[Windows Portable][Blackwell] Official MiniMax H3 Ref workflow ends in nvlddmkm DPC timeout and 0x133 reboot on a fresh v0.31.0 install

[JerryWn12]

Custom Node Testing

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

Expected Behavior

The official MiniMax H3 Reference-to-Video workflow should complete on a supported RTX 50-series GPU, or report a recoverable CUDA/application error. It should not leave the NVIDIA kernel driver in a DPC long enough to trigger DPC_WATCHDOG_VIOLATION (0x133) and reboot Windows.

Actual Behavior

After a completely fresh installation of ComfyUI 0.31.0, I ran the official MiniMax H3 Reference-to-Video workflow. During video generation, pc automatically restarted without a visible Python exception.

The minidump, 081126-17718-01.dmp, records:

DPC_WATCHDOG_VIOLATION (0x133)
Arg1: 0 (one DPC/ISR exceeded its time allotment)
Arg2: 0x500
Arg3: 0x500
DPC_TIMEOUT_TYPE: SINGLE_DPC_TIMEOUT_EXCEEDED
NVIDIA driver: nvlddmkm.sys 32.0.16.1088 / 610.88
Crash time: 2026-08-11 17:02:18 +08:00
System uptime: 1:05:53

The active stack contains 17 consecutive nvlddmkm.sys frames before returning to the Windows DPC path:

nvlddmkm+0xe9023
nvlddmkm+0x69f535
nvlddmkm+0x1252f3
nvlddmkm+0x309b2e
nvlddmkm+0x124d54
nvlddmkm+0x10ebb0
nvlddmkm+0xa98ed9
nvlddmkm+0xa97702
nvlddmkm+0xa978cd
nvlddmkm+0x14eea4
nvlddmkm+0x3a43b6
nvlddmkm+0xacf571
nvlddmkm+0xacde4a
nvlddmkm+0x129318
nvlddmkm+0x12a87c
nvlddmkm+0x12b8cc
nvlddmkm+0xe93ba

Steps to Reproduce

1. Install ComfyUI 0.31.0 on Windows 11.

2. launch.bat (I haven't install any custom nodes so --disable-all-custom-nodes is not added):

   .\python_embeded\python.exe -s ComfyUI\main.py `
       --output-directory "C:\Users\<USER>\Pictures\ComfyUI" `
       --user-directory "C:\Users\<USER>\Documents\ComfyUI" `
       --disable-auto-launch `
       --disable-api-nodes `
       --disable-metadata `
       --fast fp16_accumulation

3. Open the official workflow video_minimax_h3_r2v from comfyui_workflow_templates_json 0.1.37.

4. Use the official Comfy-Org model files:

   • minimax_h3_ref2va_pruned_int8_convrot.safetensors
   • qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
   • minimax_h3_video_vae_fp16.safetensors
   • minimax_h3_audio_vae_fp32.safetensors

