RTX 5090 support?

[quasiblob]

Hi.

So does this app currently work with RTX 5090?

I've tried to install this app, however I can't get the app generating models, it will give error after it has started generating a mesh.

NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.

I think I've got all the correct versions installed. CUDA 12.8 in Windows 11, Python 3.12 (in a venv that I created, I know there is a python folder for this app itself), and tried different (latest) versions of pytorch. I also tried recommended 2.6 Pytorch, but does it actually support 50-series card? Anyway, I didn't get this app working with that either.

I'm not that well versed with Python or its packages and package management at all, so this might be a user error.

[quasiblob]

Actually, I figured this out - if anyone else wants to simply run the standard version, that too doesn't require too much Python skills.

Someone has made a ready to run set of commands:

NOTE:
There are some steps that only work with Python 3.10!
I tried with Python 3.11 and got some errors related to removed features.
So create your venv folder / virtual environment with Python 3.10.
If you only have one Python version on your system that is 3.10. then no worries.

Also, at the time of writing this, Pytorch nightly version used by the script here worked with CUDA 12.8 Toolkit which I have installed in my system.

Run these commands one at time:

git clone https://github.com/Tencent/Hunyuan3D-2.git
cd Hunyuan3D-2
git pull
python -m venv venv
call venv\Scripts\activate.bat
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
pip install -r requirements.txt
cd hy3dgen/texgen/custom_rasterizer
python setup.py install
cd ../../..
cd hy3dgen/texgen/differentiable_renderer
python setup.py install
cd ../../..

BTW - this takes a while, setup scripts downloads gigabytes of models.

And when all the steps are done, you can create a run.bat file (call it whatever you want and place it in the root of Hunyuan 3d folder):

call venv\Scripts\activate.bat
python gradio_app.py --model_path tencent/Hunyuan3D-2 --subfolder hunyuan3d-dit-v2-0-turbo --texgen_model_path tencent/Hunyuan3D-2 --low_vram_mode --enable_flashvdm

Both the model and the texture creation seemed to be working.

Source for the solution can be found here - thanks to user Aikon404:
Tencent-Hunyuan/Hunyuan3D-2#140

[YanWenKun]

Check out the pre-release for CUDA 12.8:

https://github.com/YanWenKun/Hunyuan3D-2-WinPortable/releases/tag/v3.1