AMD 7900XTX - Windows 11 - ROCm 7.1.1 - Preview Driver 25.20.01.17 - Python 3.12 - ERROR "cumesh==0.1.0 cannot be used"

[OrsoEric]

I am running 7900XTX ROCm under bare windows that is possible since november. I use the preview driver that allows pytorch ROCm to work under bare windows, it does accelerate Zimage with ComfyUI:

• https://rocm.docs.amd.com/projects/radeon-ryzen/en/latest/docs/install/installrad/windows/install-pytorch.html

I can run Hunyuan 2.1 competently

I tried Trellis 2 with ROCm under windows.

ERROR CuMesh

CuMesh doesn't have Python 3.12 binaries.

ROCm 7.1.1 doesn't have Python 3.10/3.11 binaries

F:\ComfyUI-Windows-2026-01-04>.venv\Scripts\activate

(.venv) F:\ComfyUI-Windows-2026-01-04>cd custom_nodes

(.venv) F:\ComfyUI-Windows-2026-01-04\custom_nodes>cd ComfyUI-Trellis2

(.venv) F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2>dir
 Volume in drive F is Data
 Volume Serial Number is 8AC9-D6F0

 Directory of F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2

2026-01-04  13:44    <DIR>          .
2026-01-04  13:44    <DIR>          ..
2026-01-04  13:44                11 .gitignore
2026-01-04  13:44    <DIR>          example_workflows
2026-01-04  13:44            74,332 nodes.py
2026-01-04  13:44             4,897 README.md
2026-01-04  13:44                50 requirements.txt
2026-01-04  13:44    <DIR>          trellis2
2026-01-04  13:44    <DIR>          wheels
2026-01-04  13:44               150 __init__.py
2026-01-04  13:44    <DIR>          __pycache__
               5 File(s)         79,440 bytes
               6 Dir(s)  926,865,559,552 bytes free

uv pip install -r "requirements.txt"

(.venv) F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2>uv pip install trimesh
Using Python 3.12.10 environment at: F:\ComfyUI-Windows-2026-01-04\.venv
Resolved 2 packages in 219ms
Prepared 1 package in 97ms
░░░░░░░░░░░░░░░░░░░░ [0/1] Installing wheels...                                                                         warning: Failed to hardlink files; falling back to full copy. This may lead to degraded performance.
         If the cache and target directories are on different filesystems, hardlinking may not be supported.
         If this is intentional, set `export UV_LINK_MODE=copy` or use `--link-mode=copy` to suppress this warning.
Installed 1 package in 29ms
 + trimesh==4.10.1

Traceback (most recent call last):
  File "F:\ComfyUI-Windows-2026-01-04\nodes.py", line 2151, in load_custom_node
    module_spec.loader.exec_module(module)
  File "<frozen importlib._bootstrap_external>", line 999, in exec_module
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2\__init__.py", line 1, in <module>
    from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
  File "F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2\nodes.py", line 20, in <module>
    import cumesh as CuMesh
ModuleNotFoundError: No module named 'cumesh'

uv pip install CuMesh --no-build-isolation

(.venv) F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2>uv pip install CuMesh --no-build-isolation
Using Python 3.12.10 environment at: F:\ComfyUI-Windows-2026-01-04\.venv
  x No solution found when resolving dependencies:
  `-> Because python was not found in the package registry and cumesh==0.1.0 depends on python>=3.8, we can conclude
      that cumesh==0.1.0 cannot be used.
      And because only cumesh==0.1.0 is available and you require cumesh, we can conclude that your requirements are
      unsatisfiable.

INSTALL ROCm Windows

Installation script for ComfyUI windows

# Ensure uv is installed
if (-not (Get-Command uv -ErrorAction SilentlyContinue)) {
    Write-Host "Installing uv..."
    powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/0.9.16/install.ps1 | iex"
} else {
    Write-Host "uv already installed."
}

# Clone ComfyUI into current directory
git init
git remote add origin https://github.com/comfyanonymous/ComfyUI.git
git fetch
git checkout -t origin/master

# Create venv with Python 3.12
uv venv .venv --python 3.12
.venv\Scripts\Activate.ps1

# Install ROCm SDK and libraries
uv pip install --no-cache-dir `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_core-0.1.dev0-py3-none-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_devel-0.1.dev0-py3-none-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm_sdk_libraries_custom-0.1.dev0-py3-none-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/rocm-0.1.dev0.tar.gz

# Install ROCm-enabled PyTorch stack
uv pip install --no-cache-dir `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torch-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchaudio-2.9.0+rocmsdk20251116-cp312-cp312-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.1.1/torchvision-0.24.0+rocmsdk20251116-cp312-cp312-win_amd64.whl

# Verify installation
try {
    python -c "import torch" | Out-Null
    Write-Host "Torch import: Success"
} catch {
    Write-Host "Torch import: Failure"
}

Write-Host "CUDA available:" (python -c "import torch; print(torch.cuda.is_available())")
Write-Host "Device name:" (python -c "import torch; print(torch.cuda.get_device_name(0))")
python -m torch.utils.collect_env

# Install ComfyUI custom node manager
Set-Location custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager comfyui-manager
Set-Location ..

# Install ComfyUI requirements
uv pip install -r "requirements.txt"

# Run ComfyUI
uv run main.py

Script to collect environment information:

import torch

def check_torch():
    print("--- Torch Import ---")
    try:
        import torch
        print("Success: torch imported successfully.")
    except ImportError:
        print("Failure: torch import failed.")
        return

    print("\n--- CUDA Availability ---")
    try:
        print(f"CUDA available: {torch.cuda.is_available()}")
    except Exception as e:
        print(f"Failure: CUDA check failed. Error: {e}")

    print("\n--- Device Name ---")
    try:
        if torch.cuda.is_available():
            print(f"Device name [0]: {torch.cuda.get_device_name(0)}")
        else:
            print("No CUDA device found.")
    except Exception as e:
        print(f"Failure: Device name check failed. Error: {e}")

    print("\n--- Environment Info ---")
    try:
        # Collect environment information
        env_info = env_info = torch.utils.collect_env.get_pretty_env_info()

        # Print the environment information
        print(env_info)
        
    except Exception as e:
        print(f"Failure: Environment info collection failed. Error: {e}")

if __name__ == "__main__":
    check_torch()

Environment information

--- Torch Import ---
Success: torch imported successfully.

--- CUDA Availability ---
CUDA available: True

--- Device Name ---
Device name [0]: AMD Radeon RX 7900 XTX

--- Environment Info ---
PyTorch version: 2.9.0+rocmsdk20251116
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 7.1.52802-561cc400e1

OS: Microsoft Windows 11 Pro (10.0.22631 64-bit)
GCC version: (MinGW-W64 x86_64-ucrt-posix-seh, built by Brecht Sanders, r8) 13.2.0
Clang version: 19.0.0git (git@github.amd.com:Compute-Mirrors/llvm-project 5353ca3e0e5ae54a31eeebe223da212fa405567a)
CMake version: version 3.29.2
Libc version: N/A

Python version: 3.12.10 (main, Apr  9 2025, 04:06:22) [MSC v.1943 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-11-10.0.22631-SP0
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to:
GPU models and configuration: AMD Radeon RX 7900 XTX (gfx1100)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
Is XPU available: False
HIP runtime version: 7.1.52802
MIOpen runtime version: 3.5.1
Is XNNPACK available: True

CPU:
Name: 13th Gen Intel(R) Core(TM) i7-13700F
Manufacturer: GenuineIntel
Family: 198
Architecture: 9
ProcessorType: 3
DeviceID: CPU0
CurrentClockSpeed: 2100
MaxClockSpeed: 2100
L2CacheSize: 24576
L2CacheSpeed: None
Revision: None

Versions of relevant libraries:
[pip3] Could not collect
[conda] Could not collect

[visualbruno]

You cannot build Cumesh on Windows from the source repository (https://github.com/JeffreyXiang/CuMesh), it will crash.
I had to modify it to compile it on Windows.

Did you try to install the precompiled wheels I put in the folder "wheels" : (https://github.com/visualbruno/ComfyUI-Trellis2/tree/main/wheels/Windows)

[OrsoEric]

Thanks for the answer

Your wheel you provided installs fine with python 3.12

uv pip install "https://github.com/visualbruno/ComfyUI-Trellis2/raw/main/wheels/Windows/Torch280/cumesh-0.0.1-cp312-cp312-win_amd64.whl"

(.venv) F:\ComfyUI-Windows-2026-01-04>uv pip install "https://github.com/visualbruno/ComfyUI-Trellis2/raw/main/wheels/Windows/Torch280/cumesh-0.0.1-cp312-cp312-win_amd64.whl"
Resolved 16 packages in 1.67s
Prepared 1 package in 347ms
░░░░░░░░░░░░░░░░░░░░ [0/1] Installing wheels...                                                                         warning: Failed to hardlink files; falling back to full copy. This may lead to degraded performance.
         If the cache and target directories are on different filesystems, hardlinking may not be supported.
         If this is intentional, set `export UV_LINK_MODE=copy` or use `--link-mode=copy` to suppress this warning.
Installed 1 package in 17ms
 + cumesh==0.0.1 (from https://github.com/visualbruno/ComfyUI-Trellis2/raw/main/wheels/Windows/Torch280/cumesh-0.0.1-cp312-cp312-win_amd64.whl)

