That's 7.2 billion hashes per second per single Radeon GPU for md5. For 8 character password with numbers and letters that's 8.5 hours max and 4.25h on average.. The numbers only got better since then, I expect it's at least halved for this year's hardware. (edit: 1080ti does 32GHps, so yeah... make that 1h on average https://www.servethehome.com/password-cracking-with-8x-nvidi... )

Assumptions, not specified in the article:

- The salt is stored with the hash (this is a standard approach)

- It's only a trivial implementation with a single round of MD5. If they used multiple rounds or PBKDF1, then the hash choice doesn't matter as much.

Looking at phpass (one of the md5 algorithms), a high-end GPU can do 7M hashes per sec.

The bigger issue is if they were actually using raw MD5 (which is sadly quite common), which is benchmarked at 25 billion (with a B) hashes per sec per GPU.

The fact that the article says “believed to be” strongly suggests that things are not as simple as they’re “believed to be”, because if the passwords were easy to crack that’d be trivial to prove.

Usually the attacker will also know the salts in a breach of this type, unless the company did something clever with the salts (doubtful since they used MD5).

We don’t know how the passwords are hashed. All we have is a journo guessing.

5 years ago, a run of the mill gaming PC could crack an MD5 hash in a reasonable amount of time. Worst case, you'd have to let it run overnight.

One of the big issues is collisions. You might not find the original key, but you'd find something that hashed to the equivalent output.

AFAIK the best current preimage attack against MD5 gives you a complexity of 2^123.4. Even if you had every computer in the world working on this you'd never succeed.

>One of the big issues is collisions. You might not find the original key, but you'd find something that hashed to the equivalent output.

This is false. There does not exist a feasible preimage attack against MD5.