Don't have Telegram yet? Try it now!
https://hgpu.org/?p=5898
CudaDMA: Optimizing GPU Memory Bandwidth via Warp Specialization