Hi all,
I’m trying to run the YOLO model deployment pipeline using the Holoscan SDK (v3.1.0) through Docker on WSL2 (Ubuntu 22.04) with an RTX 3080. Everything runs successfully, and inference is completed with very low latency, but Holoviz fails to launch due to a Vulkan-related error, preventing any video output.
Here’s the specific error:
[error] [gxf_wrapper.cpp:63] Exception occurred when starting operator: 'detection_visualizer' - Failed to create the Vulkan instance.
[error] vkCreateInstance failed with ERROR_INCOMPATIBLE_DRIVER
I’ve installed the latest NVIDIA driver for Windows (v576.02), Docker is correctly configured with nvidia-container-toolkit
, and nvidia-smi
works both inside and outside the container.
Environment:
- Windows 11
- WSL2 (Ubuntu 22.04)
- Docker v28.0.4
- NVIDIA driver v576.02 (CUDA 12.9)
- GPU: RTX 3080
- Holoscan SDK v3.1.0
- Running Holohub’s
yolo_model_deployment
sample
I tried setting $DISPLAY
, confirmed /tmp/.X11-unix/X0
exists, and even installed vulkan-tools
inside the container, but vulkaninfo
fails with the same vkCreateInstance
error.
My goal is to:
- Get Holoviz working inside the container (preferably with Vulkan), or
- Understand best practices for visualizing output in real time via WSL2 + Docker setup
Any insights, suggestions, or similar experience would be greatly appreciated!
Thanks in advance!