See UDN Thread: Possibility for reducing loading times for morph targets by moving work to editor/cooker/ddc (whichever is appropriate)
Loading times: Our model has ~70k verts and ~900 morphs – On device, the FSkeletalMeshLODRenderData::InitResources routine will take about 30 seconds. Looking through the code, it seems to be performing a couple of things – consolidating the vertex/morph permutations by Tmap-ing(on 900-bit key [Image Removed] ) and then sorting/arranging for better cache. Not sure I follow what DivideAndConquerPermuations is doing but since it's called per-permutation (~11k) it's another chunk of time – is there a big runtime win here?
so... can these computations be performed during cook instead of per-load?
This does not need to be done anymore because this computation has been moved to the GPU instead.
There's no existing public thread on this issue, so head over to Questions & Answers just mention UE-56348 in the post.