facebook rss twitter

Researchers use GPUs to break passwords

by Pete Mason on 17 August 2010, 12:09

Tags: NVIDIA (NASDAQ:NVDA)

Quick Link: HEXUS.net/qazlq

Add to My Vault: x

The advent of GPGPU technology has allowed video cards to be used for a lot of different things, including scientific research.  This week, a team from the Georgia Tech Research Institute has announced that it has been putting that graphics horsepower to work at cracking passwords. 

As it happens, they're rather good at it, too.

The results of the research showed that the considerable parallel-processing power of today's high-end video cards could easily break through shorter passwords using brute-force methods.  While trying every possible combination of characters would normally be very laborious, the massively-parallel nature of graphics cards makes the task much quicker and easier.

By taking advantage of programming environments such as CUDA, the team were able to write simple programs in C that could work through character combinations and guess passwords.

According to senior research scientist Richard Boyd, "Right now we can confidently say that a seven-character password is hopelessly inadequate - and as GPU power continues to go up every year, the threat will increase."

In fact, the team suggested that anything less than twelve characters would not be sufficiently safe going forward.  Obviously longer and more complex passwords will be more secure, as they will take longer to crack using brute-force.  In the future, the researchers believe that the best solution will be to use entire sentences instead of single words.

Unfortunately the researchers didn't reveal exactly which GPU they used.  However, a GTX 295 hiding in the corner of the press shots points to NVIDIA's last-gen flagship being responsible.  Given the advancement in processing power and multi-GPU scaling, we have to imagine that the latest hardware would be even more of a threat to simple passwords.



HEXUS Forums :: 13 Comments

Login with Forum Account

Don't have an account? Register today!
No actual specifics on how long it took them to break passwords of specific lengths?

Presumably this sort of attack is only useful for encrypted files and the like - with things like online banking/email accounts, Windows logon, etc immune? Being as how all of these will likely block access for minutes or hours, or provide images to verify you're not a machine.

Interesting stuff though.
Any properly thought-out encryption software will do several, even hundreds of rounds of hashing on passwords to counter passphrase bruteforcing. And as above, it's fairly useless against websites because of the time it takes to refresh pages and the fact you will be locked out of most sites for trying a few thousand passwords. Maybe it's useful for opening passworded Word documents or something? Other than that it's a bit of a non-issue really, just a bit of scaremongering thought up by a few GPU programmers…

Edit: I've just found what I think is the original article here, and firstly I lose a lot of respect for anyone who claims to be a computing expert then uses teraflop as the singular version of teraflops, but they also hint towards this cutting down time to break a password from days to hours. TBH, if your security setup takes just days to crack there's something catastrophically wrong with it in the first place. As I said, any good encryption software should do some form of key strengthening so prevent attacks like this.
Think the article is more about the fact that GPUS are capable of doing such a thing as code breaking. Doesn't really state there is a practicle use for it, other than throwing theories around. :geek:
Hmmm, they specifically mention CUDA..
Not that I'm suspiscious like..
I'd hazard a guess from my own GPU experience and just used the GPU as a random number generator, which generated the possible passwords. Then offloaded the ‘attempts’ section to the CPU, because that cannot be done on the GPU. A 260 will happily generate 2 billion random numbers a second, depending on your algorithm (the 2b/sec was a SOBOL I think).

But you're right, any system which locks you out after x attempts will be useless, or any system which requires a round trip of latency ping will also be useless to brute force.