A uniform set of points with a dimension 128 and type cube. That has a size of 50 ** 3. Can be queried for a nearest neighbor at a speed of 10 queries per second with limit varying from 1 to 25.
It works better than when no index used at all. So gist gives here a speed up.
The documentation of postgresql doesn't mention complexity of such an index. I've got confused as to its speed.
Does postgresql allow for an approximate nearest neighbor search?
https://github.com/erikbern/ann-benchmarks has a lot of efficient implementations.