It's not too bad - I run 6x RTX 3090s on a 2nd-gen Threadripper with PCIe bifurcation cards. The energy usage is only really bad if you're training models constantly, but inference is light enough.

I use 208V power but 120V can indeed be a challenge. The USA has split phase wiring; every house has 220-240V if they need it. Bit of a misunderstanding of how our power works - we have 220-240V on tap, but typical outlets are 110-120V.

▲

flashgordon 5 days ago | parent [-]

Yeah at this point the goal is to see how to maximize for inference. For training it is impossible from the get go to compete with the frontier labs anyway. Im trying to calculate (even amortized over 2 years) the daily cost of running the equivalent rig that can get close to a single claude agent performance. (without needing a 6-digit gpu).

	▲	icelancer 4 days ago \| parent [-]
		Really the only reason to have a local setup is for 24/7 on-demand high-volume inference that can't tolerate enormous cold starts.