https://hgpu.org/?p=7533
Efficient Intranode Communication in GPU-Accelerated Systems