Error message at start trying to use triton

[Ashir555]

Custom Node Testing

• I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)

Your question

New clone, install-n. at end I got the following error

*** Checking and updating to new version if possible
Already up to date.

[START] Security scan
[DONE] Security scan
## ComfyUI-Manager: installing dependencies done.
** ComfyUI startup time: 2025-08-24 19:39:11.539
** Platform: Windows
** Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
** Python executable: G:\IMG Gen\ComfyUI-Zluda-T\venv\Scripts\python.exe
** ComfyUI Path: G:\IMG Gen\ComfyUI-Zluda-T
** ComfyUI Base Folder Path: G:\IMG Gen\ComfyUI-Zluda-T
** User directory: G:\IMG Gen\ComfyUI-Zluda-T\user
** ComfyUI-Manager config path: G:\IMG Gen\ComfyUI-Zluda-T\user\default\ComfyUI-Manager\config.ini
** Log path: G:\IMG Gen\ComfyUI-Zluda-T\user\comfyui.log

Prestartup times for custom nodes:
   4.3 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\ComfyUI-Manager

Checkpoint files will always be loaded safely.

  ::  Checking package versions...
Found pydantic: 2.11.7, pydantic-settings: 2.10.1
  ::  Pydantic packages are compatible, skipping reinstall
Installed version of comfyui-frontend-package: 1.25.10
Installed version of comfyui-workflow-templates: 0.1.65
Installed version of av: 15.0.0
Installed version of comfyui-embedded-docs: 0.2.6
  ::  Package version check complete.

  ::  ------------------------ ZLUDA -----------------------  ::
  ::  Auto-detecting AMD GPU architecture for Triton...
  ::  Detected GPU via Windows registry: AMD Radeon RX 5700 XT
  ::  Set TRITON_OVERRIDE_ARCH=gfx1010
  ::  Triton core imported successfully
  ::  Running Triton kernel test...
%%% [info] triton/runtime/build/platform_key: AMD64,Windows,64bit,WindowsPE
Using HIP SDK Clang.
%%% Triton is using the compiler 'C:\Program Files\AMD\ROCm\6.2\bin\clang.exe'
%%% [info] triton/runtime/build/_build: "C:\Program Files\AMD\ROCm\6.2\bin\clang.exe" C:\Users\Angel\AppData\Local\Temp\tmpjqtm8pxy\hip_utils.c -O3 -shared -Wno-psabi -march=native -mtune=native -o C:\Users\Angel\AppData\Local\Temp\tmpjqtm8pxy\hip_utils.cp311-win_amd64.pyd "-LG:\IMG Gen\ComfyUI-Zluda-T\venv\libs" "-LC:\Program Files\AMD\ROCm\6.2\lib" "-IG:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\include" -IC:\Users\Angel\AppData\Local\Temp\tmpjqtm8pxy -IC:\Users\Angel\AppData\Local\Programs\Python\Python311\Include
In file included from C:\Users\Angel\AppData\Local\Temp\tmpjqtm8pxy\hip_utils.c:8:
In file included from C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\Python.h:87:
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:184:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  184 |     struct timeval *tv,
      |            ^
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:190:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  190 |     struct timeval *tv,
      |            ^
2 warnings generated.
clang: error: linker command failed due to signal (use -v to see invocation)
  ::  Triton test failed: Command '['C:\\Program Files\\AMD\\ROCm\\6.2\\bin\\clang.exe', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmpjqtm8pxy\\hip_utils.c', '-O3', '-shared', '-Wno-psabi', '-march=native', '-mtune=native', '-o', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmpjqtm8pxy\\hip_utils.cp311-win_amd64.pyd', '-LG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\libs', '-LC:\\Program Files\\AMD\\ROCm\\6.2\\lib', '-IG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\Lib\\site-packages\\triton\\backends\\amd\\include', '-IC:\\Users\\Angel\\AppData\\Local\\Temp\\tmpjqtm8pxy', '-IC:\\Users\\Angel\\AppData\\Local\\Programs\\Python\\Python311\\Include']' returned non-zero exit status 1.
  ::  Triton available but failed verification
  ::  Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
  ::  Using ZLUDA with device: AMD Radeon RX 5700 XT [ZLUDA]
  ::  Applying core ZLUDA patches...
  ::  Triton optimizations skipped (not available)
  ::  Configuring PyTorch backends...
  ::  Disabled CUDA flash attention
  ::  Enabled math attention fallback
  ::  ZLUDA initialization complete
  ::  ------------------------ ZLUDA -----------------------  ::

Total VRAM 8176 MB, total RAM 16294 MB
pytorch version: 2.7.0+cu118
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 5700 XT [ZLUDA] : native
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.3.52
ComfyUI frontend version: 1.25.10
[Prompt Server] web root: G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\comfyui_frontend_package\static
Using HIP SDK Clang.
%%% Triton is using the compiler 'C:\Program Files\AMD\ROCm\6.2\bin\clang.exe'
%%% [info] triton/runtime/build/_build: "C:\Program Files\AMD\ROCm\6.2\bin\clang.exe" C:\Users\Angel\AppData\Local\Temp\tmpnr1m3aw4\hip_utils.c -O3 -shared -Wno-psabi -march=native -mtune=native -o C:\Users\Angel\AppData\Local\Temp\tmpnr1m3aw4\hip_utils.cp311-win_amd64.pyd "-LG:\IMG Gen\ComfyUI-Zluda-T\venv\libs" "-LC:\Program Files\AMD\ROCm\6.2\lib" "-IG:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\include" -IC:\Users\Angel\AppData\Local\Temp\tmpnr1m3aw4 -IC:\Users\Angel\AppData\Local\Programs\Python\Python311\Include
In file included from C:\Users\Angel\AppData\Local\Temp\tmpnr1m3aw4\hip_utils.c:8:
In file included from C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\Python.h:87:
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:184:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  184 |     struct timeval *tv,
      |            ^
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:190:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  190 |     struct timeval *tv,
      |            ^
2 warnings generated.
clang: error: linker command failed due to signal (use -v to see invocation)
Traceback (most recent call last):
  File "G:\IMG Gen\ComfyUI-Zluda-T\nodes.py", line 2129, in load_custom_node
    module_spec.loader.exec_module(module)
  File "<frozen importlib._bootstrap_external>", line 940, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "G:\IMG Gen\ComfyUI-Zluda-T\comfy_extras\nodes_canny.py", line 1, in <module>
    from kornia.filters import canny
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\__init__.py", line 25, in <module>
    from . import (
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\__init__.py", line 19, in <module>
    from kornia.augmentation import auto, container
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\__init__.py", line 18, in <module>
    from .autoaugment import AutoAugment
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\autoaugment\__init__.py", line 18, in <module>
    from .autoaugment import AutoAugment
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\autoaugment\autoaugment.py", line 22, in <module>
    from kornia.augmentation.auto.base import SUBPOLICY_CONFIG, PolicyAugmentBase
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\base.py", line 22, in <module>
    from kornia.augmentation.auto.operations.base import OperationBase
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\operations\__init__.py", line 20, in <module>
    from .policy import PolicySequential
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\operations\policy.py", line 24, in <module>
    from kornia.augmentation.container.base import ImageSequentialBase, TransformMatrixMinIn
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\container\__init__.py", line 18, in <module>
    from kornia.augmentation.container.augment import AugmentationSequential
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\container\augment.py", line 23, in <module>
    from kornia.augmentation._2d.base import RigidAffineAugmentationBase2D
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_2d\__init__.py", line 19, in <module>
    from kornia.augmentation._2d.intensity import *
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_2d\intensity\__init__.py", line 47, in <module>
    from kornia.augmentation._2d.intensity.plasma import RandomPlasmaBrightness, RandomPlasmaContrast, RandomPlasmaShadow
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_2d\intensity\plasma.py", line 22, in <module>
    from kornia.contrib import diamond_square
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\contrib\__init__.py", line 32, in <module>
    from .image_stitching import ImageStitcher
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\contrib\image_stitching.py", line 24, in <module>
    from kornia.feature import LocalFeatureMatcher, LoFTR
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\feature\__init__.py", line 24, in <module>
    from .integrated import (
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\feature\integrated.py", line 34, in <module>
    from .lightglue import LightGlue
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\feature\lightglue.py", line 48, in <module>
    from flash_attn.modules.mha import FlashCrossAttention
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\__init__.py", line 3, in <module>
    from flash_attn.flash_attn_interface import (
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_interface.py", line 13, in <module>
    from .flash_attn_triton_amd import interface_fa as flash_attn_gpu
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\interface_fa.py", line 3, in <module>
    from .fwd_prefill import attention_prefill_forward_triton_impl
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\fwd_prefill.py", line 272, in <module>
    autotune_configs, autotune_keys = get_autotune_configs()
                                      ^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\fwd_prefill.py", line 247, in get_autotune_configs
    if is_rdna():
       ^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\utils.py", line 272, in is_rdna
    return is_hip() and triton.runtime.driver.active.get_current_target().arch in ("gfx1030", "gfx1100", "gfx1101",
           ^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\utils.py", line 263, in is_hip
    return triton.runtime.driver.active.get_current_target().backend == "hip"
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\driver.py", line 30, in __getattr__
    return getattr(self._initialize_obj(), name)
                   ^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\driver.py", line 26, in _initialize_obj
    self._obj = self._init_fn()
                ^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\driver.py", line 12, in _create_driver
    return active_drivers[0]()
           ^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\driver.py", line 555, in __init__
    self.utils = HIPUtils()
                 ^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\driver.py", line 148, in __init__
    mod = compile_module_from_src(src=src, name="hip_utils", include_dirs=include_dirs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\build.py", line 167, in compile_module_from_src
    so = _build(name, src_path, tmpdir, library_dirs or [], include_dirs or [], libraries or [])
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\build.py", line 124, in _build
    subprocess.check_call(cc_cmd, stdout=subprocess.DEVNULL)
  File "C:\Users\Angel\AppData\Local\Programs\Python\Python311\Lib\subprocess.py", line 413, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['C:\\Program Files\\AMD\\ROCm\\6.2\\bin\\clang.exe', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmpnr1m3aw4\\hip_utils.c', '-O3', '-shared', '-Wno-psabi', '-march=native', '-mtune=native', '-o', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmpnr1m3aw4\\hip_utils.cp311-win_amd64.pyd', '-LG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\libs', '-LC:\\Program Files\\AMD\\ROCm\\6.2\\lib', '-IG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\Lib\\site-packages\\triton\\backends\\amd\\include', '-IC:\\Users\\Angel\\AppData\\Local\\Temp\\tmpnr1m3aw4', '-IC:\\Users\\Angel\\AppData\\Local\\Programs\\Python\\Python311\\Include']' returned non-zero exit status 1.

