Or low end consumer GPUs which can do over half a billion hashes/sec.
Re:Single-block collision for MD5
by LearnYouALisp on Monday January 30, 2012 @12:20AM
(#3527620)
Not that low-end, right? I think the cheapest "good" one was probably over $100, when I read about "Bitcoin mining" perhaps month or so ago.
$100 is pretty low end for a GPU.
MD5 is many times faster than SHA256, too.
$100 is low end. It's easily affordable for consumer purchase. You no longer need to be the NSA with a budget of $1,000,000 to do this.
These sorts of thing tend to parallelize well too. If it were written for CUDA or similar, I wouldn't be surprised to see it finish in a few hours.
Re:Single-block collision for MD5
by JoachimSchipper on Monday January 30, 2012 @03:32AM
(#3527943)
This is not a straight brute-force attack. I don't know how well it parallelizes, but Marc is aware of the existence of GPUs. (Also, it takes a few weeks. No point in messing around with CUDA...)
"Based on our complexity analysis and a number of computers available for our collision search, we estimated that it would take approximately five weeks. As this was feasible enough, we started the actual search. It was our fortune that a collision was found a bit earlier, namely after only three weeks."
This on a Core2 Q9550, a four year old processor. I wonder how fast this could be done using an EC2 temporary compute cluster.