NVIDIA Reference Index

This document is the annotated bibliography for the jetson-rt-stack build: the canonical NVIDIA, Axelera, Stereolabs, and community pages from which every magic value in the pipeline is derived. Use it to trace any CONFIG_*, patch, flash flag, or tuning value back to its upstream source.

This repo intentionally carries no scraped reference material. A prior revision shipped roughly 16 MB of scraped HTML (docs/nvidia_comprehensive/, docs/nvidia_r36.4/, docs/refs/, docs/external/) plus 8 MB of build and flash log artifacts. All of it was removed: the source pages live online at the canonical URLs below, and the values we actually consume are already captured in our own docs. Every CONFIG_*, patch, and RT tuning value is recorded in KERNEL_OPTIMIZATIONS.md, KERNEL_PATCHES.md, and RT_KERNEL_OPTIMIZATION.md.

L4T R36.4.3 / JetPack 6.2, primary references

These are the canonical pages our build pipeline references:

Page URL
Jetson Linux R36.4.3 download hub (BSP, rootfs, toolchain, public sources) https://developer.nvidia.com/embedded/jetson-linux-r3643
Release notes (PDF) https://docs.nvidia.com/jetson/archives/r36.4.3/ReleaseNotes/Jetson_Linux_Release_Notes_r36.4.3.pdf
Jetson Linux Developer Guide R36.4.3 (top) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/index.html
Quick Start https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/IN/QuickStart.html
Flashing Support (the canonical reference for l4t_initrd_flash.sh flags + XMLs) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/FlashingSupport.html
Kernel Customization (Bring Your Own Kernel) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/Kernel/KernelCustomization.html
Boot Architecture (Orin boot flow, partition layout) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/AR/BootArchitecture.html
Module Adaptation (Orin NX/Nano series, board target names) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/HR/JetsonModuleAdaptationAndBringUp/JetsonOrinNxNanoSeries.html
Clocks (devfreq paths, BPMP debug clk tree, EMC) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/Clocks.html
Backup & Restore (golden image clone) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/FlashingSupport/BackupAndRestore.html
DBT (Device-Tree-Based Tooling) https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/Kernel/DBT.html
Power Management & nvpmodel https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/PlatformPowerAndPerformance/PlatformPowerAndPerformance.html
Jetson Orin NX module datasheet https://developer.nvidia.com/downloads/jetson-orin-nx-module-series-data-sheet
CUDA GPU compute capability table (Orin NX = sm_87) https://developer.nvidia.com/cuda/gpus

What we use each page for

  • Quick Start + Flashing Support: source for the 04_flash_nvme.sh command structure (l4t_initrd_flash.sh --external-device --network usb0 ...).
  • Kernel Customization (BYOK): source for the cross-compile pattern and LOCALVERSION. The Bootlin toolchain URL we ship is taken from this page.
  • Boot Architecture: source for the MB1 -> MB2 -> CBoot -> UEFI -> Linux chain documented in COMMUNITY_POST.md Part 4.
  • Module Adaptation (Orin NX/Nano series): settles the TARGET_BOARD choice. Orin NX 16GB on a P3509-class carrier uses jetson-orin-nano-devkit, an alias of p3509-a02+p3767-0000.conf. The -super board target is the Orin Nano power table only and must not be used here. See VERIFICATION_REPORT.md section 1.1 and TARGET_BOARD in versions.env.
  • Clocks: settles the GPU devfreq path. R36.x exposes the GPU at /sys/class/devfreq/17000000.gpu/, not the R35-era .ga10b. See jetson_rt_tune.sh.
  • Backup & Restore: the canonical workflow our golden-image clone (clone_golden.sh and flash_golden.sh) is built on. See GOLDEN_IMAGE.md.
  • Module datasheet: pin counts, power envelope, JetPack matrix.

Vendor / community references

The on-device consumers of these vendor stacks are documented in DRIVERS.md (ZED X driver and daemons), ZEDX_METIS_CPP.md (C++ samples on the ZED SDK and libaxruntime), and BENCHMARKS.md (measured throughput).