Cannot import G:\IMG Gen\ComfyUI-Zluda-T\comfy_extras\nodes_canny.py module for custom nodes: Command '['C:\\Program Files\\AMD\\ROCm\\6.2\\bin\\clang.exe', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmpnr1m3aw4\\hip_utils.c', '-O3', '-shared', '-Wno-psabi', '-march=native', '-mtune=native', '-o', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmpnr1m3aw4\\hip_utils.cp311-win_amd64.pyd', '-LG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\libs', '-LC:\\Program Files\\AMD\\ROCm\\6.2\\lib', '-IG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\Lib\\site-packages\\triton\\backends\\amd\\include', '-IC:\\Users\\Angel\\AppData\\Local\\Temp\\tmpnr1m3aw4', '-IC:\\Users\\Angel\\AppData\\Local\\Programs\\Python\\Python311\\Include']' returned non-zero exit status 1.
Using HIP SDK Clang.
%%% Triton is using the compiler 'C:\Program Files\AMD\ROCm\6.2\bin\clang.exe'
%%% [info] triton/runtime/build/_build: "C:\Program Files\AMD\ROCm\6.2\bin\clang.exe" C:\Users\Angel\AppData\Local\Temp\tmphberhxxk\hip_utils.c -O3 -shared -Wno-psabi -march=native -mtune=native -o C:\Users\Angel\AppData\Local\Temp\tmphberhxxk\hip_utils.cp311-win_amd64.pyd "-LG:\IMG Gen\ComfyUI-Zluda-T\venv\libs" "-LC:\Program Files\AMD\ROCm\6.2\lib" "-IG:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\include" -IC:\Users\Angel\AppData\Local\Temp\tmphberhxxk -IC:\Users\Angel\AppData\Local\Programs\Python\Python311\Include
In file included from C:\Users\Angel\AppData\Local\Temp\tmphberhxxk\hip_utils.c:8:
In file included from C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\Python.h:87:
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:184:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  184 |     struct timeval *tv,
      |            ^
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:190:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  190 |     struct timeval *tv,
      |            ^
2 warnings generated.
clang: error: linker command failed due to signal (use -v to see invocation)
Traceback (most recent call last):
  File "G:\IMG Gen\ComfyUI-Zluda-T\nodes.py", line 2129, in load_custom_node
    module_spec.loader.exec_module(module)
  File "<frozen importlib._bootstrap_external>", line 940, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "G:\IMG Gen\ComfyUI-Zluda-T\comfy_extras\nodes_morphology.py", line 4, in <module>
    from kornia.morphology import dilation, erosion, opening, closing, gradient, top_hat, bottom_hat
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\__init__.py", line 25, in <module>
    from . import (
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\__init__.py", line 19, in <module>
    from kornia.augmentation import auto, container
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\__init__.py", line 18, in <module>
    from .autoaugment import AutoAugment
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\autoaugment\__init__.py", line 18, in <module>
    from .autoaugment import AutoAugment
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\autoaugment\autoaugment.py", line 22, in <module>
    from kornia.augmentation.auto.base import SUBPOLICY_CONFIG, PolicyAugmentBase
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\base.py", line 23, in <module>
    from kornia.augmentation.auto.operations.policy import PolicySequential
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\operations\__init__.py", line 20, in <module>
    from .policy import PolicySequential
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\auto\operations\policy.py", line 24, in <module>
    from kornia.augmentation.container.base import ImageSequentialBase, TransformMatrixMinIn
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\container\__init__.py", line 18, in <module>
    from kornia.augmentation.container.augment import AugmentationSequential
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\container\augment.py", line 24, in <module>
    from kornia.augmentation._3d.base import AugmentationBase3D, RigidAffineAugmentationBase3D
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_3d\__init__.py", line 20, in <module>
    from kornia.augmentation._3d.mix import *
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_3d\mix\__init__.py", line 18, in <module>
    from kornia.augmentation._3d.mix.transplantation import RandomTransplantation3D
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_3d\mix\transplantation.py", line 18, in <module>
    from kornia.augmentation._2d.mix.transplantation import RandomTransplantation
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_2d\__init__.py", line 19, in <module>
    from kornia.augmentation._2d.intensity import *
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_2d\intensity\__init__.py", line 47, in <module>
    from kornia.augmentation._2d.intensity.plasma import RandomPlasmaBrightness, RandomPlasmaContrast, RandomPlasmaShadow
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\augmentation\_2d\intensity\plasma.py", line 22, in <module>
    from kornia.contrib import diamond_square
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\contrib\__init__.py", line 32, in <module>
    from .image_stitching import ImageStitcher
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\contrib\image_stitching.py", line 24, in <module>
    from kornia.feature import LocalFeatureMatcher, LoFTR
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\feature\__init__.py", line 24, in <module>
    from .integrated import (
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\feature\integrated.py", line 34, in <module>
    from .lightglue import LightGlue
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\kornia\feature\lightglue.py", line 48, in <module>
    from flash_attn.modules.mha import FlashCrossAttention
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\__init__.py", line 3, in <module>
    from flash_attn.flash_attn_interface import (
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_interface.py", line 13, in <module>
    from .flash_attn_triton_amd import interface_fa as flash_attn_gpu
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\interface_fa.py", line 3, in <module>
    from .fwd_prefill import attention_prefill_forward_triton_impl
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\fwd_prefill.py", line 272, in <module>
    autotune_configs, autotune_keys = get_autotune_configs()
                                      ^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\fwd_prefill.py", line 247, in get_autotune_configs
    if is_rdna():
       ^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\utils.py", line 272, in is_rdna
    return is_hip() and triton.runtime.driver.active.get_current_target().arch in ("gfx1030", "gfx1100", "gfx1101",
           ^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\flash_attn\flash_attn_triton_amd\utils.py", line 263, in is_hip
    return triton.runtime.driver.active.get_current_target().backend == "hip"
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\driver.py", line 30, in __getattr__
    return getattr(self._initialize_obj(), name)
                   ^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\driver.py", line 26, in _initialize_obj
    self._obj = self._init_fn()
                ^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\driver.py", line 12, in _create_driver
    return active_drivers[0]()
           ^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\driver.py", line 555, in __init__
    self.utils = HIPUtils()
                 ^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\driver.py", line 148, in __init__
    mod = compile_module_from_src(src=src, name="hip_utils", include_dirs=include_dirs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\build.py", line 167, in compile_module_from_src
    so = _build(name, src_path, tmpdir, library_dirs or [], include_dirs or [], libraries or [])
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\runtime\build.py", line 124, in _build
    subprocess.check_call(cc_cmd, stdout=subprocess.DEVNULL)
  File "C:\Users\Angel\AppData\Local\Programs\Python\Python311\Lib\subprocess.py", line 413, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['C:\\Program Files\\AMD\\ROCm\\6.2\\bin\\clang.exe', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmphberhxxk\\hip_utils.c', '-O3', '-shared', '-Wno-psabi', '-march=native', '-mtune=native', '-o', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmphberhxxk\\hip_utils.cp311-win_amd64.pyd', '-LG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\libs', '-LC:\\Program Files\\AMD\\ROCm\\6.2\\lib', '-IG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\Lib\\site-packages\\triton\\backends\\amd\\include', '-IC:\\Users\\Angel\\AppData\\Local\\Temp\\tmphberhxxk', '-IC:\\Users\\Angel\\AppData\\Local\\Programs\\Python\\Python311\\Include']' returned non-zero exit status 1.

