IGX-SW 2.0 Production Release#
IGX-SW 2.0 Production Release is the production quality, enterprise level IGX-SW stack.
Release Notes#
This release is recommended for production purposes. Please ensure that firmware and software of all components, including BSP, QSPI (Boot Firmware), BMC, SMCU, CX7, IGX OS etc., are updated to the versions associated with this release before deploying production systems.
This release supports both the IGX Thor Developer Kit and the Thor Developer Kit Mini. You can find the version of your IGX Thor Developer Kit on the sticker on the developer kit, or laser etched on the bottom of the device.
Use these release notes to find information about the following:
Current Software Versions#
The following table shows the current version information of the software packages provided in the NVIDIA and Ubuntu repositories for the NVIDIA IGX Software Stack:
Component |
Version |
Additional Information |
|---|---|---|
CUDA Toolkit and GPU Driver |
CUDA Toolkit:
13.0.0GPU Driver:
580.00 |
|
CUDA |
13.0.0 |
|
TRT |
10.13.3.9-1 |
|
cuDNN |
9.12.0.46-1 |
|
IGX OS |
2.0-rc8 |
|
Boot Firmware (QSPI) |
38.5.0-gcid-44447062 (Built on 03/10/2026) |
Refer to the Quick Start Guide to check firmware versions |
BMC Firmware |
5.4 |
Refer to the Quick Start Guide to check firmware versions |
MCU Firmware |
2.00.02 |
Refer to the Quick Start Guide to check firmware versions |
ISO |
IGX-r38v2.0-rc8.iso |
Component |
Version |
|---|---|
Ubuntu |
|
BSP |
|
Kernel |
|
GPU Driver |
|
DOCA-OFED |
|
ISO |
|
ISO MD5 Checksum |
|
New Features#
IGX SW:
Support for IGX T5000 module, T7000 board kit, IGX Thor Developer Kit and and IGX Thor Developer Kit Mini
OpenRM-based stack architecture
Ubuntu 24.04 based IGX OS
Canonical signed LTS Kernel 6.8.0-1019.19
CUDA 13 based compute stack
NVIDIA optimized preemptable realtime kernel
Known Issues#
This section provides details about issues that were discovered during development and QA but have not yet been resolved in this release.
Boot / System Stability#
The following boot and system stability related issues are noted in this release.
Bug |
Description |
|---|---|
854848 |
DRAM scrubbing in MB1/MB2 and SBE error detection and correction in the DRAM hardware are enabled by default in the current release. The page retirement feature (DBE handling) in FSI is disabled by default (does not work) in the current release and will be enabled in future releases. |
5890331 |
On IGX T7000 Developer Kit with dGPU, the integrated GPU (iGPU) may fail to enumerate after extended GPU stress workloads (12+ hours). nvidia-smi reports “Unable to determine the device handle for GPU0.” A system reboot is required to recover iGPU functionality. |
5894351 |
Suspend and resume (SC7) functionality is not supported on the IGX T7000 Developer Kit in this release. If the suspend and resume operation is tried, it may result in issues like the WiFi module not getting enumerated. Hard reset the developer kit (by long press power key and release) to fix this issue. |
Display and Graphics#
The following display and graphics related issues are noted in this release.