Stereolabs ZED X / ZED Link Mono / ZED SDK:

  • Drivers (.deb downloads, kernel module included): https://www.stereolabs.com/developers/drivers
  • ZED Link install guide: https://www.stereolabs.com/docs/embedded/zed-link/install-the-drivers
  • ZED SDK release / download: https://www.stereolabs.com/developers/release
  • ZED SDK Jetson install (silent flags): https://www.stereolabs.com/docs/development/zed-sdk/jetson
  • Python API (get_python_api.py): https://www.stereolabs.com/docs/development/python/install
  • Python API repo: https://github.com/stereolabs/zed-python-api
  • Stereolabs GitHub org: https://github.com/orgs/stereolabs/repositories

Axelera Voyager SDK / Metis M.2:

  • Voyager SDK GitHub: https://github.com/axelera-ai-hub/voyager-sdk
  • Voyager 1.6 release announcement (pip wheels): https://community.axelera.ai/product-updates/voyager-sdk-new-pipeline-builder-and-more-1313
  • Metis M.2 product page: https://axelera.ai/ai-accelerators/metis-m2-ai-acceleration-card
  • Metis M.2 datasheet (PDF): https://axelera.ai/hubfs/Axelera_February2025/pdfs/axelera-ai-m2-ai-edge-accelerator-module.pdf
  • Community thread that confirmed the Metis PCI vendor ID 1f9d (our earlier 1d60 was wrong). Match on the vendor ID 1f9d: alone: rev-02 boards enumerate as 1f9d:1100, rev-01 boards as 1f9d:11aa. See verify_tuning.sh for the detection logic: https://community.axelera.ai/metis-pcie-7/axelera-metis-pcie-ai-accelerator-not-recognized-by-lspci-145
  • Bring up Metis on Jetson Orin (Axelera Help Center): https://support.axelera.ai (search “Bring up Metis M.2 Jetson Orin”)

ROS 2 Humble / Isaac ROS / Nav2:

  • Isaac ROS getting started: https://nvidia-isaac-ros.github.io/getting_started/index.html
  • isaac_ros_common: https://github.com/NVIDIA-ISAAC-ROS/isaac_ros_common
  • Nav2 (humble branch): https://github.com/ros-navigation/navigation2/tree/humble/nav2_bringup
  • rosbag2 mcap storage: https://docs.ros.org/en/humble/p/rosbag2_storage_mcap/
  • FastDDS discovery (default UDP port range 7400-7500/udp): https://fast-dds.docs.eprosima.com/en/latest/fastdds/discovery/simple.html

Linux kernel (5.15, what L4T R36.4.3 ships):

  • Tagged tree (Kconfig sources for every CONFIG_*): https://github.com/torvalds/linux/tree/v5.15
  • kernel.org cgit (same tree): https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/?h=linux-5.15.y
  • PREEMPT_RT documentation: https://wiki.linuxfoundation.org/realtime/start

How to re-fetch these pages offline

For offline copies (for example, shipping a flash station to a network-restricted lab), run wget or curl against the URLs above. The old scripts/scrape_nvidia_*.py crawlers did this in bulk and have been removed for two reasons: the bulk output goes stale the moment NVIDIA updates a page, and each scrape produced 506+ files for the handful of pages we actually use.

A focused alternative (~10 lines):

mkdir -p offline_refs
for url in \
    "https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/IN/QuickStart.html" \
    "https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/FlashingSupport.html" \
    "https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/Kernel/KernelCustomization.html" \
    "https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/SD/FlashingSupport/BackupAndRestore.html" \
    "https://docs.nvidia.com/jetson/archives/r36.4.3/DeveloperGuide/HR/JetsonModuleAdaptationAndBringUp/JetsonOrinNxNanoSeries.html" \
; do
    wget -q -P offline_refs --convert-links --adjust-extension "$url"
done

Where to find the canonical defconfig

The build mutates the L4T-stock defconfig in place at latest_jetson/Linux_for_Tegra/source/kernel/kernel-jammy-src/arch/arm64/configs/defconfig, injected by 01_extract_and_patch.sh. A prior revision also carried separate copies of defconfig.txt and tegra_defconfig.txt under docs/refs/. Those are removed: they were stale snapshots of the same file that ships in the L4T tarball.

To inspect the stock baseline before our patches:

make extract   # extracts the L4T tarball into latest_jetson/
less latest_jetson/Linux_for_Tegra/source/kernel/kernel-jammy-src/arch/arm64/configs/defconfig

Phase 1 does not currently keep a .orig copy. To diff the stock baseline against the patched config, add cp defconfig defconfig.orig to the script before its cat >> append, re-run make extract, then:

DEFCONFIG_DIR=latest_jetson/Linux_for_Tegra/source/kernel/kernel-jammy-src/arch/arm64/configs
diff -u "$DEFCONFIG_DIR/defconfig.orig" "$DEFCONFIG_DIR/defconfig"