Cannot import G:\IMG Gen\ComfyUI-Zluda-T\comfy_extras\nodes_morphology.py module for custom nodes: Command '['C:\\Program Files\\AMD\\ROCm\\6.2\\bin\\clang.exe', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmphberhxxk\\hip_utils.c', '-O3', '-shared', '-Wno-psabi', '-march=native', '-mtune=native', '-o', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmphberhxxk\\hip_utils.cp311-win_amd64.pyd', '-LG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\libs', '-LC:\\Program Files\\AMD\\ROCm\\6.2\\lib', '-IG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\Lib\\site-packages\\triton\\backends\\amd\\include', '-IC:\\Users\\Angel\\AppData\\Local\\Temp\\tmphberhxxk', '-IC:\\Users\\Angel\\AppData\\Local\\Programs\\Python\\Python311\\Include']' returned non-zero exit status 1.
### Loading: ComfyUI-Manager (V3.36)
[ComfyUI-Manager] network_mode: public
### ComfyUI Revision: 4882 [972495d9] | Released on '2025-08-24'

Import times for custom nodes:
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\cfz_vae_loader.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\cfz_cudnn.toggle.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\websocket_image_save.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\cfz_patcher.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\ComfyUI-deepcache
   1.1 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\ComfyUI-Manager

WARNING: some comfy_extras/ nodes did not import correctly. This may be because they are missing some dependencies.

IMPORT FAILED: nodes_canny.py
IMPORT FAILED: nodes_morphology.py

This issue might be caused by new missing dependencies added the last time you updated ComfyUI.
Please do a: pip install -r requirements.txt

[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/model-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/alter-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json
Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Starting server

To see the GUI go to: http://127.0.0.1:8188
FETCH ComfyRegistry Data: 5/95
FETCH ComfyRegistry Data: 10/95
FETCH ComfyRegistry Data: 15/95
FETCH ComfyRegistry Data: 20/95
FETCH ComfyRegistry Data: 25/95
FETCH ComfyRegistry Data: 30/95
FETCH ComfyRegistry Data: 35/95
FETCH ComfyRegistry Data: 40/95
FETCH ComfyRegistry Data: 45/95
FETCH ComfyRegistry Data: 50/95
FETCH ComfyRegistry Data: 55/95
FETCH ComfyRegistry Data: 60/95
FETCH ComfyRegistry Data: 65/95
FETCH ComfyRegistry Data: 70/95
FETCH ComfyRegistry Data: 75/95
FETCH ComfyRegistry Data: 80/95
FETCH ComfyRegistry Data: 85/95
FETCH ComfyRegistry Data: 90/95
FETCH ComfyRegistry Data: 95/95
FETCH ComfyRegistry Data [DONE]
[ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE]
[ComfyUI-Manager] All startup tasks have been completed.

Logs

Other

No response

[Avalre]

Not saying this will solve everything, but to get a start... follow my instructions from the other issue #259 , which I will paste again because that issue is a headache to read at this point. Once you've done this, paste your console output running the updated comfyui-n.bat and I or someone more knowledgeable can help from there.

copy & rename your comfyui-n.bat, then add this line below @echo off in the copied comfyui-n.bat and see if it helps: set TRITON_OVERRIDE_ARCH=gfx1010

(below is assuming you are using vs build tools and not the full install)
OPTIONAL: If you are encountering clang errors as I was (or are otherwise desperate), you can automate launching in the vs build tools developer command prompt (as per official troubleshooting suggestion) by adding this DIRECTLY below @echo off in the copied comfyui-n.bat, moving anything else (including set TRITON_OVERRIDE_ARCH=gfx1010) down:
call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
Doing so fixed the clang errors for me.

Ideally, this should resolve any issues with Triton passing its tests.

As for (some of) the other errors:

• navigate to your ComfyUI-Zluda directory in file explorer
• copy the full path to your ComfyUI-Zluda directory from the navigation bar
• open a command prompt or powershell window (no need to run as administrator)
• type cd , then paste your ComfyUI-Zluda directory path and hit enter
  • for my install location, the command looks like: cd C:\ComfyUI-Zluda
• copy and paste venv\scripts\activate and hit enter
• copy and paste pip uninstall flash_attn and hit enter
• close command prompt / powershell window; done 👍🏻

[Ashir555]

Not saying this will solve everything, but to get a start... follow my instructions from the other issue #259 , which I will paste again because that issue is a headache to read at this point.

Yeah, I know, that's why I choose to ask for suppport apart...

copy & rename your comfyui-n.bat, then add this line below @echo off in the copied comfyui-n.bat and see if it helps: set TRITON_OVERRIDE_ARCH=gfx1010

Same output log... If you notice, in the log, after it detects the GPU, it auto sets the override.

(assuming you are using vs build tools and not the full install) OPTIONAL: If you are encountering clang errors as I was (or are otherwise desperate), you can automate launching in the vs build tools developer command prompt (as per official troubleshooting suggestion) by adding this

By adding what??? set TRITON... it doesn't make a difference if is right after the @echo off or added later.

* copy and paste `pip uninstall flash_attn` and hit enter

That reduce the errors, but triton still doesn't pass the verification.

*** Checking and updating to new version if possible
Already up to date.

[START] Security scan
[DONE] Security scan
## ComfyUI-Manager: installing dependencies done.
** ComfyUI startup time: 2025-08-25 02:23:49.653
** Platform: Windows
** Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
** Python executable: G:\IMG Gen\ComfyUI-Zluda-T\venv\Scripts\python.exe
** ComfyUI Path: G:\IMG Gen\ComfyUI-Zluda-T
** ComfyUI Base Folder Path: G:\IMG Gen\ComfyUI-Zluda-T
** User directory: G:\IMG Gen\ComfyUI-Zluda-T\user
** ComfyUI-Manager config path: G:\IMG Gen\ComfyUI-Zluda-T\user\default\ComfyUI-Manager\config.ini
** Log path: G:\IMG Gen\ComfyUI-Zluda-T\user\comfyui.log

Prestartup times for custom nodes:
   4.2 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\ComfyUI-Manager

Checkpoint files will always be loaded safely.

  ::  Checking package versions...
Found pydantic: 2.11.7, pydantic-settings: 2.10.1
  ::  Pydantic packages are compatible, skipping reinstall
Installed version of comfyui-frontend-package: 1.25.10
Installed version of comfyui-workflow-templates: 0.1.65
Installed version of av: 15.0.0
Installed version of comfyui-embedded-docs: 0.2.6
  ::  Package version check complete.

  ::  ------------------------ ZLUDA -----------------------  ::
  ::  Auto-detecting AMD GPU architecture for Triton...
  ::  Detected GPU via Windows registry: AMD Radeon RX 5700 XT
  ::  Set TRITON_OVERRIDE_ARCH=gfx1010
  ::  Triton core imported successfully
  ::  Running Triton kernel test...
%%% [info] triton/runtime/build/platform_key: AMD64,Windows,64bit,WindowsPE
Using HIP SDK Clang.
%%% Triton is using the compiler 'C:\Program Files\AMD\ROCm\6.2\bin\clang.exe'
%%% [info] triton/runtime/build/_build: "C:\Program Files\AMD\ROCm\6.2\bin\clang.exe" C:\Users\Angel\AppData\Local\Temp\tmphk2j6no_\hip_utils.c -O3 -shared -Wno-psabi -march=native -mtune=native -o C:\Users\Angel\AppData\Local\Temp\tmphk2j6no_\hip_utils.cp311-win_amd64.pyd "-LG:\IMG Gen\ComfyUI-Zluda-T\venv\libs" "-LC:\Program Files\AMD\ROCm\6.2\lib" "-IG:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\triton\backends\amd\include" -IC:\Users\Angel\AppData\Local\Temp\tmphk2j6no_ -IC:\Users\Angel\AppData\Local\Programs\Python\Python311\Include
In file included from C:\Users\Angel\AppData\Local\Temp\tmphk2j6no_\hip_utils.c:8:
In file included from C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\Python.h:87:
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:184:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  184 |     struct timeval *tv,
      |            ^
C:\Users\Angel\AppData\Local\Programs\Python\Python311\Include\cpython/pytime.h:190:12: warning: declaration of
      'struct timeval' will not be visible outside of this function [-Wvisibility]
  190 |     struct timeval *tv,
      |            ^
2 warnings generated.
clang: error: linker command failed due to signal (use -v to see invocation)
  ::  Triton test failed: Command '['C:\\Program Files\\AMD\\ROCm\\6.2\\bin\\clang.exe', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmphk2j6no_\\hip_utils.c', '-O3', '-shared', '-Wno-psabi', '-march=native', '-mtune=native', '-o', 'C:\\Users\\Angel\\AppData\\Local\\Temp\\tmphk2j6no_\\hip_utils.cp311-win_amd64.pyd', '-LG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\libs', '-LC:\\Program Files\\AMD\\ROCm\\6.2\\lib', '-IG:\\IMG Gen\\ComfyUI-Zluda-T\\venv\\Lib\\site-packages\\triton\\backends\\amd\\include', '-IC:\\Users\\Angel\\AppData\\Local\\Temp\\tmphk2j6no_', '-IC:\\Users\\Angel\\AppData\\Local\\Programs\\Python\\Python311\\Include']' returned non-zero exit status 1.
  ::  Triton available but failed verification
  ::  Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
  ::  Using ZLUDA with device: AMD Radeon RX 5700 XT [ZLUDA]
  ::  Applying core ZLUDA patches...
  ::  Triton optimizations skipped (not available)
  ::  Configuring PyTorch backends...
  ::  Disabled CUDA flash attention
  ::  Enabled math attention fallback
  ::  ZLUDA initialization complete
  ::  ------------------------ ZLUDA -----------------------  ::