Bug |
Description |
|---|---|
5905436 |
X11 is the primary supported window system for this release. Wayland and EGLDevice stability are not guaranteed. Users should use X11 as the display server for production workloads. |
5885671 |
VPI (Vision Programming Interface) applications running inside Docker containers on IGX T7000 Developer Kit with dGPU may fail with “Failed to get EGL display” errors when targeting the dGPU. This affects VPI workloads that require EGL backend. Workaround: Run VPI on the iGPU, or run VPI natively outside Docker containers. |
5957410 |
The nvstart-weston.sh script may fail to launch Weston on IGX T7000 Developer Kit with dGPU configuration. The script selects the incorrect DRI card device (/dev/dri/card3 for dGPU instead of iGPU), resulting in “EGL_NOT_INITIALIZED” error. Workaround: Manually specify the correct DRI card for iGPU when launching Weston, or use X11 (the primary supported display server for this release). |
5929911 |
The libvdpau_nvidia.so library is not included in this release. Applications using VDPAU (e.g., mplayer -vo vdpau) will fail. Workaround: Use software decoding, GStreamer with nvdec, or alternative video output methods. |
5909111 |
Hot-switching the display between iGPU and dGPU is not supported in this release. A system reboot is required after switching the display. |
5946078 |
OpenGL onscreen rendering on the iGPU is V-Sync locked at approximately 60 FPS (matching display refresh rate). Vulkan onscreen rendering is not affected and runs at full performance (330+ FPS). Use offscreen rendering or Vulkan for performance-sensitive graphics workloads. |
5915967 |
Running nvidia-settings on IGX T7000 Developer Kit with Blackwell dGPU displays multiple error messages including “Failed to open VDPAU backend” and “The requested operation is not available on target device.” The nvidia-settings utility has limited functionality for dGPU management in this release. |
5984016 |
OpenGL/GLX applications running on the iGPU may experience hangs or crashes during extended graphics stress testing or when running under high CPU load. This affects stress-test scenarios such as the glxstress (hang after ~10 minutes) and the bubble under concurrent CPU stress. Normal interactive graphics workloads are unaffected. A fix is targeted for a future release. |
5984767 |
Display and graphics on the IGX Thor Developer Kit have limited functionality in this release. For example, two displays in separate-x-screens mode do not work. |
Connectivity#
The following connectivity related issues are noted in this release.
Bug |
Description |
|---|---|
5957420 |
The QSFP port on IGX T5000 Developer Kit currently operates at 10 Gbps. 25 Gbps Ethernet speed is not enabled by default. Users requiring 25G connectivity should refer to the Jetson developer guide for manual DTB configuration steps. The actual throughput in 25G mode is lower than 25G due to an issue. |
5881469 |
When a WiFi access point is renamed on the router side, the old AP name may persist in the WiFi scanning page on IGX and cannot be removed through the UI. Workaround: Delete the saved WiFi connection profile and rescan. |
5855268 |
During extended suspend/resume (SC7) stress testing on IGX T5000 Developer Kit Mini, network connectivity through the USB bridge interface (l4tbr0) may fail to recover after approximately 800 cycles. The device resumes successfully but SSH/network is unreachable. A full power cycle is required to restore connectivity. |
5824091 |
An intermittent “Connection failed: Activation of network connection failed” notification may appear on the desktop after rebooting IGX T7000 Developer Kit. The network connection typically recovers automatically within a few seconds. This is a cosmetic notification issue and does not indicate persistent connectivity failure. |
5701579 |
After flashing the IGX device, the first attempt to connect to a 6GHz WiFi network with WPA3 security may fail. Reboot the device after initial flash before attempting 6GHz WPA3 WiFi connections. |
5962849 |
On IGX T5000 Developer Kit Mini, TCP bidirectional throughput on the Realtek RTL8126 Ethernet (enP2p1s0) may drop to zero when the link speed is set to 2500 Mbps with MTU 9000 (jumbo frames). Workaround: Use default MTU 1500, or use 1000/5000 Mbps speed settings with jumbo frames. |
5870567 |
The Realtek RTL8126 2.5G Ethernet interface (enP2p1s0) on IGX T5000 Developer Kit Mini may experience elevated packet drop rates under high broadcast/multicast traffic loads. This primarily affects UDP applications, PTP, and real-time streaming. Workaround: Use the ConnectX-7 network interface for latency-sensitive or loss-sensitive workloads. |
5983628 |
Some audio glitches are observed when Bluetooth headphones are paired and videos are played on YouTube. |
5984016 |
Continuous messages “rtk_btcoex: count_a2dp_packet_timeout: a2dp_packet_count 60” are logged in the dmesg when Bluetooth headphones are paired. There is no impact on functionality. |
GPU and Compute#
The following GPU and compute related issues are noted in this release.
