search

bennyguo / instant-nsr-pl

Neural Surface reconstruction based on Instant-NGP. Efficient and customizable boilerplate for your research projects. Train NeuS in 10min!
MIT License
857 stars 84 forks source link

ImportError: /home/anaconda3/envs/instant-nsr-pl/lib/python3.8/site-packages/tinycudann_bindings_61/_C.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZNK2at6Tensor6deviceEv #7

Closed sunshineywz123 closed 2 years ago

sunshineywz123 commented 2 years ago

企业微信截图_16677204873554 ImportError: /home/anaconda3/envs/instant-nsr-pl/lib/python3.8/site-packages/tinycudann_bindings_61/_C.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZNK2at6Tensor6deviceEv

I have this error, according to the readme operation down, google and issue did not find a similar problem, you have encountered there?

bennyguo commented 2 years ago

What PyTorch version are you using? How did you install tiny-cuda-nn?

sunshineywz123 commented 2 years ago

What PyTorch version are you using? How did you install tiny-cuda-nn?

1.>>> torch.version '1.13.0' 2.pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch image

bennyguo commented 2 years ago

What GPU are you using? It seems that your tiny-cuda-nn is compiled under sm_61, which corresponds to NVIDIA GTX 10 series.

I suggest you open an issue in the original tiny-cuda-nn repo and see if they have some problems supporting old GPUs.

sunshineywz123 commented 2 years ago

image GeForce RTX 208..

bennyguo commented 2 years ago

You may need to check the CUDA compiler version by nvcc -V, and make sure it's consistent with the installed torch wheel.

sunshineywz123 commented 2 years ago

You may need to check the CUDA compiler version by nvcc -V, and make sure it's consistent with the installed torch wheel.

I can't install nvcc in my environment, can you provide a google colab project for us to test? Or is there any other means to confirm the environment besides nvcc?

bennyguo commented 2 years ago

nvcc is shipped with your CUDA installation. If you use Anaconda to install PyTorch, a corresponding CUDA Toolkit is automatically installed by Anaconda. You may use the following command to find the CUDA installation path:

sudo find / -name nvcc

After that, you may update your $PATH and $LD_LIBRARY_PATH to this CUDA installation and re-install tiny-cuda-nn.

sunshineywz123 commented 2 years ago

nvcc is shipped with your CUDA installation. If you use Anaconda to install PyTorch, a corresponding CUDA Toolkit is automatically installed by Anaconda. You may use the following command to find the CUDA installation path:

sudo find / -name nvcc

After that, you may update your $PATH and $LD_LIBRARY_PATH to this CUDA installation and re-install tiny-cuda-nn.

nvcc -V nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2020 NVIDIA Corporation Built on Tue_Sep_15_19:10:02_PDT_2020 Cuda compilation tools, release 11.1, V11.1.74 Build cuda_11.1.TC455_06.29069683_0

bennyguo commented 2 years ago

Where is this nvcc located? Is it Anaconda-installed?

When installing PyTorch using Anaconda, what CUDA version did you specify?

sunshineywz123 commented 2 years ago

Where is this nvcc located? Is it Anaconda-installed?

When installing PyTorch using Anaconda, what CUDA version did you specify?

which nvcc /usr/local/cuda/bin/nvcc

bennyguo commented 2 years ago

Maybe you could try installing torch1.10.1 with CUDA11.3:

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

Then re-install tiny-cuda-nn.

sunshineywz123 commented 2 years ago

Maybe you could try installing torch1.10.1 with CUDA11.3:

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

Then re-install tiny-cuda-nn.

it's ok,thanks