Total VRAM 8176 MB, total RAM 16294 MB
pytorch version: 2.7.0+cu118
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 5700 XT [ZLUDA] : native
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr  2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.3.52
ComfyUI frontend version: 1.25.10
[Prompt Server] web root: G:\IMG Gen\ComfyUI-Zluda-T\venv\Lib\site-packages\comfyui_frontend_package\static
### Loading: ComfyUI-Manager (V3.36)
[ComfyUI-Manager] network_mode: public
### ComfyUI Revision: 4882 [972495d9] | Released on '2025-08-24'

Import times for custom nodes:
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\cfz_vae_loader.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\cfz_cudnn.toggle.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\cfz_patcher.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\websocket_image_save.py
   0.0 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\ComfyUI-deepcache
   1.1 seconds: G:\IMG Gen\ComfyUI-Zluda-T\custom_nodes\ComfyUI-Manager

[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/alter-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/model-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json
Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Starting server

To see the GUI go to: http://127.0.0.1:8188
FETCH ComfyRegistry Data: 5/95
FETCH ComfyRegistry Data: 10/95
FETCH ComfyRegistry Data: 15/95
FETCH ComfyRegistry Data: 20/95
FETCH ComfyRegistry Data: 25/95
FETCH ComfyRegistry Data: 30/95
FETCH ComfyRegistry Data: 35/95
FETCH ComfyRegistry Data: 40/95
FETCH ComfyRegistry Data: 45/95
FETCH ComfyRegistry Data: 50/95
FETCH ComfyRegistry Data: 55/95
FETCH ComfyRegistry Data: 60/95
FETCH ComfyRegistry Data: 65/95
FETCH ComfyRegistry Data: 70/95
FETCH ComfyRegistry Data: 75/95
FETCH ComfyRegistry Data: 80/95
FETCH ComfyRegistry Data: 85/95
FETCH ComfyRegistry Data: 90/95
FETCH ComfyRegistry Data: 95/95
FETCH ComfyRegistry Data [DONE]
[ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE]
[ComfyUI-Manager] All startup tasks have been completed.

[Avalre]

By adding what??? set TRITON... it doesn't make a difference if is right after the @echo off or added later.

oops 😭 at some point during editing that comment the 16 times I did, I must have accidentally removed the line I wanted people to add. I updated that old comment + my first one here with the new line to add, but here it is again:
call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"

It really shouldn't be necessary to add this line; as patientx mentioned, most people shouldn't need to do it. I only included it in case you end up experiencing the same weird clang errors I was. That said, I don't imagine it hurts anything?

Same output log... If you notice, in the log, after it detects the GPU, it auto sets the override.

Yes, it auto sets the variable, but you'll still get errors that go away if you set it in the .bat file. Or, at least, that's what happened for me and several others. Look at how my output differs from yours:

  ::  ------------------------ ZLUDA -----------------------  ::
  ::  TRITON_OVERRIDE_ARCH already set to: gfx1031
  ::  Triton core imported successfully
  ::  Running Triton kernel test...
%%% [info] triton/runtime/build/platform_key: AMD64,Windows,64bit,WindowsPE
  ::  Triton kernel test passed successfully
  ::  Triton initialized successfully
  ::  Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
  ::  Using ZLUDA with device: AMD Radeon RX 6700 XT [ZLUDA]
  ::  Applying core ZLUDA patches...
  ::  Initializing Triton optimizations
  ::  Configuring Triton device properties...
  ::  Triton device properties configured
  ::  Flash attention components found
  ::  AMD flash attention enabled successfully
  ::  Configuring PyTorch backends...
  ::  Disabled CUDA flash attention
  ::  Enabled math attention fallback
  ::  ZLUDA initialization complete
  ::  ------------------------ ZLUDA -----------------------  ::

:: TRITON_OVERRIDE_ARCH already set to: gfx1031 tells me that I set the variable in the .bat, whereas
:: Set TRITON_OVERRIDE_ARCH=gfx1010 tells me the variable isn't set in the .bat - unless there's some discrepancy I'm unaware of that results in your manually-set TRITON_OVERRIDE_ARCH variable being ignored?

[EDIT]
Here's my comfyui-n.bat, just to make absolutely sure everything is clear:

@echo off

call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"

set FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE
set FLASH_ATTENTION_TRITON_AMD_AUTOTUNE=TRUE
set TRITON_OVERRIDE_ARCH=gfx1031

set MIOPEN_FIND_MODE=2
set MIOPEN_LOG_LEVEL=3

set PYTHON="%~dp0/venv/Scripts/python.exe"
set GIT=
set VENV_DIR=./venv

set COMMANDLINE_ARGS=--auto-launch --use-sage-attention --reserve-vram 0.9

set ZLUDA_COMGR_LOG_LEVEL=1

echo *** Checking and updating to new version if possible 

copy comfy\customzluda\zluda-default.py comfy\zluda.py /y >NUL
git pull
copy comfy\customzluda\zluda.py comfy\zluda.py /y >NUL

echo.
.\zluda\zluda.exe -- %PYTHON% main.py %COMMANDLINE_ARGS%
pause

beyond changing the gfx to match yours, you should(?) be able to just copy/paste that into a separate .bat file in the comfyui-zluda folder and run it. Only assumption made is that you have vs build tools installed (to the same directory, check this), but that's a listed dependency for ComfyUI anyways.

'struct timeval' will not be visible outside of this function [-Wvisibility]
| struct timeval *tv,
| ^

I believe this error is a separate (minor) issue that can either be safely ignored entirely or dealt with by following sfinktah's instructions here: python/cpython#82863 (comment) (thanks sfinktah)

[Ashir555]

oops 😭 at some point during editing that comment the 16 times I did, I must have accidentally removed the line I wanted people to add. I updated that old comment + my first one here with the new line to add, but here it is again: call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"

Added, no difference.

:: TRITON_OVERRIDE_ARCH already set to: gfx1031 tells me that I set the variable in the .bat, whereas :: Set TRITON_OVERRIDE_ARCH=gfx1010 tells me the variable isn't set in the .bat - unless there's some discrepancy I'm unaware of that results in your manually-set TRITON_OVERRIDE_ARCH variable being ignored?

Added, no difference.

[EDIT] Here's my comfyui-n.bat, just to make absolutely sure everything is clear:

Used, change only the GFX to 1010, no difference.

I believe this error is a separate (minor) issue that can either be safely ignored entirely or dealt with by following sfinktah's instructions here: python/cpython#82863 (comment) (thanks sfinktah)

clang: error: linker command failed due to signal (use -v to see invocation)

I think that is the source of the problem, I spend 2 hours back and forward between ChatGPT and Deepseek to no avail.

[Avalre]

oops 😭 at some point during editing that comment the 16 times I did, I must have accidentally removed the line I wanted people to add. I updated that old comment + my first one here with the new line to add, but here it is again: call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"

Added, no difference.

Wasn't expecting it to in terms of errors, but are you at least getting the

**********************************************************************
** Visual Studio 2022 Developer Command Prompt v17.14.13
** Copyright (c) 2025 Microsoft Corporation
**********************************************************************
[vcvarsall.bat] Environment initialized for: 'x64'

bit at the top of your execution now? That's completely unrelated to anything we're doing; just an environment for everything we do to run in. Your graphics card model is completely irrelevant to this bit. So, at the very least, if you are doing everything correctly, this should pop up before anything else in the console. If it isn't, something is very wrong. My first question would be if you are running the same .bat file that you are applying the changes to.

:: TRITON_OVERRIDE_ARCH already set to: gfx1031 tells me that I set the variable in the .bat, whereas :: Set TRITON_OVERRIDE_ARCH=gfx1010 tells me the variable isn't set in the .bat - unless there's some discrepancy I'm unaware of that results in your manually-set TRITON_OVERRIDE_ARCH variable being ignored?

