@openrisk @humanetech I’m no expert but this idea is interesting to me. I would hypothesize the opposite - I think it’s generally more efficient to share resources on one larger machine than divide it up - you would have more resources wasted by idling. This seems apparent with memory, but maybe there’s more nuance with electricity.
@jaredmoody @email@example.com its an optimisation problem of sorts and to solve properly one must indeed figure all the important elements. Losses in idling infrastructure would be definitely adding on the penalty side of extreme decentralization. On the other hand the gains would probably be quick to recover if one had, say, neighborhood shared clusters pooling compute for ~100 or so.
Another interesting angle is whether cooling is more efficient with less aggregation