FATL| Cuda Error: CUDA driver version is insufficient for CUDA runtime version

[jiadingfang]

Hi,

I ran into this error when trying to run "start normal" in the live input section. But the application window crashes and terminal shows the error info:

Stack trace:
8       0x7f291da2388f clone + 63
7       0x7f291b7ae6db /lib/x86_64-linux-gnu/libpthread.so.0(+0x76db) [0x7f291b7ae6db]
6       0x7f291e36666f /usr/lib/x86_64-linux-gnu/libstdc++.so.6(+0xbd66f) [0x7f291e36666f]
5       0x7f292035103d vis::ImageInputRealSense::ThreadMain(std::vector<std::shared_ptr<vis::AvailableInput>, std::allocator<std::shared_ptr<vis::AvailableInput> > > const&) + 6925
4       0x7f29203e8289 vis::LiveImageConsumer::NewImageset(std::vector<vis::Image<Eigen::Matrix<unsigned char, 3, 1, 0, 3, 1> >, std::allocator<vis::Image<Eigen::Matrix<unsigned char, 3, 1, 0, 3, 1> > > > const&) + 601
3       0x7f292033c625 vis::FeatureDetectorTaggedPattern::DetectFeatures(vis::Image<Eigen::Matrix<unsigned char, 3, 1, 0, 3, 1> > const&, std::vector<vis::PointFeature>*, vis::Image<Eigen::Matrix<unsigned char, 3, 1, 0, 3, 1> >*) + 501
2       0x7f2920332a62 vis::FeatureDetectorTaggedPattern::PrepareCUDAResources(int, int) + 66
1       0x7f2920340cde vis::CUDABuffer<unsigned char>::CUDABuffer(int, int) + 206
0       0x7f291f912fe2 loguru::StreamLogger::~StreamLogger() + 162
23:30:27.332       cuda_buffer_inl.h:40    FATL| Cuda Error: CUDA driver version is insufficient for CUDA runtime version

So I checked my cuda and driver version:

root@568df16a4713:/# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Wed_Oct_23_19:24:38_PDT_2019
Cuda compilation tools, release 10.2, V10.2.89
root@568df16a4713:/# cat /proc/driver/nvidia/version
NVRM version: NVIDIA UNIX x86_64 Kernel Module  465.19.01  Fri Mar 19 07:44:41 UTC 2021
GCC version:  gcc version 9.3.0 (Ubuntu 9.3.0-17ubuntu1~20.04)

According to https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#major-components, my driver version (465.19) is more than enough for my cuda version (10.2).

Could you suggest how this could happen and what could possibly the way to solve it? Thanks!

[puzzlepaint]

Unfortunately, I don't know what could cause this given that the driver version should theoretically be sufficient. Two things to try might be:

• In case you haven't, restart the computer after installing the driver.
• Try running any other CUDA program to see whether it is a general issue with the CUDA installation, or only affects the camera calibration program.