Concurrent task execution from multiple processes on Jetson TX2

Since CUDA Multi-Process Service is not available on the Jetson TX2, are there any other features that will enable concurrent task execution from multiple processes on this platform?

Ordinary threads are supported the same on this as any desktop PC Linux (e.g., see “man pthread_create”). You won’t get nearly as much power out of an ARM core as you would out of a desktop PC core.


Want to confirm first: your task is a CUDA job, right?

Yes, I meant multiple CUDA jobs/kernels being launched from multiple “CPU” processes.


You can launch multiple CUDA jobs at the same time.
If you want these jobs executed concurrently, please start them with different CUDA stream.

Here is a sample to demonstrate multple CUDA stream: