I think it’s not quite as well known or prevalent as other services (as say SSH) so likely doesn’t have anything automated attacking it yet. If you check something like http://shodan.io/ against your ip, I’d guess the service has been found.

Home Assistant likely won’t come under any kind of attack until there’s a very easy to exploit, unpatched zero-day vulnerability in the wild. Given how many people (myself included) who have HA exposed publicly it’s really a matter of time. The best mitigation is not exposing publicly if possible, and staying up to date.

In my case I don’t expose HA over 8123, I have a proxy on 443 where HA is not the default host name, meaning if you don’t use the right host HA doesn’t receive the traffic. As I’d expect that automated attackers wouldn’t what my host is it’s a reasonable layer in the security onion. I don’t expect anything would realistically protect from a targeted attack but I’m also not important enough to be targeted.

You don’t need cards to have full bandwidth, they only time it will matter is when you’re loading the models on the card. You need a motherboard with x16 slots but even x4 connections would be good enough. Running the model doesn’t need a lot of bandwidth. Remember you only load the model once then reuse it.

An x4 pcie gen 4 slot has ~7.8 GiB/s theoretical transfer rate (after overhead), a x16 has ~31.5GiB/s - so disk I/O is likely your limit even for a x4 slot.

• overhead was already in calculations

We can’t ever stop this kind of stuff, but with something like fail2ban you can set it up to block on too many failures.

Really though - ensuring your system is kept up to date and uses strong passwords or use a SSH keys is the best defence. Blocking doesn’t prevent them from trying a few times. Moving SSH to a non standard port will stop most of the automated attacks but it won’t stop someone who is dedicated.