Hi ,
I am wondering that somehow is it possible to read/write host data from kernel while running? Or maybe some features of cuda is doing this consciously?
Thanks in advance
Hi ,
I am wondering that somehow is it possible to read/write host data from kernel while running? Or maybe some features of cuda is doing this consciously?
Thanks in advance
You could make use of CUDA’s zero-copy feature that allows the device to read directly from pinned host memory. Note that these accesses go across the PCIe bus, which means high latency and low bandwidth, so this may or may not be a realistic option depending on your use case.
Thank you for answer. Yes you are right. is it make sense for numa devices?