The wheel is for torch 2.8.0 and seems to depend on a CUDA DLL that is incompatible with torch rocm as far as I can tell

uv run main.py

Traceback (most recent call last):
  File "F:\ComfyUI-Windows-2026-01-04\nodes.py", line 2151, in load_custom_node
    module_spec.loader.exec_module(module)
  File "<frozen importlib._bootstrap_external>", line 999, in exec_module
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2\__init__.py", line 1, in <module>
    from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
  File "F:\ComfyUI-Windows-2026-01-04\custom_nodes\ComfyUI-Trellis2\nodes.py", line 20, in <module>
    import cumesh as CuMesh
  File "F:\ComfyUI-Windows-2026-01-04\.venv\Lib\site-packages\cumesh\__init__.py", line 1, in <module>
    from . import remeshing
  File "F:\ComfyUI-Windows-2026-01-04\.venv\Lib\site-packages\cumesh\remeshing.py", line 4, in <module>
    from . import _C
ImportError: DLL load failed while importing _C: The specified module could not be found.

SCRIPT

import os
import importlib.util

import torch

print("PyTorch location:", os.path.dirname(importlib.util.find_spec("torch").origin))

def check_torch():
    print("--- Torch Import ---")
    try:
        import torch
        print("Success: torch imported successfully.")
    except ImportError:
        print("Failure: torch import failed.")
        return

    print("\n--- CUDA Availability ---")
    try:
        print(f"CUDA available: {torch.cuda.is_available()}")
    except Exception as e:
        print(f"Failure: CUDA check failed. Error: {e}")

    print("\n--- Device Name ---")
    try:
        if torch.cuda.is_available():
            print(f"Device name [0]: {torch.cuda.get_device_name(0)}")
        else:
            print("No CUDA device found.")
    except Exception as e:
        print(f"Failure: Device name check failed. Error: {e}")

    print("\n--- Environment Info ---")
    try:
        # Collect environment information
        env_info = env_info = torch.utils.collect_env.get_pretty_env_info()

        # Print the environment information
        print(env_info)
        
    except Exception as e:
        print(f"Failure: Environment info collection failed. Error: {e}")


check_torch()

import cumesh as CuMesh

print("CuMesh location:", os.path.dirname(importlib.util.find_spec("cumesh").origin))

SCRIPT OUTPUT

(.venv) F:\ComfyUI-Windows-2026-01-04>uv run verify_trellis.py
PyTorch location: F:\ComfyUI-Windows-2026-01-04\.venv\Lib\site-packages\torch
--- Torch Import ---
Success: torch imported successfully.

--- CUDA Availability ---
CUDA available: True

--- Device Name ---
Device name [0]: AMD Radeon RX 7900 XTX

--- Environment Info ---
PyTorch version: 2.9.0+rocmsdk20251116
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 7.1.52802-561cc400e1

OS: Microsoft Windows 11 Pro (10.0.22631 64-bit)
GCC version: Could not collect
Clang version: 19.0.0git (git@github.amd.com:Compute-Mirrors/llvm-project 5353ca3e0e5ae54a31eeebe223da212fa405567a)
CMake version: version 4.2.1
Libc version: N/A

Python version: 3.12.10 (main, Apr  9 2025, 04:06:22) [MSC v.1943 64 bit (AMD64)] (64-bit runtime)
Python platform: Windows-11-10.0.22631-SP0
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to:
GPU models and configuration: AMD Radeon RX 7900 XTX (gfx1100)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
Is XPU available: False
HIP runtime version: 7.1.52802
MIOpen runtime version: 3.5.1
Is XNNPACK available: True

CPU:
Name: 13th Gen Intel(R) Core(TM) i7-13700F
Manufacturer: GenuineIntel
Family: 198
Architecture: 9
ProcessorType: 3
DeviceID: CPU0
CurrentClockSpeed: 2100
MaxClockSpeed: 2100
L2CacheSize: 24576
L2CacheSpeed: None
Revision: None

Versions of relevant libraries:
[pip3] Could not collect
[conda] Could not collect
Traceback (most recent call last):
  File "F:\ComfyUI-Windows-2026-01-04\verify_trellis.py", line 48, in <module>
    import cumesh as CuMesh
  File "F:\ComfyUI-Windows-2026-01-04\.venv\Lib\site-packages\cumesh\__init__.py", line 1, in <module>
    from . import remeshing
  File "F:\ComfyUI-Windows-2026-01-04\.venv\Lib\site-packages\cumesh\remeshing.py", line 4, in <module>
    from . import _C
ImportError: DLL load failed while importing _C: The specified module could not be found.

DEPENDENCIES

The binary is there, I do wonder if it fails to link something (more likely I think since I don't have the CUDA dlls) or if the issue is somewhere else.

remeshing.py

from typing import *
import torch
from tqdm import tqdm
from . import _C

I used dll tool to check the dependencies

I kept looking since the pytorch ROCm bindings seems to work well, and ROCm does have torch-python.dll I wonder if it's just matter to make remeshing look into the right folder