Hacker News
new
past
comments
ask
show
jobs
points
by
theLiminator
5 hours ago
|
comments
by
tjwebbnorfolk
5 hours ago
|
[-]
Inefficient per watt, yes, but local inference capacity is greatly underutilized in aggregate. If a model can run on a machine that already exists, that's a bunch of additional chips that don't need to be built.
reply