I am learning TensorRT and PyCUDA to do inference with the TX2.
I would like to know what type of memory I am using with my code?
I think it is the page-locked.
a) Traditional memory copy: mem_alloc (Python) = malloc ©
b) page-locked memory
d) Unified management
It’s page-locked memory.
You can find more information in pyCUDA’s document here:
Is there a function for “zero-copy” for the TX2? I read that it is like: Pinned data transfer. Check the image.
YES. You can use a pinned memory to avoid memory copy.
Some example can be found in our CUDA document:
I guess there is nothing similar in PyCUDA, right?
It should be similar since pyCUDA is wrapping from the C++ function.