CUDA IPC Low Level VMM, cuMemImportFromShareableHandle returns invalid device ordinal

Accelerated Computing CUDA CUDA Programming and Performance
1

Hey there,

I am using ros2 to send the handle exported from cuMemExportToShareableHandle to the subscriber. When I tried to import the handle using cuMemImportFromShareableHandle, it returns “invalid device ordinal”. This error only happens during inter process communication between the publisher and and the subscriber node, but not during intra process communication.

I am under the impression that this API can be used for IPC. Does anyone know what could be the reason and is there anyway to solve this?

Thanks in advance

CUDA11.4
Ubuntu 20.04

1 Like
2

Facing same issue!