pycuda._driver.Error: cuInit failed: unknown error

Accelerated Computing Intelligent Video Analytics TAO Toolkit
, ,
1

Hey,
I’m trying to follow Jupyter-notbook yolo_v4.ipynb tutorial. I got the error described in the image that is attached here. I don’t know how to solve it. please help!!!

the specs are:
• Hardware RX 3070 GPU
• Network Type Yolo_v4
• TAO Version 3.21.11

image
image1081×674 202 KB

2

Refer to Train with my own tlt model #2 - #17 by tmid0819
and No CUDA-capable device is detected on tao detectnet_v2 dataset convert - #4 by NilsAI

3

Hey,

Thanks for fast reply. I tried to erase the “tao” from the command line, didn’t work. the second suggestion, to add --privileged, I don’t run “docker run” command, just using the jupyter notebook as it is. How can I add --privileged to the notebook?

I run a docker with --privileged and I ran a dataset_convert command inside the container. this is what I got.

Screenshot from 2021-12-29 23-59-33
Screenshot from 2021-12-29 23-59-331920×1080 378 KB

Thanks a lot

4

May I know the result of below
$ nvidia-smi

See TAO container fails on Google Vertex AI - #58 by mattcarp88, that topic has similar error.
For that topic, need to update nvidia-driver.

5

The results of nvidia-smi command:

Screenshot from 2021-12-30 22-25-23
Screenshot from 2021-12-30 22-25-231920×1080 164 KB

What version of the driver should I install?

Thanks a lot for your assist!

6

Can you add “--runtime=nvidia” when you run docker run?

More, may I know if you can reproduce the error with below steps?

$ python
>>> from modulus.export._tensorrt import ONNXEngineBuilder

7

Well, it looks like the problem is that I run Ubuntu 20.04. Now that I switched to Ubuntu 18.04 everything is ok.

8

Thanks for the info. Currently only Ubuntu 18.04 is full verified.