Recently, research students at Georgia Tech released a report outlining the dangers that GPUs pose to the current state of password security. There are a number of ways to crack a password, all with their different pros and cons, but when it comes down to it, the limiting factor in all of these methods is processing complexity. The more operations that need to be run, the longer it takes, and the less useful each tool is for cracking passwords. In the past, most recommendations for password security revolved around making sure your password wasn’t something predictable, such as “password” or your birthday. With today’s (and tomorrows) GPUs, this may no longer be enough.
Although the article never mentions them by name, the newest tools in password cracking are based around two tools, nVidia’s CUDA and AMD’s Stream SDKs. These tools allow programs to be written in C that can be broken up and utilize the parallel nature of the hardware that is usually optimized for graphics. GPUs are much better at large-scale mathematical operations than CPUs because of this parallel layout. Chances are, if you have a somewhat recent graphics card, it is probably compatible with either CUDA or Stream, and if you already know C, you have all the tools necessary to get started.
The lesson to learn here, the longer or more complex a password is, generally the safer it is. Because of this, a number of tools, both software and hardware, may become more and more popular, or necessary, to accommodate this need.
ruh-roh
“GPUs are much better at large-scale mathematical operations than CPUs because of this parallel layout. “.
This is generally wrong. This is true only for mathematical operations which can be splitted into a large number of small independant computations, which is not at all the most common situation.
Password cracking is not exactly the typical “large scale mathematical operation”. The most typical ones are operations on large matrices, for which GPU computing is often not suitable.
Pyrit password “testing” program has a CUDamA version that really shows the great speed benefit of GPUs
Well, implement a super simple 3 wrong pass’s and need to wait 30 min or 1h, there goes the super hiper mega fast bruteforcing, even 4 letters passwords would take days.
Yes, I’m reviving a long-dead comment thread. Still.
I believe you misunderstand the purpose of GPU hash-cracking. In order to crack a password via this method, you need to already have the hash of the password that is stored in the system in question. The brute-force part comes when you un-hash the password: un-hashing takes a LOT of power.
@Tiago
But then it would be super-easy to do a DDoS attack on you
@Tiango
Think about WPA/WPA2 Passwords.
You can listen to the data sent over the WLAN for a few hours.
Then take this Dump to your CUDA application and let it crack the password within a very short time.
Then go back to the WLAN hotspot and type in the password and you have full acces with only 1 TRY !!!!
Just scary.
i have dual nvidia GTX 480s in sli in my desktop and it makes rendering and film processing a breeze so password cracking would be simple no?
i mean brute force attacking and hash cracking is nothing more than a bunch of small not so complex tasks and with over 900 cores per GPU than it should speed things up right?
WPA cracking is a poor example. You still have to “listen to the data sent over the WLAN for a few hours,” so cracking gramma’s wifi password isn’t the bottleneck, the acquisition of data remains the hurdle.
Are there any opensource cuda/stream crackers out there? They’d be fun to play around with.
As long as people pick good passwords that aren’t going to be on a wordlist, things are pretty secure. A plain old bruteforce attack will still take a very long time.
I sometimes think about the fact that, given enough time, all these encryptions will be a computational breeze to bruteforce. I could just sniff your encrypted data, wait 10 – 15 years, and get your secret then. It usually doesn’t matter, since your secret probably won’t be very important after 10 years. I do, however, imagine it should matter to governments and people who like to send files called, “encrypted_pics_of_my_gf.rar”.
It’s not that hard to outpace technology with passwords.
Using a very simplistic (and generally incorrect, but whatever) assumption with Moore’s Law, we expect computing power to double every 18 months.
On the other hand, if we add ONE random letter/special character/number, the time to brute force your password is multiplied by about 40.
The same applies for hashing and encryption algorithms. Adding more bits is relatively easy, but it makes cracking a whole lot harder.
http://www.golubev.com/hashgpu.htm
Using GPU’s to crack passwords is old news.
@NatureTM
@biozz
People who use encryption for government-type secrets or even just people who know what they are talking about (though I do not claim to be one of them) design cryptography algorithms (hashes, symmetric and asymmetric cryptography, etc.) so that not only the computational power for current computers is far to weak, but they also look into what could be possible in the future.
Given a pure brute force attack and no further breakthrough in CS/mathematics (no attacks in the algorithm allowing to select a subset of possible plaintexts without actually testing them for example), there is such a thing as a “big enough key” for a given algorithm, so that the computational power of silicon CPU will NEVER reach a point at which it can brute force it in reasonable time, no matter how big and fast you make it.
