Multicore-TSNE
Multicore-TSNE copied to clipboard
Low CPU utilization during second phase
It says in "future work" that phase 2 can be improved, so maybe this is what you're referring to?
On phase 2 I see very low CPU utilization (see attached screenshot). Running Multicore-TSNE on 38 cores only utilize 1 core to 100% while the others lie at around 8%.
Exactly, phase 2 is not paralellized at all.
i am finding that during stage two my CPUs are used in parallel in a periodic manner. If this stage is not parallelized, why do I see this kind of CPU utilization?
Is there parallelization within each iteration? so different nodes are allocated to different CPU within each gradient descent iteration?