Inquiries Regarding NVIDIA's Support for PREEMPT_RT

**• Hardware Platform: (GPU) P100 (Old GPU)
**• DeepStream Version: 6.3, 6.4
**• NVIDIA GPU Driver Version: nvidia-driver-535
**• Issue Type:questions:
**• How to reproduce the issue ? :
Install the preempt_rt kernel on Ubuntu 22.04 using sudo apt install linux-realtime, then set it as the boot kernel. After that, install the NVIDIA driver with sudo apt install -y nvidia-driver-535. During this process, do not set IGNORE_PREEMPT_RT_PRESENCE=1.

**• Requirement details
Background:

I am aware that I cannot install the software without setting IGNORE_PREEMPT_RT_PRESENCE=1 when PREEMPT_RT is the boot kernel, as I receive an error message.

I also know that a similar message is written in the source code of open-gpu-kernel-modules. open-gpu-kernel-modules/kernel-open/conftest.sh at bb2dac1f20a06f78e028c4acdc4df1c7908dd91c · NVIDIA/open-gpu-kernel-modules · GitHub

Furthermore, I am aware that the guard process for PREEMPT_RT with IGNORE_PREEMPT_RT_PRESENCE=1 has been in place since around 2015. Linux Audio Production: How to Patch and Compile the Nvidia Graphics Driver (352.41) for the Realtime Kernel

Additionally, I know that NVIDIA has released a Linux-based OS that appears to support PREEMPT_RT for automotive applications. Kernel Configuration | NVIDIA Docs

I am also aware of documentation for applying PREEMPT_RT to Jetson. Kernel Customization — Jetson Linux<br/>Developer Guide 34.1 documentation

What I want to know:

For what reasons does this installation guard with IGNORE_PREEMPT_RT_PRESENCE=1 exist? (For example, I imagine it could be because PREEMPT_RT is not yet stable and NVIDIA cannot fully support it.)

Is there any plan to support PREEMPT_RT on Ubuntu 22.04 for a wide range of IA servers (Intel x86_64 server), not limited to Jetson? If not, is it possible that Jetson alone might be officially supported in advance?

3 Likes