Cannot transform tensor to CUDA after "from diff_gaussian_rasterization import GaussianRasterizationSettings, GaussianRasterizer"

[RyanPham19092002]

Hi! Thanks for your great work!
But I have some issues with this pacakage.
I try to import torch and transform tensor to CUDA after I install package, but it error like the image below

(hierarchical_3d_gaussians) vinai@gausian-splatting:/data/Phat/hierarchical-3d-gaussians$ nvidia-smi
Sun Aug 18 08:29:35 2024
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.80 Driver Version: 460.80 CUDA Version: 11.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 GeForce RTX 208... Off | 00000000:17:00.0 Off | N/A |
| 44% 44C P0 56W / 260W | 0MiB / 11019MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
(hierarchical_3d_gaussians) vinai@gausian-splatting:/data/Phat/hierarchical-3d-gaussians$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:33:58_PDT_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0
P/s : g++ and gcc have version 11.2
Can you help me for this issues ? Thanks for your help

[ameuleman]

Hi,
Would it be possible to update the NVIDIA driver?

[RyanPham19092002]

Thanks for your reply.
I'm sorry but I cannot do that, I don't have permission for updating NVIDIA driver on this server.
Do you have any solutions for this problem but I don't need to update NVIDIA driver ?

[ameuleman]

There seems to be a mismatch between the CUDA compiler and the NVIDIA driver. If updating the driver is not an option, downgrading NVCC to 11.1 and installing a pytorch version compiled with cuda 11.1 should solve the issue. Note that we have not tested our codebase with these versions.

[RyanPham19092002]

Thanks for your suggestion.
But I based on repo hierarchical-3d-gaussians , and I follow the guide and it need to install torch CUDA version 11.8 ? Downgrading NVCC to 11.1 is suitable ?

[ameuleman]

Given your constraints, this is the best solution that comes to mind and it is likely to work. Again, we have not tested with CUDA 11.1 or pytorch 1.x.

[RyanPham19092002]

Thanks for your help, I will try this
P/s : my pytorch version is 2.0.0+cu118

[ameuleman]

To ensure pytorch is also compatible with your driver, I would install an older pytorch version.
After uninstalling pytorch 2.0.0+cu118, run:

 conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=11.1 -c pytorch -c conda-forge

[RyanPham19092002]

Hi, I try to install but it cannot install

Do you have another solutions ? Thanks for your help

[ameuleman]

This version of pytorch requires an older python version:

 conda install python==3.9
 pip install torch==1.10.1+cu111 torchvision==0.11.2 torchaudio==0.10.1 -f https://download.pytorch.org/whl/cu111/torch_stable.html
 pip install -r requirements.txt 

[RyanPham19092002]

Hi
Thanks for your help, it's work.
Thank you so much