Hello Step1X-3D Team,
First, thank you for this incredible project. I'm writing to request support for the new NVIDIA Blackwell (RTX 50-series) GPUs.
I've been attempting to run the Step1X-3D model on an NVIDIA GeForce RTX 5090 but have run into a hard-stop incompatibility issue.
The Core Problem: CUDA Version Conflict
The new RTX 5090 (Blackwell architecture, compute capability sm_120) requires CUDA 13 or newer to function.
The project's current dependencies appear to be hard-coded to CUDA 12.x (specifically torch==2.5.1+cu124 and the associated torch-cluster builds as seen in the ComfyUI-Step1X-3D node requirements).
Actual Behavior (Error Log)
When attempting to load the model, PyTorch correctly identifies this hardware/software mismatch and fails to initialize the GPU. This is the critical error message from the log:
NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
Feature Request
Would it be possible for the team to update the project's dependencies to support CUDA 13+ and the Blackwell sm_120 architecture?
Based on the installation files, this would likely involve:
Upgrading the PyTorch dependency to a newer build (e.g., a cu13x build).
Updating the torch-cluster dependency to a version compatible with the new PyTorch build.
Potentially re-compiling the custom step1x3d_texture/custom_rasterizer against the new CUDA 13 libraries.
As more users begin to adopt the RTX 50-series cards, adding Blackwell support would be a huge benefit for the community and ensure the project remains accessible.
Thank you for your time and consideration!
Hello Step1X-3D Team,
First, thank you for this incredible project. I'm writing to request support for the new NVIDIA Blackwell (RTX 50-series) GPUs.
I've been attempting to run the Step1X-3D model on an NVIDIA GeForce RTX 5090 but have run into a hard-stop incompatibility issue.
The Core Problem: CUDA Version Conflict
The new RTX 5090 (Blackwell architecture, compute capability sm_120) requires CUDA 13 or newer to function.
The project's current dependencies appear to be hard-coded to CUDA 12.x (specifically torch==2.5.1+cu124 and the associated torch-cluster builds as seen in the ComfyUI-Step1X-3D node requirements).
Actual Behavior (Error Log)
When attempting to load the model, PyTorch correctly identifies this hardware/software mismatch and fails to initialize the GPU. This is the critical error message from the log:
NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
Feature Request
Would it be possible for the team to update the project's dependencies to support CUDA 13+ and the Blackwell sm_120 architecture?
Based on the installation files, this would likely involve:
Upgrading the PyTorch dependency to a newer build (e.g., a cu13x build).
Updating the torch-cluster dependency to a version compatible with the new PyTorch build.
Potentially re-compiling the custom step1x3d_texture/custom_rasterizer against the new CUDA 13 libraries.
As more users begin to adopt the RTX 50-series cards, adding Blackwell support would be a huge benefit for the community and ensure the project remains accessible.
Thank you for your time and consideration!