Kinda similar to above, would have really expected to see a change in output here. But I'm not referencing the code in saying that, just my own experience. I couldn't tell you what's going wrong without spending more time than I am willing to spend (college resumes today) - hopefully someone more knowledgeable can help out.

clang: error: linker command failed due to signal (use -v to see invocation)

I don't remember seeing this error specifically, but given that it's clang... I would want to verify that you're running everything in the vs developer command prompt (ie, the line I previously forgot to include).

If, for some reason that line just won't work for you, you can try searching for "developer command prompt for vs 2022" in your start menu. If installed correctly, an application with a console icon should show up. From there, cd to wherever your custom copy of comfyui-n.bat is. Then, just type comfyui-n.bat (replace with your custom version's file name) to run comfyui. For me, I run:
cd C:\ComfyUI-Zluda
popsoda.bat

If you're still getting the same error after running everything in the actual application, I'm not immediately sure how to help. My favorite catch phrase: hopefully someone more knowledgeable can help out

(my responses from now on will be much less frequent, as college is resuming and I take classes very seriously)

[Ashir555]

Wasn't expecting it to in terms of errors, but are you at least getting the

**********************************************************************
** Visual Studio 2022 Developer Command Prompt v17.14.13
** Copyright (c) 2025 Microsoft Corporation
**********************************************************************
[vcvarsall.bat] Environment initialized for: 'x64'

Yes, that pops right at the beginning.

If, for some reason that line just won't work for you, you can try searching for "developer command prompt for vs 2022" in your start menu. If installed correctly, an application with a console icon should show up. From there, cd to wherever your custom copy of comfyui-n.bat is. Then, just type comfyui-n.bat (replace with your custom version's file name) to run comfyui. For me, I run: cd C:\ComfyUI-Zluda popsoda.bat

I didn't knew I had that. I try it, same error.

If you're still getting the same error after running everything in the actual application, I'm not immediately sure how to help. My favorite catch phrase: hopefully someone more knowledgeable can help out

One can only hope...

(my responses from now on will be much less frequent, as college is resuming and I take classes very seriously)

I understand, thank you for the efford. Best of luck.

[zereshkealt]

By adding what??? set TRITON... it doesn't make a difference if is right after the @echo off or added later.

oops 😭 at some point during editing that comment the 16 times I did, I must have accidentally removed the line I wanted people to add. I updated that old comment + my first one here with the new line to add, but here it is again: call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"

It really shouldn't be necessary to add this line; as patientx mentioned, most people shouldn't need to do it. I only included it in case you end up experiencing the same weird clang errors I was. That said, I don't imagine it hurts anything?

Same output log... If you notice, in the log, after it detects the GPU, it auto sets the override.

Yes, it auto sets the variable, but you'll still get errors that go away if you set it in the .bat file. Or, at least, that's what happened for me and several others. Look at how my output differs from yours:

  ::  ------------------------ ZLUDA -----------------------  ::
  ::  TRITON_OVERRIDE_ARCH already set to: gfx1031
  ::  Triton core imported successfully
  ::  Running Triton kernel test...
%%% [info] triton/runtime/build/platform_key: AMD64,Windows,64bit,WindowsPE
  ::  Triton kernel test passed successfully
  ::  Triton initialized successfully
  ::  Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
  ::  Using ZLUDA with device: AMD Radeon RX 6700 XT [ZLUDA]
  ::  Applying core ZLUDA patches...
  ::  Initializing Triton optimizations
  ::  Configuring Triton device properties...
  ::  Triton device properties configured
  ::  Flash attention components found
  ::  AMD flash attention enabled successfully
  ::  Configuring PyTorch backends...
  ::  Disabled CUDA flash attention
  ::  Enabled math attention fallback
  ::  ZLUDA initialization complete
  ::  ------------------------ ZLUDA -----------------------  ::

:: TRITON_OVERRIDE_ARCH already set to: gfx1031 tells me that I set the variable in the .bat, whereas :: Set TRITON_OVERRIDE_ARCH=gfx1010 tells me the variable isn't set in the .bat - unless there's some discrepancy I'm unaware of that results in your manually-set TRITON_OVERRIDE_ARCH variable being ignored?

[EDIT] Here's my comfyui-n.bat, just to make absolutely sure everything is clear:

@echo off

call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"

set FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE
set FLASH_ATTENTION_TRITON_AMD_AUTOTUNE=TRUE
set TRITON_OVERRIDE_ARCH=gfx1031

set MIOPEN_FIND_MODE=2
set MIOPEN_LOG_LEVEL=3

set PYTHON="%~dp0/venv/Scripts/python.exe"
set GIT=
set VENV_DIR=./venv

set COMMANDLINE_ARGS=--auto-launch --use-sage-attention --reserve-vram 0.9

set ZLUDA_COMGR_LOG_LEVEL=1

echo *** Checking and updating to new version if possible 

copy comfy\customzluda\zluda-default.py comfy\zluda.py /y >NUL
git pull
copy comfy\customzluda\zluda.py comfy\zluda.py /y >NUL

echo.
.\zluda\zluda.exe -- %PYTHON% main.py %COMMANDLINE_ARGS%
pause

beyond changing the gfx to match yours, you should(?) be able to just copy/paste that into a separate .bat file in the comfyui-zluda folder and run it. Only assumption made is that you have vs build tools installed (to the same directory, check this), but that's a listed dependency for ComfyUI anyways.

'struct timeval' will not be visible outside of this function [-Wvisibility]
| struct timeval *tv,
| ^

I believe this error is a separate (minor) issue that can either be safely ignored entirely or dealt with by following sfinktah's instructions here: python/cpython#82863 (comment) (thanks sfinktah)

the result for using the same bat file content as urs:

---

** Visual Studio 2022 Developer Command Prompt v17.14.13
** Copyright (c) 2025 Microsoft Corporation

---

[vcvarsall.bat] Environment initialized for: 'x64'
*** Checking and updating to new version if possible
remote: Enumerating objects: 15, done.
remote: Counting objects: 100% (15/15), done.
remote: Compressing objects: 100% (12/12), done.
remote: Total 12 (delta 8), reused 0 (delta 0), pack-reused 0 (from 0)
Unpacking objects: 100% (12/12), 3.92 KiB | 9.00 KiB/s, done.
From https://github.com/patientx/ComfyUI-Zluda
e0a2d8d..cc9bf76 master -> origin/master
Updating e0a2d8d..cc9bf76
Fast-forward
install-for-older-amd.bat | 6 +++++-
install-legacy.bat | 6 +++++-
2 files changed, 10 insertions(+), 2 deletions(-)

Adding extra search path checkpoints G:\SD.Next\sdnext\models\Stable-diffusion
Adding extra search path configs G:\SD.Next\sdnext\models\Stable-diffusion
Adding extra search path vae G:\SD.Next\sdnext\models\VAE
Adding extra search path loras G:\SD.Next\sdnext\models\Lora
Adding extra search path loras G:\SD.Next\sdnext\models\LyCORIS
Adding extra search path upscale_models G:\SD.Next\sdnext\models\ESRGAN
Adding extra search path upscale_models G:\SD.Next\sdnext\models\RealESRGAN
Adding extra search path upscale_models G:\SD.Next\sdnext\models\SwinIR
Adding extra search path embeddings G:\SD.Next\sdnext\embeddings
Adding extra search path hypernetworks G:\SD.Next\sdnext\models\hypernetworks
Adding extra search path controlnet G:\SD.Next\sdnext\models\ControlNet
[START] Security scan
[DONE] Security scan

ComfyUI-Manager: installing dependencies done.

** ComfyUI startup time: 2025-08-25 17:13:39.859
** Platform: Windows
** Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
** Python executable: G:\ComfyUI\ComfyUI-Zluda\venv\Scripts\python.exe
** ComfyUI Path: G:\ComfyUI\ComfyUI-Zluda
** ComfyUI Base Folder Path: G:\ComfyUI\ComfyUI-Zluda
** User directory: G:\ComfyUI\ComfyUI-Zluda\user
** ComfyUI-Manager config path: G:\ComfyUI\ComfyUI-Zluda\user\default\ComfyUI-Manager\config.ini
** Log path: G:\ComfyUI\ComfyUI-Zluda\user\comfyui.log

Prestartup times for custom nodes:
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Easy-Use
2.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager

Checkpoint files will always be loaded safely.

:: Checking package versions...
Found pydantic: 2.11.7, pydantic-settings: 2.10.1
:: Pydantic packages are compatible, skipping reinstall
Installed version of comfyui-frontend-package: 1.25.10
Installed version of comfyui-workflow-templates: 0.1.65
Installed version of av: 15.0.0
Installed version of comfyui-embedded-docs: 0.2.6
:: Package version check complete.

