-opencl is the same
of one single processor of my box, which is pretty slow. Also, I was unable
to find opencl modes for netntlmv2 and mssql05 algorithm. Is there any
working in progress? Or any beta code that I could try?

Some would call everything in my GitHub repo beta...

The "bleeding-mask" branch currently lags behind bleeding-jumbo but it
has candidate generation on GPU for some fast formats (including NT),
although only when using mask mode. I think it supports some combination
of wordlist + mask though. Sayantan just picked up where he left so
hopefully it will evolve soon!

We do have a GPU version of "NETNTLMv2", but it's called ntlmv2-opencl
just to confuse people ;-) ...or rather because I never quite understood
the reasoning behind prefixing it with "NET".
Anyway it's slower than Hashcat but on the other hand it fully supports
Unicode as opposed to the tiny ISO-8859-1 subset Hashcat and others support.

MSSQL05 and others will be implemented as soon as that mask mode branch
has merged to our main branch. There's no point in implementing more
fast formats before that.

magnum

[Richard Miles]

thanks for answer.

By masking + wordlist do you mean the same masking mode used by hashcat?
What about the tradicional and good rules mode from john? :)

So, is this one (https://github.com/magnumripper/JohnTheRipper.git) which
supports generation on GPU? For netntlmv2 is it supported as well?

LOL, it may be weird, but just to keep compatibility with the old legacy,
why not rename the ntlmv2-opencl to netntlmv2-opencl?

Just for curious, why not? I mean, the speed that my computer is able to
read the wordlist file is much faster in comparison with my GPU and
wordlist with rules on john...so, why it's not worth?

Thanks and sorry if I misunderstood some points.

Post by magnum

Post by Richard Miles
I love john the ripper rules and in my opinion they are better in
comparison with hashcat rules, and I would like to take benefit from my GPU
card (opencl), however I noticied that the speed of nt-opencl is the same
of one single processor of my box, which is pretty slow. Also, I was unable
to find opencl modes for netntlmv2 and mssql05 algorithm. Is there any
working in progress? Or any beta code that I could try?

Some would call everything in my GitHub repo beta...
The "bleeding-mask" branch currently lags behind bleeding-jumbo but it has
candidate generation on GPU for some fast formats (including NT), although
only when using mask mode. I think it supports some combination of wordlist
+ mask though. Sayantan just picked up where he left so hopefully it will
evolve soon!
We do have a GPU version of "NETNTLMv2", but it's called ntlmv2-opencl
just to confuse people ;-) ...or rather because I never quite understood
the reasoning behind prefixing it with "NET".
Anyway it's slower than Hashcat but on the other hand it fully supports
Unicode as opposed to the tiny ISO-8859-1 subset Hashcat and others support.
MSSQL05 and others will be implemented as soon as that mask mode branch
has merged to our main branch. There's no point in implementing more fast
formats before that.
magnum

[magnum]

Post by Richard Miles
By masking + wordlist do you mean the same masking mode used by hashcat?

I suppose it's basically like Hashcat's "hybrid attack" but I have tried
neither.

Post by Richard Miles
What about the tradicional and good rules mode from john? :)

They are generated on CPU. This might change some day but don't hold
your breath waiting for it.

Post by Richard Miles
So, is this one (https://github.com/magnumripper/JohnTheRipper.git) which
supports generation on GPU? For netntlmv2 is it supported as well?

It's the (non-default) bleeding-mask branch of that repo, yes. It's
quirky but it's pretty fast.

Post by Richard Miles
LOL, it may be weird, but just to keep compatibility with the old legacy,
why not rename the ntlmv2-opencl to netntlmv2-opencl?

Like I said, I consider netntlmv2 a bogus name. The CPU format should be
renamed to NTLMv2, if anything.

Post by Richard Miles
Just for curious, why not? I mean, the speed that my computer is able to
read the wordlist file is much faster in comparison with my GPU and
wordlist with rules on john...so, why it's not worth?

You suck royally at quoting. The reason there's no point in implementing
fast formats until we have GPU-side candidate generation is they won't
be any faster than the already existing CPU formats.

magnum

[Richard Miles]

I suppose it's basically like Hashcat's "hybrid attack" but I have tried
neither.

What about the tradicional and good rules mode from john? :)

They are generated on CPU. This might change some day but don't hold your
breath waiting for it.

Is there any incentive we could do to help? Maybe donate some money for
this specific task? :)

You suck royally at quoting. The reason there's no point in implementing
fast formats until we have GPU-side candidate generation is they won't be
any faster than the already existing CPU formats.

LOL! Sorry :)