DMA between GPU and other peripherals

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi, all
Is it possible to perform DMA between GPU and other peripherals (hard disk, NIC, and so on) directly (bypassing main memory)?
Thanks!

2

This is frequently requested, but NVIDIA employees have only said they are “considering” it for future CUDA releases. No promises, unfortunately.

3

Thank you very much! Is there any other API that can do this?