https://hgpu.org/?p=17577
Out-of-core Implementation for Accelerator Kernels on Heterogeneous Clouds