I’m trying to understand how to use low-level virtual memory APIs.
While I’m reading this post,
there exists the allocation type CU_MEM_ALLOCATION_TYPE_PINNED.
I wonder what does Pinned memory means in device memory.
Inside of document, CUDA Driver API :: CUDA Toolkit Documentation
it says that this allocation can’t migrate.
Does this mean that this memory can’t be copied out to the host memory by cudaMemcpy?


Pinned in this context simply means that this allocation is always located on that specific device. It does not affect memory copies.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.