STATIC PARTITIONING FAILED WITH NVIDIA MPS ON JETSON ORIN NANO

hello there,

i managed to successfully launch NVIDIA MPS on my Jetson Orin Nano. the MPS control daemon starts normally and multiple CUDA processes can share the GPU context

However, when I attempt to enable static partitioning using the commands described in the NVIDIA documentation, the MPS daemon fails to launch.

When starting the daemon I get the following output:

CUDA MPS Control for Linux/Unix.
Copyright (C) 2003 - 2024 NVIDIA Corporation.
This program allows multiple CUDA processes to share a single GPU context.
For information on how to use the CUDA MPS, see the nvidia-cuda-mps-control manual page.

According to the documentation, MPS static partitioning should work on Ampere or newer architectures. The Jetson Orin Nano GPU is Ampere-based, so I expected this feature to be supported :)))

has anyone encountered the same problem? Is MPS static partitioning not supported on Jetson Orin Nano? are there additional requirements or configuration steps specific to Jetson devices?

any help will be apperciated :)

here is what i used:

https://docs.nvidia.com/deploy/mps/when-to-use-mps.html#static-sm-partitioning

Hi,

Could you share a sample to reproduce the error when using static partitioning?
We want to reproduce this issue locally to gather more information.

Thanks.

Hello, thanks for replying,

i just executed nvidia-cuda-mps-control -d -S and it is not being recognized (the -S // –static-partitioning flag is unknown to the system)

with no static partitioning:

nvidia-cuda-mps-control -d 2>&1; echo “exit code: $?”

returns: exit code: 0

with static partitioning flag:

nvidia-cuda-mps-control -d -S 2>&1; echo “exit code: $?”

returns the following:

CUDA MPS Control for Linux/Unix.
Copyright (C) 2003 - 2024 NVIDIA Corporation.

This program allows multiple CUDA processes to share a single GPU context. For information on how to use the CUDA MPS, see the nvidia-cuda-mps-control manual page.

exit code: 1