https://hgpu.org/?p=10583
Pipeline strategies to accelerate range query processing on a multi-GPU environment