Releases: nvidia-holoscan/meta-tegra-holoscan
Releases · nvidia-holoscan/meta-tegra-holoscan
v2.4.0
v2.3.0
2024-08-05
Added
- Support for Kata Containers (3.7.0)
Changed
- Updated L4T to 36.3.0
- Updated Yocto and all layer dependencies to Scarthgap (5.0.2)
- Updated dGPU drivers to 555.42.02
- Updated CuPy to 13.2.0
- Updated FastRLock to 0.8.2
- Updated Holoscan SDK and Holohub apps to 2.3.0
- Restored IGX Orin Devkit machine configuration, which used to be provided
by meta-tegra layer. - Removed use of
PREFERRED_RPROVIDER
to install dGPU drivers, instead fixing
MACHINE_EXTRA_RDEPENDS
to remove the Tegra iGPU display drivers then install
the dGPU drivers instead.
v2.2.0
v2.1.0
v2.0.0
v1.0.0
2024-02-09
Added
- A
core-image-holoscan
reference image, based oncore-image-sato
with some
Holoscan-specific modifications. - Matchbox desktop customizations for Holoscan SDK and Holohub, including a
Holoscan application group and desktop icons for the sample applications.
Changed
- Updated L4T to 36.2.0
- Updated layer for Yocto Nanbield (4.3.2)
- Replaced
TEGRA_DGPU
build variable withdgpu
machine override. - Removed support for Clara AGX and Holoscan devkits.
- Common iGPU and dGPU components now use the same major/minor/patch versions.
- Most component versions were updated; see component version table below.
Component Versions
Component | Version |
---|---|
L4T | 36.2.0 |
Kernel | 5.15.122 |
dGPU Drivers | 535.154.05 |
CUDA | 12.2.2 |
cuDNN | 8.9.4.25 |
CuPy | 12.3.0 |
GXF | 3.1-24.01 |
Holoscan SDK | 1.0.3 |
MLNX OFED | 23.10-1.1.9.0 |
Nsight Systems | 2023.3.3.42 |
ONNX Runtime | 1.16.3 |
PyTorch | 2.1.2 |
Rivermax | 1.40.11 |
TensorRT | 8.6.2 |
Torchvision | 0.16.2 |
UCX | 1.15.0 |
AJA NTV2 | 17.1 |
v0.6.0
2023-08-01
Added
- Various dependency recipes
- Clara Viz 0.3.1
- PyTorch 2.0.1
- torchvision 0.15.2
- libnvjpeg 12.2.0.2
- nlohmann-json 3.9.1
- Jansson 2.14
- cloudpickle 2.2.1
nvidia_deb_pkgfeed
class to simplify fetching NVIDIA packages from the
SBSA repository.
Changed
- Updated dGPU components
- Drivers 530.30.02
- CUDA 12.1.1
- cuDNN 8.9.0.131
- TensorRT 8.6.0.12
- Nsight Systems 2023.1.2
- Other updated components
- MLNX OFED 5.9-0.5.6.0
- Rivermax 1.21.10
- mstflint 4.24.0-1
- NVIDIA GPU Stress Test 2.4.0
- UCX 1.14.1
- CuPy 12.1.0
- GXF 23.05
- Use public download locations for the following NVIDIA components to avoid the
need to manually download them outside of the bitbake build process- cuDNN
- TensorRT
- GXF
- Nsight Systems
- Replaces deprecated
nv_peer_mem
module withnvidia-peermem
module built
from thenvidia-open-gpu-kernel-modules
source.
Fixed
- Added patch to
nvidia-display-driver
to fixnvidia-p2p
symbol conflicts
withnvidia.ko
, which fixes RDMA for the iGPU. - Added missing fan control settings for the IGX Orin Devkit.
- Fixed the iGPU build of the AJA drivers with RDMA enabled.
- Added kernel patch to fix the color format for the Lontium HDMI input board
supplied with the IGX Orin Devkit. - Added kernel parameter to disable ACS for the CX7 device in the IGX Orin
Devkit since RDMA does not work with ACS enabled.
Known Issues
- The
h264
Holohub applications do not work with the dGPU configuration. - Emergent cameras and the
high_speed_endoscopy
Holohub application
do not work with the dGPU configuration.
Component Versions
iGPU and dGPU | |
---|---|
L4T | 35.3.1 |
Holoscan SDK | 0.6.0 |
MLNX OFED | 5.9-0.5.6.0 |
Rivermax | 1.21.10 |
AJA NTV2 | 16.2.0 |
iGPU | dGPU | |
---|---|---|
GPU Drivers | 35.3.1 | 530.30.02 |
CUDA | 11.4.19 | 12.1.1 |
cuDNN | 8.6.0.166 | 8.9.0.131 |
TensorRT | 8.5.2.1 | 8.6.0.12 |
Nsight Systems | 2022.5.2 | 2023.1.2 |
Emergent Camera | 2.37.05 | N/A |
v0.5.1
2023-04-25
Added
- Secure Boot documentation section
- Patch to append the Microsoft UEFI certificate and UEFI revocation
list to the Secure Boot key enrollment list
Changed
- Updated L4T to 35.3.1
- Updated IGX Orin Devkit configuration for latest board revision
- Switched mlnx-ofed kernel patch to config fragment
Known Issues
- Enabling Secure Boot when using a dGPU causes the UEFI boot menu
to no longer be displayed on the dGPU video output. - Fan control configuration for IGX Orin Devkit is missing.
