https://hgpu.org/?p=4500
Hardware accelerated symmetric condensed node TLM procedure for NVIDIA graphics processing units