:: ------------------------ ZLUDA ----------------------- ::
:: TRITON_OVERRIDE_ARCH already set to: gfx1031
:: Triton core imported successfully
:: Enabled cuDNN
:: Running Triton kernel test...
%%% [info] triton/runtime/build/platform_key: AMD64,Windows,64bit,WindowsPE
:: Triton test failed: Triton Error [HIP]: Code: 209, Messsage: no kernel image is available for execution on the device
:: Triton available but failed verification
:: Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
:: Using ZLUDA with device: AMD Radeon RX 5700 XT [ZLUDA]
:: Applying core ZLUDA patches...
:: Triton optimizations skipped (not available)
:: Configuring PyTorch backends...
:: Disabled CUDA flash attention
:: Enabled math attention fallback
:: ZLUDA initialization complete
:: ------------------------ ZLUDA ----------------------- ::

Total VRAM 8176 MB, total RAM 16297 MB
pytorch version: 2.7.0+cu118
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 5700 XT [ZLUDA] : native
Using sage attention
Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.3.52
ComfyUI frontend version: 1.25.10
[Prompt Server] web root: G:\ComfyUI\ComfyUI-Zluda\venv\Lib\site-packages\comfyui_frontend_package\static
[ComfyUI-Easy-Use] server: v1.3.3 Loaded
[ComfyUI-Easy-Use] web root: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Easy-Use\web_version/v2 Loaded
ComfyUI-GGUF: Partial torch compile only, consider updating pytorch

Loading: ComfyUI-Impact-Pack (V8.22.2)

[Impact Pack] Wildcards loading done.

Loading: ComfyUI-Manager (V3.36)

[ComfyUI-Manager] network_mode: public

ComfyUI Revision: 4898 [cc9bf76] | Released on '2025-08-25'

[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/alter-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/model-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
FantasyPortrait nodes not available due to error in importing them: No module named 'onnx'

Import times for custom nodes:
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\cfz_vae_loader.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\websocket_image_save.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\cfz_cudnn.toggle.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\cfz_patcher.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-deepcache
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Custom-Scripts
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Frame-Interpolation
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-KJNodes
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-GGUF
0.1 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Impact-Pack
0.2 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-VideoHelperSuite
0.2 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Florence2
0.7 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager
1.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-WanVideoWrapper
1.1 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Easy-Use

Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Starting server

To see the GUI go to: http://127.0.0.1:8188
[ComfyUI-Manager] An error occurred while fetching 'https://api.comfy.org/nodes?page=1&limit=30&comfyui_version=unknown&form_factor=git-windows': Expecting value: line 2 column 1 (char 1)
Cannot connect to comfyregistry.
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE]
[ComfyUI-Manager] All startup tasks have been completed.

but deepseek told me to built it this way:

@echo off

set TRITON_OVERRIDE_ARCH=gfx1010
set FLASH_ATTENTION_TRITON_AMD_ENABLE=FALSE
set FLASH_ATTENTION_TRITON_AMD_AUTOTUNE=TRUE

set MIOPEN_FIND_MODE=2
set MIOPEN_LOG_LEVEL=3

set PYTHON="%~dp0/venv/Scripts/python.exe"
set GIT=
set VENV_DIR=./venv

set COMMANDLINE_ARGS=--auto-launch --use-quad-cross-attention --reserve-vram 0.9

set ZLUDA_COMGR_LOG_LEVEL=1

echo *** Checking and updating to new version if possible

copy comfy\customzluda\zluda-default.py comfy\zluda.py /y >NUL
git pull
copy comfy\customzluda\zluda.py comfy\zluda.py /y >NUL

echo.
call "G:\AI\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
.\zluda\zluda.exe -- %PYTHON% main.py %COMMANDLINE_ARGS%
pause

because of the echo command nearly at the end starting a new thing(i don't remember what was the reason and i'm not knowledgeable) and the result is the same:

*** Checking and updating to new version if possible
Already up to date.

---

** Visual Studio 2022 Developer Command Prompt v17.14.13
** Copyright (c) 2025 Microsoft Corporation

---

[vcvarsall.bat] Environment initialized for: 'x64'
Adding extra search path checkpoints G:\SD.Next\sdnext\models\Stable-diffusion
Adding extra search path configs G:\SD.Next\sdnext\models\Stable-diffusion
Adding extra search path vae G:\SD.Next\sdnext\models\VAE
Adding extra search path loras G:\SD.Next\sdnext\models\Lora
Adding extra search path loras G:\SD.Next\sdnext\models\LyCORIS
Adding extra search path upscale_models G:\SD.Next\sdnext\models\ESRGAN
Adding extra search path upscale_models G:\SD.Next\sdnext\models\RealESRGAN
Adding extra search path upscale_models G:\SD.Next\sdnext\models\SwinIR
Adding extra search path embeddings G:\SD.Next\sdnext\embeddings
Adding extra search path hypernetworks G:\SD.Next\sdnext\models\hypernetworks
Adding extra search path controlnet G:\SD.Next\sdnext\models\ControlNet
[START] Security scan
[DONE] Security scan

ComfyUI-Manager: installing dependencies done.

** ComfyUI startup time: 2025-08-25 17:17:32.903
** Platform: Windows
** Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
** Python executable: G:\ComfyUI\ComfyUI-Zluda\venv\Scripts\python.exe
** ComfyUI Path: G:\ComfyUI\ComfyUI-Zluda
** ComfyUI Base Folder Path: G:\ComfyUI\ComfyUI-Zluda
** User directory: G:\ComfyUI\ComfyUI-Zluda\user
** ComfyUI-Manager config path: G:\ComfyUI\ComfyUI-Zluda\user\default\ComfyUI-Manager\config.ini
** Log path: G:\ComfyUI\ComfyUI-Zluda\user\comfyui.log

Prestartup times for custom nodes:
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Easy-Use
2.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager

Checkpoint files will always be loaded safely.

:: Checking package versions...
Found pydantic: 2.11.7, pydantic-settings: 2.10.1
:: Pydantic packages are compatible, skipping reinstall
Installed version of comfyui-frontend-package: 1.25.10
Installed version of comfyui-workflow-templates: 0.1.65
Installed version of av: 15.0.0
Installed version of comfyui-embedded-docs: 0.2.6
:: Package version check complete.

:: ------------------------ ZLUDA ----------------------- ::
:: TRITON_OVERRIDE_ARCH already set to: gfx1010
:: Triton core imported successfully
:: Enabled cuDNN
:: Running Triton kernel test...
%%% [info] triton/runtime/build/platform_key: AMD64,Windows,64bit,WindowsPE
:: Triton kernel test failed (incorrect output)
:: Triton available but failed verification
:: Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
:: Using ZLUDA with device: AMD Radeon RX 5700 XT [ZLUDA]
:: Applying core ZLUDA patches...
:: Triton optimizations skipped (not available)
:: Configuring PyTorch backends...
:: Disabled CUDA flash attention
:: Enabled math attention fallback
:: ZLUDA initialization complete
:: ------------------------ ZLUDA ----------------------- ::

Total VRAM 8176 MB, total RAM 16297 MB
pytorch version: 2.7.0+cu118
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 5700 XT [ZLUDA] : native
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.3.52
ComfyUI frontend version: 1.25.10
[Prompt Server] web root: G:\ComfyUI\ComfyUI-Zluda\venv\Lib\site-packages\comfyui_frontend_package\static
[ComfyUI-Easy-Use] server: v1.3.3 Loaded
[ComfyUI-Easy-Use] web root: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Easy-Use\web_version/v2 Loaded
ComfyUI-GGUF: Partial torch compile only, consider updating pytorch

Loading: ComfyUI-Impact-Pack (V8.22.2)

[Impact Pack] Wildcards loading done.

Loading: ComfyUI-Manager (V3.36)

[ComfyUI-Manager] network_mode: public

ComfyUI Revision: 4898 [cc9bf76] | Released on '2025-08-25'

[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/alter-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/model-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json
FantasyPortrait nodes not available due to error in importing them: No module named 'onnx'

Import times for custom nodes:
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\cfz_vae_loader.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\cfz_cudnn.toggle.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\websocket_image_save.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\cfz_patcher.py
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-deepcache
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Custom-Scripts
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Frame-Interpolation
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-GGUF
0.0 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-KJNodes
0.1 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Impact-Pack
0.2 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-VideoHelperSuite
0.2 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Florence2
0.7 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager
0.9 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-WanVideoWrapper
1.1 seconds: G:\ComfyUI\ComfyUI-Zluda\custom_nodes\ComfyUI-Easy-Use

Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Starting server

To see the GUI go to: http://127.0.0.1:8188

[zereshkealt]

@Ashir555
tried to ask deepseek for this problem with triton test failing, it gave a couple of solutions, one was to run this command before running the comfyui:
"set TRITON_INTERPRET=1" in the cmd, so the comfyui always runs with sage attention, no matter the the test result failing or not.
the other one was to find the code for the kernel test in the files and set the return line for it to "True" so it always runs with triton enabled!(basically the same as the first one)
the last which i'm interested in, is to downgrade the triton itself! i don't know which version of triton we are running, i did a bit of digging and realize there are other versions for different generations of nvidia gpus, so it might work with older generations for our older gpus.

