This covers the installation of updated Kernel, out-of-tree nvidia kernel modules package, how to get GNOME desktop running and installation/run of glmark2 benchmark. Also it describes how to get some CUDA and TensorRT samples running.

SP6

Download SLE-15-SP6 (Arm) installation image. This you can put on a regular USB stick or on an SD card using dd command.

Boot from the USB stick/SD card, that you wrote above and install SP6. You need to install via serial console, since the monitor won’t get any signal without the out-of-tree nvidia kernel modules, which are installed later in the process.

Make sure you select the following modules during installation:

  • Basesystem (enough for just installing the kernel driver)
  • Containers (needed for podman for CUDA libraries)
  • Desktop Applications (needed for running a desktop)
  • Development Tools (needed for git for CUDA samples)

Select SLES with GNOME for installation.

Kernel + KMP drivers

Continue installation with serial console.

Now update kernel and install our KMP (kernel module package) for all nvidia kernel modules.

We plan to make the KMP available as a driver kit via the SolidDriver Program. For now please install an updated kernel and the KMP after checking the build status (rebuilding can take a few hours!) from our open buildservice:

sudo zypper ar https://download.opensuse.org/repositories/home:/sndirsch:/sidecar/SLE_15_SP6/ home:sndirsch:sidecar
sudo zypper ref
# flavor either default or 64kb (check with `uname -r` command)
sudo zypper up kernel-<flavor>
sudo zypper in -r home:sndirsch:sidecar nvidia-open-driver-G06-signed-sidecar-kmp-<flavor>

Userspace/Desktop

Unfortunately installing the userspace is a non-trivial task.

Installation

Download Jetpack 6 Driver Package (BSP) from this location. Extract jetson_linux_r36.3.0_aarch64.tbz2.

tar xf jetson_linux_r36.3.0_aarch64.tbz2

Then you need to convert debian packages from this content into tarballs.

pushd Linux_for_Tegra
sed -i 's/lbzip2/bzip2/g' nv_tools/scripts/nv_repackager.sh
./nv_tools/scripts/nv_repackager.sh -o ./nv_tegra/l4t_tar_packages --convert-all
popd

From the generated tarballs you only need these:

nvidia-l4t-3d-core_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-camera_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-core_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-cuda_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-firmware_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-gbm_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-multimedia-utils_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-multimedia_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-nvfancontrol_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-nvml_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-nvpmodel_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-nvsci_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-pva_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-tools_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-vulkan-sc-sdk_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-wayland_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-x11_36.3.0-20240404104251_arm64.tbz2
nvidia-l4t-nvml_36.3.0-20240404104251_arm64.tbz2

And from this tarball nvidia-l4t-init_36.3.0-20240404104251_arm64.tbz2 you only need these files:

etc/asound.conf.tegra-ape
etc/asound.conf.tegra-hda-jetson-agx
etc/asound.conf.tegra-hda-jetson-xnx
etc/nvidia-container-runtime/host-files-for-container.d/devices.csv
etc/nvidia-container-runtime/host-files-for-container.d/drivers.csv
etc/nvsciipc.cfg
etc/sysctl.d/60-nvsciipc.conf
etc/systemd/nv_nvsciipc_init.sh
etc/systemd/nvpower.sh
etc/systemd/nv.sh
etc/systemd/system.conf.d/watchdog.conf
etc/systemd/system/multi-user.target.wants/nv_nvsciipc_init.service
etc/systemd/system/multi-user.target.wants/nvpower.service
etc/systemd/system/multi-user.target.wants/nv.service
etc/systemd/system/nv_nvsciipc_init.service
etc/systemd/system/nvpower.service
etc/systemd/system/nv.service
etc/udev/rules.d/99-tegra-devices.rules
usr/share/alsa/cards/tegra-ape.conf
usr/share/alsa/cards/tegra-hda.conf
usr/share/alsa/init/postinit/00-tegra.conf
usr/share/alsa/init/postinit/01-tegra-rt565x.conf
usr/share/alsa/init/postinit/02-tegra-rt5640.conf

So first let’s repackage nvidia-l4t-init_36.3.0-20240404104251_arm64.tbz2:

pushd Linux_for_Tegra/nv_tegra/l4t_tar_packages/
cat > nvidia-l4t-init.txt << EOF
etc/asound.conf.tegra-ape
etc/asound.conf.tegra-hda-jetson-agx
etc/asound.conf.tegra-hda-jetson-xnx
etc/nvidia-container-runtime/host-files-for-container.d/devices.csv
etc/nvidia-container-runtime/host-files-for-container.d/drivers.csv
etc/nvsciipc.cfg
etc/sysctl.d/60-nvsciipc.conf
etc/systemd/nv_nvsciipc_init.sh
etc/systemd/nvpower.sh
etc/systemd/nv.sh
etc/systemd/system.conf.d/watchdog.conf
etc/systemd/system/multi-user.target.wants/nv_nvsciipc_init.service
etc/systemd/system/multi-user.target.wants/nvpower.service
etc/systemd/system/multi-user.target.wants/nv.service
etc/systemd/system/nv_nvsciipc_init.service
etc/systemd/system/nvpower.service
etc/systemd/system/nv.service
etc/udev/rules.d/99-tegra-devices.rules
usr/share/alsa/cards/tegra-ape.conf
usr/share/alsa/cards/tegra-hda.conf
usr/share/alsa/init/postinit/00-tegra.conf
usr/share/alsa/init/postinit/01-tegra-rt565x.conf
usr/share/alsa/init/postinit/02-tegra-rt5640.conf
EOF
tar xf nvidia-l4t-init_36.3.0-20240404104251_arm64.tbz2
rm nvidia-l4t-init_36.3.0-20240404104251_arm64.tbz2
tar cjf nvidia-l4t-init_36.3.0-20240404104251_arm64.tbz2 $(cat nvidia-l4t-init.txt)
popd

