ERROR: Cuda initialization failure with error 30

Accelerated Computing CUDA CUDA Programming and Performance
1

I got "ERROR: Cuda initialization failure with error 30. " when I tried to run sample_mnist code on TensorRT. What does that error mean?

Environment:
Ubuntu 16.04
TensorRT 5.0.2.6-1+cuda9.0
CUDA 9.0
cuDNN 7
Python3

Thanks!

2

Hello,

cudaErrorUnknown = 30 “This indicates that an unknown internal error has occurred.” (I know, not very helpful).

To rule out issues specific to your configuration, consider trying the TensorRT container (available at https://www.nvidia.com/en-us/gpu-cloud/).

Moving to CUDA for better support coverage.

3

This is often an indicator of a broken CUDA install.

After installing CUDA, the install should be verified using the instructions in the linux install guide.

4

Thanks for informing that.

5

I found the same problem.
The indication of the issue is:
$ modprobe nvidia
If it’s not silent then check/change alternatives of x86_64-linux-gnu_gl_conf by
$ sudo update-alternatives --config x86_64-linux-gnu_gl_conf
It should point to nvidia file and not mesa one.