4 ms·
Dedicated single-purpose hardware with models would be even less energy-intensive. It's theoretically possible to design chips which run neural networks and ali
by WhyNotHugo 1y ago
Dedicated single-purpose hardware with models would be even less energy-intensive. It's theoretically possible to design chips which run neural networks and alike using just resistors (rather than transistors).
Such hardware is not general-purpose, and upgrading the model would not be possible, but there's plenty of use-cases where this is reasonable.
- amelius 1y agoBut resistors are, even in theory, heat dissipating devices. Unlike transistors, which can in theory be perfectly on or off (in both cases not dissipating heat).
- divamgupta 1y agoThe thing is that the new models keep coming every day. So it’s economically not feasible to make chips for a single model
- regularfry 1y agoIt's theoretically possible but physical "neurons" is a terrible idea. The number of connections between two layers of an FF net is the product of the number of weights in each, so routing makes every other problem a rounding error.