You can cycle the smaller drives to cold backup, that's not a waste. You do have backups, which RAID is not, right?
MalReynolds
joined 2 years ago
Sure, works fine for inference with tensor parallelism, USB4 / thunderbolt 4/5 is a better (40Gbit+ and already there) bet than ethernet (see distributed-llama). Trash for training / fine tuning, that needs higher inter GPU speed, or better a bigger GPU VRAM.
Seems like data integrity is your highest priority, and you're doing pretty well, the next step is keeping a copy offsite. It's the 3-2-1 backup strategy, 3 copies, 2 media (used to mean CDs etc but now think offline drives) 1 offsite (in case of fire, meteor strike etc), so look to that, stash a copy at a friends or something.
In your case I'd look at getting some online storage to fill the offsite role while you're overseas (paid probably, but a year of 1 or 2 Tb is quite reasonable) leaving you with no pressure on the selfhosting side, just tailscale in, muck around and have fun, and if something breaks, no harm done, data safe.
I've done it for what seems like forever and I'd still be worried about leaving a system out of physical control for any extended period of time, at the very least having someone to reboot it if connectivity or power fails will be invaluable, but talking them through a broken update is another thing entirely, and you shouldn't make that a critical necessity, too much stress.
You don't say what else other than gaming you want to do, so it's harder to advise. Still, consider bazzite KDE, easy, stable, relatively close to the bleeding edge without all the cuts, everything you need for gaming in the tin.
Perhaps not saved, but I'd venture the most significant nail in the coffin of the scientific publishing mafia so far, pursued with integrity and honor. The rise of open publishing that followed is very telling, and in my mind directly attributable to Alexandra's work and it's popularity, they know they need to adapt or (probably and) die.
Still need to work on the publish or perish mentality, getting negative results published, and getting corporate propaganda out of the mix, to name a few.