undefined | Better HN

0 pointsmoyix2y ago0 comments

Since that post was published, the 4090 came out, which can (according to this hashcat benchmark [1]) do 50,638.7 million SHA1 hashes per second, so now it would only take a single 4090 GPU 11.55 years. Or you could buy 12 of them and do it in a year, etc. So it's definitely not cheap but 15 years is definitely an overestimate (and presumably GPUs will keep getting faster...).

SHA2-256 is "only" 21975.5 MH/s so you'd have to double the number of GPUs or amount of time.

[1] https://gist.github.com/Chick3nman/32e662a5bb63bc4f51b847bb4...

0 comments

gruez2y ago

Generating 2^64 hashes isn't guaranteed to produce a collision, and even if a collision did exist in that set, you're not going to find it by getting a bunch of GPUs to compute 2^64 hashes. There's a huge difference between a haystack that maybe contains a needle, and a needle that's been pulled from the haystack and presented to you. To actually find and identify the collisions you'll need to hook those GPUs up to some sort of storage/retrieval system. Just to store 2^64 128-bit hashes would take 295.1 exabytes. That's an order of magnitude more storage than NSA's utah datacenter[1].

[1] https://en.wikipedia.org/wiki/Utah_Data_Center

moyixOP2y ago

You can generate pairs of hashes for random inputs and check for collision without storing all of the outputs, no?

adrian_b2y ago

True, which is why 256-bit or bigger hashes are recommended to make sure that the time for finding a collision is alone great enough to make this impossible.

Finding a collision for an 160-bit hash by brute force, on a big cloud or supercomputer, is about the maximum that someone with unlimited financial resources could do.

gruez2y ago

That requires way more hashes to be computed though.

1 more reply

j / k navigate · click thread line to collapse

0 pointsmoyix2y ago0 comments

SHA2-256 is "only" 21975.5 MH/s so you'd have to double the number of GPUs or amount of time.

[1] https://gist.github.com/Chick3nman/32e662a5bb63bc4f51b847bb4...

0 comments

gruez2y ago

[1] https://en.wikipedia.org/wiki/Utah_Data_Center

moyixOP2y ago

You can generate pairs of hashes for random inputs and check for collision without storing all of the outputs, no?

adrian_b2y ago

True, which is why 256-bit or bigger hashes are recommended to make sure that the time for finding a collision is alone great enough to make this impossible.

Finding a collision for an 160-bit hash by brute force, on a big cloud or supercomputer, is about the maximum that someone with unlimited financial resources could do.

gruez2y ago

That requires way more hashes to be computed though.

1 more reply

j / k navigate · click thread line to collapse