Help with gaussian splatting torch 4090 torch Version: 2.1.0

windows
1

Hello guys, I am trying to run the comand line on a Gaussian Splatting test.

python train.py -s

After this comand I got the message:

C:\Users\marcos.costa\AppData\Local\anaconda3\envs\gaussian_splatting\lib\site-packages\torch\cuda_init_.py:106: UserWarning:
NVIDIA GeForce RTX 4090 with CUDA capability sm_89 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 compute_37.
If you want to use the NVIDIA GeForce RTX 4090 GPU with PyTorch, please check the instructions at Start Locally | PyTorch

warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))
Output folder: ./output/1ea37bd9-2 [13/10 20:19:08]
Tensorboard not available: not logging progress [13/10 20:19:08]
Reading camera 149/149 [13/10 20:19:09]
Loading Training Cameras [13/10 20:19:09]
[ INFO ] Encountered quite large input images (>1.6K pixels width), rescaling to 1.6K.
If this is not desired, please explicitly specify ‘–resolution/-r’ as 1 [13/10 20:19:09

Can anyone help me?

2

Based on the error message you’ve installed a PyTorch binary with CUDA 10.2, which is too old for your GPU. Install any current stable or nightly release and it will work.

3

I have cuda 11.8 and 12.2 installed and the same error proceeds.

Is there a problem to have more than one version?

4

I doubt it, since e.g. 2.1.0+cu121 shows these architectures:

torch.__version__
Out[111]: '2.1.0+cu121'

torch.cuda.get_arch_list()
Out[112]: ['sm_50', 'sm_60', 'sm_70', 'sm_75', 'sm_80', 'sm_86', 'sm_90']

while your error message shows the ones built with CUDA 10.2.