Bug |
Description |
|---|---|
5943778 |
MIG (for both iGPU and dGPU) is not supported in this release. Enabling MIG (Multi-Instance GPU) mode on the dGPU using |
5711804 |
Running cuDNN test suites on IGX T5000 Developer Kit Mini may trigger kernel OOM (Out of Memory) conditions under heavy GPU memory allocation. Users should monitor memory usage and limit concurrent GPU workloads to prevent OOM-triggered process kills. |
5929862 |
Running VPI test suites concurrently (e.g., vpi_test_system_core and vpi_test_unit_target in parallel) on IGX Thor iGPU may cause a kernel-level RT mutex deadlock after approximately 130 iterations, requiring a system reboot. Workaround: Avoid running multiple VPI stress workloads simultaneously on the iGPU. |
5699079 |
If excessively large CUDA memory allocation is attempted, the device might reboot. Ensure that CUDA or other applications do not request more memory than is physically available. Another workaround is to launch CUDA processes with a higher OOM score so that, under memory pressure, the kernel preferentially kills those processes rather than critical system processes. |
Security and Hardening#
The following security and hardening related issues are noted in this release.
Bug |
Description |
|---|---|
5915767 |
IGX developer kit images do not enable a host firewall by default. All listening services (SSH, rpcbind, etc.) are reachable on all interfaces.
For lab environments behind perimeter firewalls, this is expected. For untrusted networks, enable the ufw firewall with the following command: |
5915755 |
The IGX base image may contain packages with pending security updates. Users should run |
SDK#
The following SDK related issues are noted in this release.
Bug |
Description |
|---|---|
5956594 |
PVA SDK 2.9.1 L4T cross-compilation fails due to incorrect ASIP Programmer path in CMake files. The default path references /opt/nvidia/pva-sdk/ but the actual location is /opt/nvidia/cupva/.
Workaround: Set ASIP paths manually with the following: |
Power and Performance#
The following power and performance related issues are noted in this release.
Bug |
Description |
|---|---|
5953312 |
Running |
5957538 |
During boot up on IGX Thor Developer Kit Mini, the kernel reports “irq 345: nobody cared” and disables IRQ #345 (regmap_irq_thread associated with the NVVRS PSEQ power sequencer). There is no functional impact; the NVVRS PSEQ driver and power sequencing function correctly despite this message. |
Camera#
The following camera related issues are noted in this release.
Bug |
Description |
|---|---|
5853942 |
The IGX BaseOS ISO image ships with DualIMX274 camera sensor enabled by default. To use other supported CSI sensors (IMX185, IMX390, Hawk, Owl, etc.), users must manually configure the device tree overlay files. Refer to the camera section of the developer guide for DTB overlay configuration steps. |
5700598 |
Errors BAD_INPUT_STREAM and ISP_PROCESSING_FAILURE_RECOVERABLE are observed with the nvsiple_coe_camera application on the IGX Thor Developer Kit Mini and VB1940 HSB camera. |
BMC#
The following BMC related issues are noted in this release.
Bug |
Description |
|---|---|
5498176 |
BMC KVM is not supported in this release. |
5893651 |
The BMC tcu_muxer and obmc-console services may sometimes fail to start after a BMC firmware update, factory reset, or reboot, resulting in loss of Tegra serial console access through BMC. Workaround: Restart the tcu_muxer service manually using the following command from the BMC shell: |
4811908 |
Some information such as the asset tag is not updated on the System Inventory page in the BMC WebUI. |
5912027 |
Loading an image using the BMC Virtual Media from an external server is very slow as compared to IGX Orin. |
5915761 |
Tegra/SMCU console hangs after SMCU flash using the smcu-auto-flash.sh. Do not use the smcu-auto-flash.sh in this BMC FW release. |
5911508 |
Sometimes the BMC WebUI shows “BMC is still initializing. Use at your own risk.” Workaround: Power cycle the unit. |