[Ashir555]

set TRITON_INTERPRET=1

This just gives me an error and ComfyUI-Zluda didn't even start.

the last which i'm interested in, is to downgrade the triton itself! i don't know which version of triton we are running, i did a bit of digging and realize there are other versions for different generations of nvidia gpus, so it might work with older generations for our older gpus

If you use install-n you it will install triton-3.4.0+gita9c80202 but for Python 3.11. You can check this by going into \venv\scripts\ and running pip show triton, but you should not change it because it's patched for torch 2.7.0.

Remember what I said before: you shouldn't customize your ComfyUI-Zluda installation, if you do, the more you do it the more prone for errors you will be.

[zereshkealt]

set TRITON_INTERPRET=1

This just gives me an error and ComfyUI-Zluda didn't even start.

the last which i'm interested in, is to downgrade the triton itself! i don't know which version of triton we are running, i did a bit of digging and realize there are other versions for different generations of nvidia gpus, so it might work with older generations for our older gpus

If you use install-n you it will install triton-3.4.0+gita9c80202 but for Python 3.11. You can check this by going into \venv\scripts\ and running pip show triton, but you should not change it because it's patched for torch 2.7.0.

Remember what I said before: you shouldn't customize your ComfyUI-Zluda installation, if you do, the more you do it the more prone for errors you will be.

correct, i just wanted to tell u, this thing, doesn't have a solution yet. at least not for us.
BTW, u can still put this line "set TRITON_INTERPRET=1" in the comfyui-n.bat to force to run each time and there's this folder ".triton" in the "user/your account name/" location in which i think u should remove before trying to test anything. it resets the triton test.

[Ultracelestial]

Add "set TRITON_OVERRIDE_ARCH=gfx1010" in comfyui-n.bat
gfx1100 if you have 7900XT or XTX, gfx1201 for 9070XT.
For 5700XT it must be gfx1010.

[Ashir555]

Add "set TRITON_OVERRIDE_ARCH=gfx1010" in comfyui-n.bat gfx1100 if you have 7900XT or XTX, gfx1201 for 9070XT. For 5700XT it must be gfx1010.

Already try that, didn't work.

[Rando717]

@Ashir555
Trial and error is the key when it comes to old stuff like RX5000 series.
Just you wait once we switch to rocm 6.4.2, new wave of issues likely incoming for us.

This might not be helpful but I gonna share my experience,
I do remember having issues with triton 3.4.0 in past with compiling or on every first generation (after stopping and hitting run again it worked), so I am still rocking old v3.0.0 because of it.
Perhaps worth experimenting if you up for it.

Maybe worth noting I am using gfx1010xnack- version with HIP SDK extension applied over it, and adrenalin 25.8.1 (driver only).

Anyway, make sure VS BuildTools are actually installed.

• clear cache (C:\Users\<username>\.triton) or run cache-clean.bat if you wanna wipe all cache (zluda, triton, miopen...)
• double check VS BuildTools are installed, open PowerShell or cmd type cl ,it should print (depending on version):
  Microsoft (R) C/C++ Optimizing Compiler Version 19.44.35214 for x64
  Copyright (C) Microsoft Corporation. All rights reserved.
• check environment variables for PATH (C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\<your version>\bin\Hostx64\x64)
  EDIT: this is assuming default install path, replace <your version> with correct folder
• cmd pip cache purge doesn't hurt if you wanna start fresh and force package re-download

PS. install-n.bat is like pandora's box without much output info of what is actually going on in background.
After comparing files I noticed sometimes it fails Patch DLLs without notice, so I had to patch (move and rename) files manually.

Attached my messy .bat if you wanna try it, works on my RX5700 (non-XT). sidenote: I do keep triton cache inside comfy folder.

comfyui-n - CUSTOM.zip

Log (EasyCache go brrr)

*** Checking and updating to new version if possible
Already up to date.

** Checking ZLUDA version
ZLUDA 3.9.5

** Checking installed Triton/FlashAttention/SageAttention packages
Name: triton
Version: 3.0.0+git2b248f14
Name: flash_attn
Version: 2.7.4.post1
Name: sageattention
Version: 1.0.6

[START] Security scan
[DONE] Security scan

ComfyUI-Manager: installing dependencies done.

** ComfyUI startup time: 2025-08-30 23:42:09.920
** Platform: Windows
** Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
** Python executable: D:\ComfyUI-Zluda\venv\Scripts\python.exe
** ComfyUI Path: D:\ComfyUI-Zluda
** ComfyUI Base Folder Path: D:\ComfyUI-Zluda
** User directory: D:\ComfyUI-Zluda\user
** ComfyUI-Manager config path: D:\ComfyUI-Zluda\user\default\ComfyUI-Manager\config.ini
** Log path: D:\ComfyUI-Zluda\user\comfyui.log
[ComfyUI-Manager] Windows event loop policy mode enabled

Prestartup times for custom nodes:
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\rgthree-comfy
2.3 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager

Checkpoint files will always be loaded safely.

:: Checking package versions...
Found pydantic: 2.11.7, pydantic-settings: 2.10.1
:: Pydantic packages are compatible, skipping reinstall
Installed version of comfyui-frontend-package: 1.25.11
Installed version of comfyui-workflow-templates: 0.1.70
Installed version of av: 14.3.0
Installed version of comfyui-embedded-docs: 0.2.6
:: Package version check complete.

:: ------------------------ ZLUDA ----------------------- ::
:: Auto-detecting AMD GPU architecture for Triton...
:: Detected GPU via Windows registry: AMD Radeon RX 5700
:: Set TRITON_OVERRIDE_ARCH=gfx1010
:: Triton core imported successfully
:: Enabled cuDNN
:: Running Triton kernel test...
:: Triton kernel test passed successfully
:: Triton initialized successfully
:: Patching ONNX Runtime for ZLUDA — disabling CUDA EP.
:: Using ZLUDA with device: AMD Radeon RX 5700 [ZLUDA]
:: Applying core ZLUDA patches...
:: Initializing Triton optimizations
:: Configuring Triton device properties...
:: Triton device properties configured
:: Flash attention components found
:: AMD flash attention enabled successfully
:: Configuring PyTorch backends...
:: Disabled CUDA flash attention
:: Enabled math attention fallback
:: ZLUDA initialization complete
:: ------------------------ ZLUDA ----------------------- ::

Total VRAM 8176 MB, total RAM 32693 MB
pytorch version: 2.7.0+cu118
Enabled fp16 accumulation.
Set vram state to: LOW_VRAM
Disabling smart memory management
Device: cuda:0 AMD Radeon RX 5700 [ZLUDA] : native
Using async weight offloading with 2 streams
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
Python version: 3.11.9 (tags/v3.11.9:de54cf5, Apr 2 2024, 10:12:12) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.3.56
ComfyUI frontend version: 1.25.11
[Prompt Server] web root: D:\ComfyUI-Zluda\venv\Lib\site-packages\comfyui_frontend_package\static
[Crystools INFO] Crystools version: 1.27.3
[Crystools INFO] Platform release: 10
[Crystools INFO] JETSON: Not detected.
[Crystools INFO] CPU: AMD Ryzen 7 5800X3D 8-Core Processor - Arch: AMD64 - OS: Windows 10
[Crystools INFO] pynvml (NVIDIA) initialized.
[Crystools INFO] GPU/s:
[Crystools INFO] 0) AMD Radeon RX 5700
[Crystools INFO] NVIDIA Driver: AMD Driver: Unknown
ComfyUI-FreeMemory nodes loaded: Image, Latent, Model, CLIP, and String versions available.

Loading: ComfyUI-Impact-Pack (V8.22)

[Impact Pack] Wildcards loading done.

Loading: ComfyUI-Impact-Subpack (V1.3.5)

[Impact Pack/Subpack] Using folder_paths to determine whitelist path: D:\ComfyUI-Zluda\user\default\ComfyUI-Impact-Subpack\model-whitelist.txt
[Impact Pack/Subpack] Ensured whitelist directory exists: D:\ComfyUI-Zluda\user\default\ComfyUI-Impact-Subpack
[Impact Pack/Subpack] Loaded 17 model(s) from whitelist: D:\ComfyUI-Zluda\user\default\ComfyUI-Impact-Subpack\model-whitelist.txt
[Impact Subpack] ultralytics_bbox: D:\ComfyUI-Zluda\models\ultralytics\bbox
[Impact Subpack] ultralytics_segm: D:\ComfyUI-Zluda\models\ultralytics\segm

Loading: ComfyUI-Manager (V3.36)

[ComfyUI-Manager] network_mode: private

ComfyUI Revision: 4948 [7c17c18] | Released on '2025-08-30'

[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/alter-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/model-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json
[ComfyUI-Manager] The private comfyregistry is not yet supported in network_mode=private.
[ComfyUI-Manager] All startup tasks have been completed.
[ReActor] - STATUS - Running v0.6.1-b3 in ComfyUI
Torch version: 2.7.0+cu118
[ComfyUI-RMBG] v2.7.0 | 31 nodes Loaded
(pysssss:WD14Tagger) [DEBUG] Available ORT providers: DmlExecutionProvider, CPUExecutionProvider
(pysssss:WD14Tagger) [DEBUG] Using ORT providers: CUDAExecutionProvider, CPUExecutionProvider
[D:\ComfyUI-Zluda\custom_nodes\comfyui_controlnet_aux] | INFO -> Using ckpts path: D:\ComfyUI-Zluda\custom_nodes\comfyui_controlnet_aux\ckpts
[D:\ComfyUI-Zluda\custom_nodes\comfyui_controlnet_aux] | INFO -> Using symlinks: False
[D:\ComfyUI-Zluda\custom_nodes\comfyui_controlnet_aux] | INFO -> Using ort providers: ['CUDAExecutionProvider', 'DirectMLExecutionProvider', 'OpenVINOExecutionProvider', 'ROCMExecutionProvider', 'CPUExecutionProvider', 'CoreMLExecutionProvider']
FaceDetailer: Model directory already exists
FaceDetailer: Model already exists
Efficiency Nodes: Attempting to add Control Net options to the 'HiRes-Fix Script' Node (comfyui_controlnet_aux add-on)...Success!