This kind of calculation is generally not done vs. the latest i7, GTXblabla, …, but more in terms of total energy and time needed compared to “big values”, in the order of the age of the universe. So, like I said, without breakthroughs and research, you can’t just throw computational power at brute force and hope it will solve itself…
For those who are interested, I’d recommend a book like Bruce Schneier’s Applied Cryptography, which has everything from the basic concepts to the theory and implementation, though I’m sure the same information can be found on the net somewhere.
@Alexander Rossie
how is this “old news”? its only been 2 years seance it became popular and only recently became a concern
tell me whats the new and better way to crack passwords .. im all ears
@lolzertank yeah, it is easy for encryption to out pace technology, BUT as someone else already pointed out, that only works with current encryption technology — if I sniff your data now, and crack it in 10 or 20 years on my cell phone, I still have it cracked — though — as was also already said, it’s probably not going to be worth anything in that time.
As NatureTM and lolzertank pointed out, hashed passwords are still pretty secure because at a length of say 20 it will take quite a long while to crack a password.
Even with time/memory tradeoff of rainbow tables you’ll need several terabytes of storage and many many GPU’s to generate 20 character passwords in a reasonable length of time.
@biozz
Two years is old news. If I told you that I just started my job.. two years ago. You would wonder why I am telling you this now, because it is old.
The ability and relative ease of using GPUs for encryption breaking has been known and used far longer than two years ago.
It’s the reason why the US gov’t get’s all flustered when a connex container full of playstatons and video ipods are shipped to certain countries that don’t like us.
The developement was on the software side of things with the release to the public of SDKs that lower the implementaion hurdle considerably.
I hope people start releasing programs that take advantage of the power of GPUs. Even those not aimed at encryption breaking would benefit.
@Spork
thats a poor analogy if you told me 2 years ago than told me again its old news
if your telling me for the first time than its just news that’s a one time event this is an ongoing event
there not telling us that it is possible its telling us that people are doing it NOW and it poses a security threat NOW and the article is only a few weeks old
There is a hidden point that’s being missed here.
It’s true, superfluous characters added to a password can make it quite more difficult to brute force, and an arbitrarily long password can possibly make a brute force attempt futile, BUT —
if an arbitrarily long password is changed at an arbitrarily small interval, any brute force attempt, no matter how fast (up to a certain, currently unachievable point) will be rendered useless, even if it provides the correct secret.
In practical application, if you have a system that generates a 256-bit password at random, at least 1024 times per second, and requires a perfectly synchronized password to gain access to an otherwise implausibly secure computer system, any brute force attack would be mostly useless.
Some cryptos slow on GPUs e.g. Blue Midnight Wish
http://upcommons.upc.edu/pfc/bitstream/2099.1/7933/1/Masteoppgave.pdf
Hak#5 done an interesting article on GPU password cracking. Might be of use for those wanting to know more.
If you’re thinking about cracking WIFI passwords, hotmail accounts or even paypal accounts – you’re thinking small time.
Check out this forum thread (particularly post 43)
http://forum.gsmhosting.com/vbb/f83/how-sl3-unlock-codes-calculated-1079689/index3.html
@goldscott
You’re way off. All you need is that 4way handshake. Deauth a client and capture = few minutes at the most.
@payne
You dont need to listen to a Wlan for a few hours in order to be able to crack it, you only need to capture a 4way handshake which can be forced with various ddos or deauth methods.
Once you have the handshake there are various methods used (tables/cowpatty/etc..) which can be used to crack it.
longer and more complex passwords will help fight it off for a while, but eventually its not going to be enough.
@haltux
My understanding is that brute force password cracking operations is highly scalable. A naive approach is working on the same cyphertext in parallel with different key ranges across several compute domains.
And why do you think matrix operations can not be parallelised? They can be split up into several compute domains in various ways. There is several decades worth of research in this area.
Things have never been so sweet now that OpenCL and CUDA is becoming more popular.
I test-drove a CUDA based RAR password cracker on my system. My single GTX280 achieved over 3000 calculations per second, compared to just 30 calculations per second for a 3.2GHz P4
http://www.distributed.net/Download_prerelease
I used to run the stream client on dual 4870×2’s, and would crunch about 2billion keys/sec… And that was with 3200 stream cores.
It isn’t difficult to put 12800 cores in a box now, so 8billion keys/sec isnt too bad!
further malicious activities on GPUs: http://it.slashdot.org/story/10/09/27/1422205/Malware-Running-On-Graphics-Cards?from=rss
I think one of the biggest flaw of lots of password protected systems are the “security questions”. It’s usually so easy to guess or find the answers that it’s like having no protection at all!
