Don't have Telegram yet? Try it now!
https://hgpu.org/?p=18341
Improving tasks throughput on accelerators using OpenCL command concurrency