CUDA error 801 with RT kernel (nvbuild.sh -r) on JetPack 6.2.2 — expected behavior?

Robotics & Edge Computing Jetson Systems Jetson Orin Nano
, ,
1

Hi,

We are building a PREEMPT_RT kernel on Jetson Orin Nano using the official build script and have a question about CUDA support on the resulting RT kernel.

Environment:

  • Hardware: Jetson Orin Nano
  • JetPack: 6.2.2
  • RT kernel built with: ./nvbuild.sh -r -m (resulting in kernel 5.15.185-rt-tegra)
  • CUDA Runtime: 12.6

Observed behavior:

After booting the RT kernel and running the CUDA deviceQuery sample, it fails with error 801:

$ ./deviceQuery
./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 801
→ unknown error
Result = FAIL

The same binary runs successfully on the standard (non-RT) kernel included with JetPack 6.2.2.

Question:

Is CUDA expected to work on the PREEMPT_RT kernel when built using nvbuild.sh -r? Is this a known limitation, or is there official guidance on running CUDA workloads alongside a
real-time kernel on Jetson Orin?

Thank you.

4

Hi,

Please run the command below and share the output with us.

$ sudo nvidia-smi

More, could you share the detailed steps you used for setup so we can check if there are any issues?

Thanks.

5

Thank you for the quick reply.

I first tried running nvidia-smi, and the output was:

$ sudo nvidia-smi
Unable to determine the device handle for GPU0002:00:00.0: Unknown Error

More, could you share the detailed steps you used for setup so we can check if there are any issues?

I created a bash script to reproduce the issue and ran it on Ubuntu 22.04 running under WSL2 for cross-compilation.

#!/bin/bash

set -euxo pipefail

## define environment variables
SCRIPT_DIR=$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" &> /dev/null && pwd)

export CROSS_COMPILE=/usr/bin/aarch64-linux-gnu-
export INSTALL_MOD_PATH="${SCRIPT_DIR}/Linux_for_Tegra/rootfs"
export IGNORE_PREEMPT_RT_PRESENCE=1
export KERNEL_HEADERS="${SCRIPT_DIR}/Linux_for_Tegra/source/kernel/kernel-jammy-src"

## download necessary tarballs
wget https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v5.0/sources/public_sources.tbz2
wget https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v5.0/release/Jetson_Linux_r36.5.0_aarch64.tbz2
wget https://developer.nvidia.com/downloads/embedded/l4t/r36_release_v5.0/release/Tegra_Linux_Sample-Root-Filesystem_r36.5.0_aarch64.tbz2

## expand tarballs
tar jxvf Jetson_Linux_r36.5.0_aarch64.tbz2
tar jxvf public_sources.tbz2

## expand rootfs
cp Tegra_Linux_Sample-Root-Filesystem_r36.5.0_aarch64.tbz2 Linux_for_Tegra/rootfs/
cd Linux_for_Tegra/rootfs
tar jxvf Tegra_Linux_Sample-Root-Filesystem_r36.5.0_aarch64.tbz2

## move to source directory
cd ../source

tar jxvf kernel_src.tbz2
tar jxvf kernel_oot_modules_src.tbz2
tar jxvf nvidia_kernel_display_driver_source.tbz2

## build and install to rootfs
./nvbuild.sh -r
./nvbuild.sh -r -i

## scp kernel image and modules to jetson
cd ../rootfs

## edit this variable
remote_work_dir_path=

cd boot
scp Image jetson:${remote_work_dir_path}/Image-rt

cd ../lib/modules

sudo tar jcvf 5.15.185-rt-tegra.tbz2 5.15.185-rt-tegra
scp 5.15.185-rt-tegra.tbz2 jetson:${remote_work_dir_path}

Next, I installed the kernel image and the modules on the Jetson Orin Nano using the following script. This script is located in the same directory as the kernel image and modules.

#!/bin/bash

set -euxo pipefail

sudo cp Image-rt /boot/

tar jxvf 5.15.185-rt-tegra.tbz2
sudo cp -r 5.15.185-rt-tegra /lib/modules/

sudo depmod -a

sudo update-initramfs -c -k 5.15.185-rt-tegra
sudo cp /boot/initrd.img-5.15.185-rt-tegra /boot/initrd-rt

My extlinux.conf is as follows. Note that I had already enabled the I2S function using jetson-io.py, so the FDT and OVERLAYS entries were added.

DEFAULT JetsonIORT

LABEL JetsonIORT
        MENU LABEL Custom Header Config: <HDR40 User Custom [2026-06-26-162555]>
        LINUX /boot/Image-rt
        INITRD /boot/initrd-rt
        FDT /boot/dtb/kernel_tegra234-p3768-0000+p3767-0005-nv-super.dtb
        APPEND ${cbootargs} root=PARTUUID=${UUID} rw rootwait rootfstype=ext4 mminit_loglevel=4 console=ttyTCU0,115200 firmware_class.path=/etc/firmware fbcon=map:0 video=efifb:off console=tty0 efi=runtime pci=pcie_bus_perf nvme.use_threaded_interrupts=1 nv-auto-config
        OVERLAYS /boot/jetson-io-hdr40-user-custom.dtbo

After rebooting, I confirmed that the RT kernel was running:

$ uname -a
Linux ubuntu 5.15.185-rt-tegra #1 SMP PREEMPT_RT Thu Jul 2 09:29:30 JST 2026 aarch64 aarch64 aarch64 GNU/Linux
6

Hi,

We don’t maintain nvbuild.sh on JetPack 6.
Could you follow the steps shared in the two comments below?

Thanks.