My application needs to launch many processes in parallel that use NPP function. I found that the maximum number of processes can be started is 48 on 1 GPU before NPP fails.
A NPP context will be created for each process.
Is 48 the magic number for the maximum?
OS: Ubuntu 18.04
CUDA: 10.2 and 11.
GPU: Tesla T4
The GPU has 15 GB, there are still a lot of memory available.
±------------------------------±---------------------±---------------------+
| 1 Tesla T4 On | 00000000:13:00.0 Off | 0 |
| N/A 51C P0 27W / 70W | 4694MiB / 15109MiB | 0% E. Process |
| | | N/A |