@tedu ahhh hmmm, I had not considered the actual password hashing as well. I was only testing the connection setup (aka, RSA sign/verify and a bunch of (E)?DH KEX)

My best guess is on a reasonably loaded Intel SP1 system, a SSH connection setup takes 0.007 Joules of power.

A random machine I picked out has 15200 SSH connection setups a day

0.007 * 15200 = 106.4J ~ 0.0000279 kwH

A rough estimate on the amount of accessible SSH servers is around 16,280,000 (based on some scanning stuff from 2 years ago)

16,280,000 * 0.0000279 kwH = 454.212 kwH (a day)

or a constant ish 18.925kw of power.

And that is only assuming one party , so likely 2x, so all of the SSH bruteforces going around is costing around 38kW of power.

on one hand, not that bad (there are some DCs that can put that into a single rack), on the other hand, that's quite a lot of wasted CPU energy

@wolf480pl Yeah "only" 54 Nvidia H100's

It's worth keeping in mind that this final number also extremely depends on the 0.007J per SSH connection number, any real change to that (I think 0.007J is on the low side) will radically change the final KwH for the worse

@benjojo and how much power can be saved by changing the SSH port to a random one. Once one figures out the power consumption, we can claim that IPv6 is officially more power efficient as well, because the bots won't bother to scan it ;)