https://hgpu.org/?p=969
GPU support for batch oriented workloads