[rgthree-comfy] Loaded 48 extraordinary nodes. 🎉

Import times for custom nodes:
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\cfz_vae_loader.py
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\websocket_image_save.py
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\cfz_cudnn.toggle.py
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\cfz_patcher.py
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-deepcache
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-FreeMemory
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui-wd14-tagger
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui_ipadapter_plus
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui_essentials
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\RyuuNoodles
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-Gallery
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-ExtendIPAdapterClipVision
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\facerestore_cf
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui_controlnet_aux
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\efficiency-nodes-comfyui
0.0 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui-impact-subpack
0.1 seconds: D:\ComfyUI-Zluda\custom_nodes\rgthree-comfy
0.1 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui-custom-scripts
0.1 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui-impact-pack
0.2 seconds: D:\ComfyUI-Zluda\custom_nodes\DZ-FaceDetailer
0.3 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-Crystools
0.5 seconds: D:\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager
0.6 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui-reactor-node
0.7 seconds: D:\ComfyUI-Zluda\custom_nodes\comfyui-rmbg

Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Using LRU cache
Starting server

To see the GUI go to: http://127.0.0.1:8188
disable_logs False
use_polling_observer False
Serving static files from D:\ComfyUI-Zluda\output at /static_gallery
FileSystemMonitor: Watchdog monitoring thread started.
disable_logs False
use_polling_observer False
FileSystemMonitor: Monitor already running, stopping previous monitor.
FileSystemMonitor: Watchdog monitoring thread stopped.
Serving static files from D:\ComfyUI-Zluda\output at /static_gallery
FileSystemMonitor: Watchdog monitoring thread started.
got prompt
Using split attention in VAE
Using split attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.float16
CFZ VAE: Loaded with forced precision fp16
got prompt
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using split attention in VAE
Using split attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.float16
Requested to load SDXLClipModel
loaded completely 9.5367431640625e+25 1560.802734375 True
CLIP/text encoder model load device: cpu, offload device: cpu, current: cpu, dtype: torch.float16
Requested to load SDXLClipModel
[CUDNN_TOGGLE] torch.backends.cudnn.enabled set to False (was True)
[CUDNN_TOGGLE] torch.backends.cudnn.benchmark still set to False
EasyCache enabled - threshold: 1.0, start_percent: 0.0, end_percent: 0.95
Requested to load SDXL
loaded completely 6323.93201171875 4897.0483474731445 True
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [00:24<00:00, 1.04it/s]
EasyCache - skipped 11/25 steps (1.79x speedup).
Attempting to free GPU VRAM and system RAM...
GPU VRAM: Initial usage: 5.22 GB, Final usage: 5.22 GB, Freed: 0.00 GB
Garbage collector: collected 1688 objects.
System RAM: Initial usage: 31.50%, Final usage: 31.50%, Freed: 0.00%
Requested to load AutoencoderKL
loaded completely 4123.1875 159.55708122253418 True
Prompt executed in 36.98 seconds
EasyCache enabled - threshold: 1.0, start_percent: 0.0, end_percent: 0.95
Requested to load SDXL
loaded completely 6137.18201171875 4897.0483474731445 True
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [00:23<00:00, 1.05it/s]
EasyCache - skipped 11/25 steps (1.79x speedup).
Attempting to free GPU VRAM and system RAM...
GPU VRAM: Initial usage: 5.21 GB, Final usage: 5.21 GB, Freed: 0.00 GB
Garbage collector: collected 3223 objects.
System RAM: Initial usage: 31.50%, Final usage: 31.50%, Freed: 0.00%
Requested to load AutoencoderKL
loaded completely 4118.6875 159.55708122253418 True
Prompt executed in 32.14 seconds
EasyCache enabled - threshold: 1.0, start_percent: 0.0, end_percent: 0.95
Requested to load SDXL
loaded completely 6137.18201171875 4897.0483474731445 True
100%|██████████████████████████████████████████████████████████████████████████████████| 25/25 [00:23<00:00, 1.05it/s]
EasyCache - skipped 11/25 steps (1.79x speedup).
Attempting to free GPU VRAM and system RAM...
GPU VRAM: Initial usage: 5.21 GB, Final usage: 5.21 GB, Freed: 0.00 GB
Garbage collector: collected 1380 objects.
System RAM: Initial usage: 30.90%, Final usage: 30.90%, Freed: 0.00%
Requested to load AutoencoderKL
loaded completely 4118.6875 159.55708122253418 True
Prompt executed in 32.14 seconds

[zereshkealt]

@Ashir555 Trial and error is the key when it comes to old stuff like RX5000 series. Just you wait once we switch to rocm 6.4.2, new wave of issues likely incoming for us.

This might not be helpful but I gonna share my experience, I do remember having issues with triton 3.4.0 in past with compiling or on every first generation (after stopping and hitting run again it worked), so I am still rocking old v3.0.0 because of it. Perhaps worth experimenting if you up for it.

Maybe worth noting I am using gfx1010xnack- version with HIP SDK extension applied over it, and adrenalin 25.8.1 (driver only).

Anyway, make sure VS BuildTools are actually installed.

• clear cache (C:\Users\<username>\.triton) or run cache-clean.bat if you wanna wipe all cache (zluda, triton, miopen...)
• double check VS BuildTools are installed, open PowerShell or cmd type cl ,it should print (depending on version):
  Microsoft (R) C/C++ Optimizing Compiler Version 19.44.35214 for x64
  Copyright (C) Microsoft Corporation. All rights reserved.
• check environment variables for PATH (C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\<your version>\bin\Hostx64\x64)
• cmd pip cache purge doesn't hurt if you wanna start fresh and force package re-download

PS. install-n.bat is like pandora's box without much output info of what is actually going on in background. After comparing files I noticed sometimes it fails Patch DLLs without notice, so I had to patch (move and rename) files manually.

Attached my messy .bat if you wanna try it, works on my RX5700 (non-XT). sidenote: I do keep triton cache inside comfy folder.

comfyui-n - CUSTOM.zip

Log (EasyCache go brrr)

interesting! so i don't have VS installed? i ruined my windows installation 3 times already and the damn thing, isn't even installed?

but it says, i have the damn thing, installed:

[Ashir555]

@Ashir555 Trial and error is the key when it comes to old stuff like RX5000 series. Just you wait once we switch to rocm 6.4.2, new wave of issues likely incoming for us.

I'm already living them: I wasn't able to use triton but at least I could generate images and videos, after upgrading HIP to v6.4.2 I can't...

Maybe worth noting I am using gfx1010xnack- version with HIP SDK extension applied over it, and adrenalin 25.8.1 (driver only).

Same I was using, no problems.

Anyway, make sure VS BuildTools are actually installed.

They are, but the cl command doesn't responds unless you are in the installation folder.

* clear cache (`C:\Users\<username>\.triton`) or run `cache-clean.bat` if you wanna wipe all cache (zluda, triton, miopen...)

Done, no change.

* double check VS BuildTools are installed, open PowerShell or cmd type `cl` ,it should print (depending on version):
  Microsoft (R) C/C++ Optimizing Compiler Version 19.44.35214 for x64
  Copyright (C) Microsoft Corporation.  All rights reserved.

cl doesn't work on CMD or PowerShell unless I'm inside the installation folder.

* check environment variables for `PATH` (`C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Tools\MSVC\<your version>\bin\Hostx64\x64`)

It is, but it's like it isn't...

PS. install-n.bat is like pandora's box without much output info of what is actually going on in background. After comparing files I noticed sometimes it fails Patch DLLs without notice, so I had to patch (move and rename) files manually.

That's because of the --quiet parameters. I delete them so I can see the whole installation process.

Attached my messy .bat if you wanna try it, works on my RX5700 (non-XT). sidenote: I do keep triton cache inside comfy folder.

comfyui-n - CUSTOM.zip
Log (EasyCache go brrr)

Why you are not using sage attention?

How much performance did you notice after using triton?