https://hgpu.org/?p=8901
Using Hybrid Shared and Distributed Caching for Mixed-Coherency GPU Workloads