3.14.by forum

OpenCL changes the dev environment a bit. However, atom of oclhashcat claims the nVidia OpenCL runtime is massively buggy.

What type of work are you doing? For crypto, ATI is often faster because it has a hardware barrel rotator. For other stuff, it just depends.

1. You don't understand rainbow tables if you think they're pass:hash pairs.
2. Have you looked at the size requirements for your proposed system with any reasonable password coverage? There's a reason it's not done.

Is installing a 64-bit Linux that much trouble? The modern Ubuntu installers have support for shrinking a Windows partition and installing alongside. The Cryptohaze GRTs may eventually be supported under Windows, but my search currently relies on mmap, which is not supported on Windows (at least not...

You may have luck with the Cryptohaze multiforcer - you can use the per-position character set support to emulate this type of behavior, at least for short salts/passwords. In this case, you'd set the following per-position character sets: 0: t 1: e 2: s 3: t 4: [full US charset/etc] 5: [full US cha...

Are you discussing the dictionary mutation tables, or something different?

0.55 is out with NTLM/MD4 support as well as MD5. Over at cryptohaze!

http://cryptohaze.com/gpurainbowcracker.php

This is, to my knowledge, the first ground-up rainbow tables implementation using CUDA.

The tables are NOT compatible with the existing rcrack tables - I use a different (much faster) reduction function.

Have fun!

Impressive, though I question the CUDA-Multiforcer compare - if you're using 0.7, I think the autotuner is broken (so non-optimal block/thread sizes are being used), and you should probably be using the '-l' switch for 500k MD5 hashes, as it improves speed on large hashlists dramatically. I may test...

I'd add that the Cryptohaze multiforcer 0.7 has this feature.

^++ You can't directly compare the strength of symmetric encryption algorithms (DES, AES) with public key algorithms (RSA) - the bit count does NOT in any way correlate between the two. AES256 (256 bit key) is theoretically secure well, well beyond feasible futures if the key is kept secure. RSA512 ...

Does this attack a single hash, or full hash lists?

It's far more complex than that. It's some initial operations, then 1000 iterations of a fairly annoying algorithm that does some string concatenation (GPUs are awful at this) depending on which operation you're on (using modulus, which GPUs are really bad at too). Then it does some final operations...

Cisco IOS MD5 is just MD5CRYPT with a short hash value, IIRC.

No GPU brute forcers that I'm aware of support this yet.

The algorithm really kind of sucks for GPUs.

in regards to a crossover point - how could reversal NOT be beneficial? Even if you re-reverse all hashes quite frequently, wouldn't it still be faster than computing the entire hash algorithm for every plaintext? Re-calculation of the last round of MD4 every 26^2 tries should still be less expensi...

I believe this is a known issue with Vista - it only makes GPUs with displays attached available to CUDA.

Another solution would be to install Linux or XP for cracking - they will both outperform Vista significantly due to dramatically reduced kernel launch times.

I am curious as to how the current events in the USA (massive bailouts, nationalization of the auto industry, "spend now, tax the rich later," etc) are viewed by those who have lived through such things in Russia over the past 40 years or so. It seems to be a very similar road that Russia tried and ...

There's no way a single 4890 should be beating a 4870x2 by a factor of 2.5 - it should be around 60% the performance or so, or around 1200M/s

What driver revision do you have installed?

budden wrote:I can't run it on my XP x86 CUDA2.2 station with 8800GTS 512mb - could you please check and report if it works?
In that case please specify your CUDA version, drivers, OS and GPU. On Vista it works fine AFAIK.

I believe he has fixed the XP issue and will be releasing a revision shortly.

http://3.14.by/forum/viewtopic.php?f=8&t=30 Go there.

I'd add that Cryptohaze as a very nice Windows/Linux NTLM brute forcer - check the forum for the 0.7 binaries, as it's a LOT faster than the 0.6 ones on the front page.

I noticed the branching switch statement in your md5 code. If you are interested, I have a method that places the 0x80 in the correct place without branching, at the cost of about 20 more ops. I am afraid I don't have the code on me, but I can get it to you soon if you are interested. I've tested i...

Macs for desktops, the real servers doing the CUDA work are Linux servers in the other room. I'll try to get a pic of my research area & the i7 box too.

Since it's buried in the other post:

http://www.cryptohaze.com/releases/CUDA ... .7.tar.bz2

It should compile if you put it in the projects directory of the CUDA SDK on a 64-bit Linux box. Beyond that, you're on your own.

Discussion over at the cryptohaze forums as well.

Eh, comments welcome.

http://www.cryptohaze.com/releases/CUDA ... .7.tar.bz2

Comment here, or over on the cryptohaze forum.