Run nvcr.io/nvidia/deepstream:5.1-21.02-triton on GeForce RTX 3090 GPU with CUDA 11.4

I’m trying to use nvcr.io/nvidia/deepstream:5.1-21.02-triton on computer with cuda 11.4. Everything builds fine but then I get error:

ERROR: ../nvdsinfer/nvdsinfer_func_utils.cpp:33 [TRT]: CUDA initialization failure with error 2. Please check your CUDA installation:  http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html
deepstream-app: ../nvdsinfer/nvdsinfer_model_builder.cpp:693: nvdsinfer::TrtModelBuilder::TrtModelBuilder(int, nvinfer1::ILogger&, const std::shared_ptr<nvdsinfer::DlLibHandle>&): Assertion `m_Builder' failed.
Aborted (core dumped)

Steps:

  1. docker pull nvcr.io/nvidia/deepstream:5.1-21.02-triton
  2. docker run --gpus all -it --rm --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 nvcr.io/nvidia/deepstream:5.1-21.02-triton
  3. cd /opt/nvidia/deepstream/deepstream-5.1/samples/configs/deepstream-app
  4. deepstream-app -c source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt
  5. And now I get the error

• Hardware Platform (Jetson / GPU) GPU NVIDIA GeForce RTX 3090 GPU
• DeepStream Version 5.1
• NVIDIA GPU Driver Version (valid for GPU only) 470.63.01

I can run well with this docker.

KLT Tracker Init
**PERF: 32.03 (31.89) 32.37 (32.22) 32.03 (31.89) 32.03 (31.89)
**PERF: 29.97 (30.89) 29.97 (31.05) 29.97 (30.89) 29.97 (30.89)

Did you reinstall cuda?