Then extract the generated tarballs to your system.

pushd Linux_for_Tegra/nv_tegra/l4t_tar_packages
for i in \
nvidia-l4t-core_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-3d-core_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-cuda_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-firmware_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-gbm_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-multimedia-utils_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-multimedia_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-nvfancontrol_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-nvpmodel_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-tools_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-x11_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-nvsci_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-pva_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-wayland_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-camera_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-vulkan-sc-sdk_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-nvml_36.3.0-20240404104251_arm64.tbz2 \
nvidia-l4t-init_36.3.0-20240404104251_arm64.tbz2; do
  sudo tar xjf $i -C /
done
popd

Then you still need to move

/usr/lib/xorg/modules/drivers/nvidia_drv.so
/usr/lib/xorg/modules/extensions/libglxserver_nvidia.so

to

/usr/lib64/xorg/modules/drivers/nvidia_drv.so
/usr/lib64/xorg/modules/extensions/libglxserver_nvidia.so

and add /usr/lib/aarch64-linux-gnu to /etc/ld.so.conf.d/nvidia-tegra.conf.

sudo mv /usr/lib/xorg/modules/drivers/nvidia_drv.so \
          /usr/lib64/xorg/modules/drivers/
sudo mv /usr/lib/xorg/modules/extensions/libglxserver_nvidia.so \
          /usr/lib64/xorg/modules/extensions/
sudo rm -rf /usr/lib/xorg
sudo echo /usr/lib/aarch64-linux-gnu >> /etc/ld.so.conf.d/nvidia-tegra.conf

Run ldconfig 

sudo ldconfig

Video group for regular users

A regular user needs to be added to the group video to be able to log in to the GNOME desktop as regular user. This can be achieved by using YaST, usermod or editing /etc/group manually.

Reboot the machine with the previously updated kernel

sudo reboot

In Mokmanager (Perform MOK management) select Continue boot. Although Secureboot is enabled by default in BIOS it seems it hasn’t been implemented yet (BIOS from 04/04/2024). Select first entry SLES 15-SP6 for booting.

Basic testing

First basic testing will be running nvidia-smi

sudo nvidia-smi

Graphical desktop (GNOME) should work as well. Unfortunately Linux console is not available. Use either a serial console or a ssh connection if you don’t want to use the graphical desktop or need remote access to the system.

glmark2

Install phoronix-test-suite

sudo zypper ar https://cdn.opensuse.org/distribution/leap/15.6/repo/oss/ repo-oss
sudo zypper ref
sudo zypper in phoronix-test-suite

Run phoronix-test-suite

sudo zypper in gcc gcc-c++
phoronix-test-suite benchmark glmark2

CUDA/Tensorflow

Containers

NVIDIA provides containers available for Jetson that include SDKs such as CUDA. More details here. These containers are Ubuntu based, but can be used from SLE as well. You need to install the NVIDIA container runtime for this. Detailed information here.

1. Install podman and nvidia-container-runtime
sudo zypper install podman
sudo zypper ar https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo
sudo zypper modifyrepo --enable nvidia-container-toolkit-experimental
sudo zypper --gpg-auto-import-keys install -y nvidia-container-toolkit
sudo nvidia-ctk cdi generate --mode=csv --output=/var/run/cdi/nvidia.yaml
sudo nvidia-ctk cdi list
2. Download the CUDA samples
sudo zypper install git
cd
git clone https://github.com/NVIDIA/cuda-samples.git
cd cuda-samples
git checkout v12.4
3. Start X
sudo rcxdm stop
sudo Xorg -retro &> /tmp/log &
export DISPLAY=:0
xterm &

Monitor should now show a Moiree pattern with an unframed xterm on it. Otherwise check /tmp/log.

4. Download and run the JetPack6 container
sudo podman run --rm -it -e DISPLAY --net=host --device nvidia.com/gpu=all --group-add keep-groups --security-opt label=disable -v $HOME/cuda-samples:/cuda-samples nvcr.io/nvidia/l4t-jetpack:r36.2.0 /bin/bash

CUDA

5. Build and run the samples in the container
cd /cuda-samples
make -j$(nproc)
pushd ./Samples/5_Domain_Specific/nbody
make
popd
./bin/aarch64/linux/release/deviceQuery
./bin/aarch64/linux/release/nbody

Tensorrt

6. Build and run Tensorrt in the container

This is both with the GPU and DLA (deep-learning accelerator).

cd /usr/src/tensorrt/samples/
make -j$(nproc)
cd ..
./bin/sample_algorithm_selector
./bin/sample_onnx_mnist
./bin/sample_onnx_mnist --useDLACore=0
./bin/sample_onnx_mnist --useDLACore=1

Misc

Performance

You can improve the performance by giving the clock a boost. For best performance you can run jetson_clocks to set the device to max clock settings

sudo jetson_clocks --show
sudo jetson_clocks
sudo jetson_clocks --show

The 1st and 3rd command just prints the clock settings.

MaxN Power

For maximum performance you also need to set MaxN Power. This can be done by running

sudo nvpmodel -m 0

Afterwards you need to reboot the system though.

sudo reboot

In order to check for the current value run

sudo nvpmodel -q