I always write dummy answers to those questions…
@hello @cknopp thanks for the links.
@smoker_dave excellent picture. Who’d of known that clusters aren’t just for researchers any more. I always wondered how they did code unlocking. Looks like a great HAD article in there somewhere.
This is only an issue for encryted data, it is nothing against passwords for simple reasons.
1. Lock out, sorry you wanna try 50 passwords to login? Sorry buddy lock out 24 hours. Good luck brute forcing that.
2. Don’t allow access to the password database file in the first place, once again no brute forcing.
There is no concern for passwords being brute forced online if all the two steps above are done. There is also no concern of said rainbow tables on windows passwords if the said password data base file can’t be accessed. This stuff is a no brainer. Passwords will always be safe with these systems.
SEATEC ASTRONOMY
Not that ANYONE gives a flying **** about AMD/ATI “Streams” — talk about stillborn.
What I’m surprised to not see mentioned is OpenCL, since that’s cross platform.
Though NishaKitty hit it on the head — most sites/apps where security matters locks you out for 24 hours after THREE tries!
… and if you are sending ANYTHING that gives access to the passwords database you’re a retard!
@deathshadow
ATI Stream is not still born… Actually, OpenCL is now a part of it!
And what I want to see is someone harness these GPU’s to crack the unbreakable Nagra3!
@deathshadow
“… and if you are sending ANYTHING that gives access to the passwords database you’re a retard!”
Yeah, if you have a single vulnerability in your website, you’re retarded! How could you not notice that one of the many apps your company runs or hosts has a vulnerability in it?
I work for an asset management company interning for web application security, and we have A LOT of applications. It only takes one of them to have a vulnerability to allow people to start mining database information, possibly including hashed passwords. At that point, they don’t really them, but they can get a user’s name, email, all other kinds of information. With that, and if they assume the user uses the same or a similar password for their accounts, they can start really damaging individuals.
@Rob
That’s an already lost situation anyway reguardless of password databases. Layering the services on your server can help with this etc sure this is quite obvious.
I only read the short comments.
@goldscott
You can get a WPA/WPA2 hash in about 5 min or less. Use airepley to deauth the client, when they reconnect you get the hash. It takes maybe a minute.
The people who really should be shaking in their boots are the people that use the default passwords on ATT’s 2WIRE modems as they have a 10 digit password. Without GPU cracking it would take forever to crack, with GPU cracking (pyrit) crunch to generate the numbers (so you dont need a huge password file) you can be done in a couple days.
In any decently written web application, even with a full copy of the database brute forcing just is not an option without also having file access, which usually means physical access, and at that point all is lost anyways. I’ve been writing a CMS for personal use, with my salting and hashing routine each password has around 100 trillion permutations per password before you even get to try breaking the hash, and thats if you’ve happened to grab the salt lists without them it’s around (turns out I can’t tell you, my online permutation calculator responds with “INFINITY”) I doubt that without physical access the password “cat” could be brute forced, add to that that each site has there own different method of salting, and getting a job at the web host seams like the easy way.
Someone should use the methods above to decypher the Zodiac Killer cyphers that are still undecrypted.
well, on a more positive side of this, we’ll probably be seeing some games that take advantage of extra GPUs pretty soon. even if not, just pop in another GPU, write a driver that registers it to windows/linux/mac as an extra CPU, then there you go, high speed games!
can any one post cuda suppored gui password cracker?
I think we are all overlooking the implications this has in regard to cyberwarfare. According to my calculations, the Quadro 6000, nVidia’s top of the line simulation and hardcore 3D rendering solution, has an effective speed approaching 500GHz.
so… what about rainbow tables? and table generation?
what if you could get the cpu + gpu(s) to work together? i.e. cpu working on part 1 of 2 and gpu working on part 2 of 2 etc.
Considering elcomsoft was doing this back in 2007 and continue to apply it to new GPUs and encryption processes it is old news. They wrote a paper to increase awareness of this “vulnerability” for people who have had their head buried in the sand for the past 3 years. If waving your hands and shouting about something that happens is the best you can do as a research scientist we are screwed, but I already had the figured when the picture showed three white guys instead of an asian.
Now if some hacker group would apply this to a certain satellite tv network it would make my day. I have no interest in using something like this to run malware on a random strangers computer.
Now you have ighashGPU, a GPU hash cracker. My Radeon 5770 does 3.3 Billion passwords per second.
wowww nokia SLI ati amd 790 core!!!!! is very fasstt