4 ms·
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.
by snovv_crash 7mo ago
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.
- xavxav 7mo agosearch trees tend not to scale well to higher dimensions though, right? from what I've seen I had the impression that Yinyang k-means was the best way to take advantage of the sparsity.
- snovv_crash 7mo agoMost 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.