AGX Xavier -> Unified Memory questions

Hello Everyone,

I am using an AGX Xavier which has integrated GPU and CPU memory.

If I use unified memory (i.e. cudaMallocManaged) will this result in zero memory copy as I alternate accessing the memory between host and device?

Hi,

Unified memory use same buffer point between CPU memory and GPU memory.
The underlying synchronization is handled by GPU driver.

There is another zero-copy memory (Pinned Memory) available on Jetson.
You can choose it depends on the use case:
https://docs.nvidia.com/cuda/cuda-for-tegra-appnote/index.html#memory-management

Thanks.

This topic was automatically closed 2 days after the last reply. New replies are no longer allowed.