Comment by kgeist

Are there vector DBs with 100B vectors in production which work well? There was a paper which showed that there's 12% loss in accuracy at just 1 mln vectors. Maybe some kind of logical sharding is another option, to improve both accuracy and speed.