Hi,
I’m integrating the YOLO model deployment app into a Windows system where real-time GPU display is required. For this, I’m using Holoscan within WSL2 + CUDA (GPU passthrough works well), but I’m hitting a hard limitation due to Holoviz relying on Vulkan. Because of the real-time performance requirement, I cannot use packages like OpenCV.
Currently:
vulkaninfo
under WSL2 showsMicrosoft Direct3D12 (NVIDIA RTX 3080)
via D3D12/DZN (testing only).- Vulkan-capable NVIDIA GPU isn’t detected in Holoviz due to missing extensions like
VK_KHR_push_descriptor
.
- Is there a way to disable Vulkan fallback in Holoviz or patch it to work with D3D12 devices?
- If I want to build one myself that reads CUDA buffers and displays them directly, how complex would it be?
- Can I use or contribute to the unreleased WebRTC/WebSocket operators mention in ( HoloViz output to web-based app via RTSP - Healthcare & Life Sciences / Holoscan SDK - NVIDIA Developer Forums)?
Thanks in advance!