For CPU with bigger K you would put the centroids in a search tree, so take advantage of the sparsity, while a GPU would calculate the full NxK distance matrix. So from my understanding the bottleneck they are fixing doesn't show up on CPU.
replysearch trees tend not to scale well to higher dimensions though, right?
replyfrom what I've seen I had the impression that Yinyang k-means was the best way to take advantage of the sparsity.
Most data I've used is for geospatial with D<=4 (xyzt) so for me search trees worked great. But for things like descriptor or embedding clustering yes, trees wouldn't be useful.
reply[dead]
reply