Failed to load CUDA modules. maybe the NVIDIA driver is not installed? Failed to initialize CUDA context. Failed to open VA display. Maybe the AMD video driver or libva-dev/libdrm-dev is not installed? Failed to open VAAPI display. Failed to load CUDA modules. maybe the NVIDIA driver is not installed? Failed to initialize CUDA context.

Thanks for reaching out, Billy. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

LiveKit cloud

Thanks for reaching out! Could you please let me know if you're seeing these errors while using LiveKit Cloud for a particular project or room? If you can share the project ID or room ID, that would really help us look into this for you.

im running lk agent logs

Let's work together to troubleshoot the CUDA and VAAPI errors you're seeing in your LiveKit agent logs. Here’s a step-by-step guide to help you through it: 1. Check CUDA Module Loading and Driver Version: • First, please make sure your NVIDIA driver is installed and updated to support at least CUDA 11.0. • You can verify the CUDA driver version by using

cuDriverGetVersion

. 2. Monitor and Collect Logs: • Try using the LiveKit CLI to monitor the agent status and collect logs. For example, running

lk agent logs

will let you follow runtime logs in real-time. • Also, enabling debug logging by setting the log level to 'debug' can provide more detailed insights. 3. Review Specific Error Messages: • Take a look at the error codes returned by CUDA APIs or NVENC (NVIDIA Encoder) — these often give valuable clues. 4. Check GPU and CUDA Context Initialization: • Make sure the CUDA context is properly initialized with

cuInit

and that a valid device is selected with

cuDeviceGet

. • Also, confirm that a CUDA context is successfully created with

cuCtxCreate

. 5. External Monitoring Integration: • If you want, you might consider forwarding runtime logs to external monitoring services like Datadog for more advanced analytics. If you find that the issue is still there after these steps, please don’t hesitate to share more details with us — things like the specific error messages, project ID, or room ID will really help us dig deeper and assist you better. Sources: Deploying to production | LiveKit Docs | Log collection | LiveKit Docs | client/web/.github/ISSUE_TEMPLATE/bug_report.yaml | Community & Support | livekit/livekit_composite | DeepWiki | webrtc-sys/src/nvidia/NvCodec/include/nvEncodeAPI.h

It is safe to ignore the CUDA messages.

Ok thanks

hey @refined-appointment-81829, nice to see you again. I am running the

from livekit.plugins.turn_detector.multilingual import MultilingualModel

turn detector and I am seeing the same error in my livekit cloud agent (might be unrelated, not sure). But since I am using a transformer based detector, I assume the GPU is needed. Should I ignore that error?

GPU is not required for the transformer turn detector from LiveKit

@refined-appointment-81829 not required but won't it be slower to detect without a GPU?

We current intend the inference to run on CPU and not GPU.

Thank you 🙏