CUDA device

Question

CUDA device

Closed this issue 2 years ago · 3 comments

How do we ensure that the CUDA device is properly set for API calls?

Answer 1 · 2022-08-22T07:34:03.000Z

Hi @bashimao , please check here, https://github.com/NVIDIA-Merlin/merlin-kv/blob/master/include/merlin_hashtable.cuh#L169

Answer 2 · 2022-08-22T09:44:32.000Z

Ok, got it. So initialization is done in that device. But for downstream calls to the API, we simply assume the user is smart enough to set the device and provides pointers in the right memory area? Right?

Answer 3 · 2022-08-23T03:48:34.000Z

Ok, got it. So initialization is done in that device. But for downstream calls to the API, we simply assume the user is smart enough to set the device and provides pointers in the right memory area? Right?

Yes