- Fixed in the v0.5.1-fixes branch by this change.
Component Versions
iGPU and dGPU | |
---|---|
L4T | 35.3.1 |
Holoscan SDK | 0.5.0 |
MLNX OFED | 5.8-1.0.1.1 |
Rivermax | 1.20.10 |
AJA NTV2 | 16.2.0 |
Emergent Camera | 2.37.05 |
iGPU | dGPU | |
---|---|---|
GPU Drivers | 35.3.1 | 525.85.05 |
CUDA | 11.4.19 | 11.8.0 |
cuDNN | 8.6.0.166 | 8.7.0.84 |
TensorRT | 8.5.2.1 | 8.5.3.1 |
nvidia-peermem | N/A | 1.3.0 |
Nsight Systems | 2022.5.2 | 2023.1.1 |
v0.5.0
2023-03-31
Added
- Support for IGX Orin Developer Kit (
igx-orin-devkit
) - HoloHub Applications
- Holoscan Test Suite
- Holoscan UEFI boot image
- dGPU container support
- NVIDIA Container Toolkit 1.9.0
- libnvidia-container 1.11.0
- Nsight Systems 2023.1.1
- Documentation and support for debugging using GDB
PREEMPT_RT
patch support (iGPU only)- Rivermax generic-sender and generic-receiver samples
- Various dependency recipes
- Emergent Camera 2.37.05
- FastRLock 0.8.1
- CuPy 11.5.0
- UCX 1.13.1
- NVIDIA GPU Stress Test 2.3.0
Changed
- Switched to using Yocto/Poky (Kirkstone 4.0.5)
- Updated documentation to use
core-image-sato
image by default - Added
kernel-modules
to build template for convenience - Set default power mode to MAXN
Fixed
- Added missing dGPU driver library symlinks
- Fixed nvidia-settings application
- Added missing Rivermax library symlinks
- Added runtime check to flash script for required dtc command
- Added missing Mellanox firmware updater tool
- Disabled MetaModes in xorg config to fix Vulkan exclusive display mode
Component Versions
iGPU and dGPU | |
---|---|
L4T | 35.2.1 |
Holoscan SDK | 0.5.0 |
MLNX OFED | 5.8-1.0.1.1 |
Rivermax | 1.20.10 |
AJA NTV2 | 16.2.0 |
Emergent Camera | 2.37.05 |
iGPU | dGPU | |
---|---|---|
GPU Drivers | 35.2.1 | 525.85.05 |
CUDA | 11.4.19 | 11.8.0 |
cuDNN | 8.6.0.166 | 8.7.0.84 |
TensorRT | 8.5.2.1 | 8.5.3.1 |
nvidia-peermem | N/A | 1.3.0 |
Nsight Systems | 2022.5.2 | 2023.1.1 |
v0.4.0
2022-12-20
Added
- ONNX Runtime 1.12.1 recipe
- Missing mlnx-ofed components (libibmad5, libibumad3, libibnetdisc5, etc.)
- mstflint 4.22.0-1 recipe
Changed
- Renamed layer to meta-tegra-holoscan
- Updated Holsocan SDK to 0.4.0 using new GitHub repository
(https://github.com/nvidia-holoscan/holoscan-sdk) - Updated GXF to 2.5.0
- glad is now built as a shared library.
- Updated mlnx-ofed to 5.8-1.0.1.1
- Updated Rivermax to 1.20.10
Fixed
- Added host network option to docker run commands to work with VPN.
- Added missing Vulkan ICD to GPU drivers.
- The AJA kernel module will now load automatically if an AJA device is
found, eliminating the need to add it toKERNEL_MODULE_AUTOLOAD
.
Known Issues
- A build-time race condition may lead to Rivermax packaging conflicts.
See the "Building and Flashing" section of README.md for a solution. - The display may not be detected during first boot.
See the "Building and Flashing" section of README.md for a solution.
Component Versions
iGPU and dGPU | |
---|---|
L4T | 35.1.0 |
Holoscan SDK | 0.4.0 |
GXF | 2.5.0 |
AJA NTV2 | 16.2.0 |
MLNX OFED | 5.8-1.0.1.1 |
Rivermax | 1.20.10 |
iGPU | dGPU | |
---|---|---|
GPU Drivers | 35.1.0 | 515.65.01 |
CUDA | 11.4.239 | 11.7.1 |
cuDNN | 8.4.1.50 | 8.5.0.96 |
TensorRT | 8.4.1 | 8.4.3.1 |
nvidia-peermem | N/A | 1.3.0 |