5. Set the generation parameters to 0.8 MP, 3:4, 5 seconds, 20 steps, simple scheduler, and res_multistep sampler.

6. Queue the generation. PC may automatically reboot during generation without a Python exception being printed.

Debug Logs

C:\Users\<USER>\AppData\Local\Programs\ComfyUI>.\python_embeded\python.exe -s ComfyUI\main.py     --output-directory "C:\Users\<USER>\Pictures\ComfyUI"     --user-directory "C:\Users\<USER>\Documents\ComfyUI"     --disable-auto-launch     --disable-api-nodes     --disable-metadata     --fast fp16_accumulation
[INFO] setup plugin alembic.autogenerate.schemas
[INFO] setup plugin alembic.autogenerate.tables
[INFO] setup plugin alembic.autogenerate.types
[INFO] setup plugin alembic.autogenerate.constraints
[INFO] setup plugin alembic.autogenerate.defaults
[INFO] setup plugin alembic.autogenerate.comments
[INFO] Adding extra search path checkpoints C:\Users\<USER>\Files\Models\checkpoints
[INFO] Adding extra search path configs C:\Users\<USER>\Files\Models\configs
[INFO] Adding extra search path loras C:\Users\<USER>\Files\Models\loras
[INFO] Adding extra search path vae C:\Users\<USER>\Files\Models\vae
[INFO] Adding extra search path text_encoders C:\Users\<USER>\Files\Models\text_encoders
[INFO] Adding extra search path text_encoders C:\Users\<USER>\Files\Models\clip
[INFO] Adding extra search path diffusion_models C:\Users\<USER>\Files\Models\unet
[INFO] Adding extra search path diffusion_models C:\Users\<USER>\Files\Models\diffusion_models
[INFO] Adding extra search path clip_vision C:\Users\<USER>\Files\Models\clip_vision
[INFO] Adding extra search path style_models C:\Users\<USER>\Files\Models\style_models
[INFO] Adding extra search path embeddings C:\Users\<USER>\Files\Models\embeddings
[INFO] Adding extra search path diffusers C:\Users\<USER>\Files\Models\diffusers
[INFO] Adding extra search path vae_approx C:\Users\<USER>\Files\Models\vae_approx
[INFO] Adding extra search path controlnet C:\Users\<USER>\Files\Models\controlnet
[INFO] Adding extra search path controlnet C:\Users\<USER>\Files\Models\t2i_adapter
[INFO] Adding extra search path gligen C:\Users\<USER>\Files\Models\gligen
[INFO] Adding extra search path upscale_models C:\Users\<USER>\Files\Models\upscale_models
[INFO] Adding extra search path latent_upscale_models C:\Users\<USER>\Files\Models\latent_upscale_models
[INFO] Adding extra search path datasets C:\Users\<USER>\Files\Models\datasets
[INFO] Adding extra search path hypernetworks C:\Users\<USER>\Files\Models\hypernetworks
[INFO] Adding extra search path photomaker C:\Users\<USER>\Files\Models\photomaker
[INFO] Adding extra search path classifiers C:\Users\<USER>\Files\Models\classifiers
[INFO] Adding extra search path model_patches C:\Users\<USER>\Files\Models\model_patches
[INFO] Adding extra search path audio_encoders C:\Users\<USER>\Files\Models\audio_encoders
[INFO] Adding extra search path background_removal C:\Users\<USER>\Files\Models\background_removal
[INFO] Adding extra search path frame_interpolation C:\Users\<USER>\Files\Models\frame_interpolation
[INFO] Adding extra search path geometry_estimation C:\Users\<USER>\Files\Models\geometry_estimation
[INFO] Adding extra search path optical_flow C:\Users\<USER>\Files\Models\optical_flow
[INFO] Adding extra search path detection C:\Users\<USER>\Files\Models\detection
[INFO] Setting output directory to: C:\Users\<USER>\Pictures\ComfyUI
[INFO] Setting user directory to: C:\Users\<USER>\Documents\ComfyUI
[INFO] Found comfy_kitchen backend hip: {'available': False, 'disabled': False, 'unavailable_reason': 'PyTorch ROCm/HIP runtime not available', 'capabilities': []}
[INFO] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_embedding', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'int8_linear', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']}
[INFO] Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []}
[INFO] Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'int8_linear', 'na3d', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']}
[INFO] Checkpoint files will always be loaded safely.
[INFO] Total VRAM 12227 MB, total RAM 31935 MB
[INFO] pytorch version: 2.13.0+cu130
[INFO] Enabled fp16 accumulation.
[INFO] Set vram state to: NORMAL_VRAM
[INFO] Device: cuda:0 NVIDIA GeForce RTX 5070 Ti Laptop GPU : cudaMallocAsync
[INFO] Using async weight offloading with 2 streams
[INFO] Enabled pinned memory 12773.0
[INFO] Using pytorch attention
[INFO] aimdo: src-win/cuda-detour.c:38:INFO:aimdo_setup_hooks: installing 6 hooks
[INFO] aimdo: src/control.c:262:INFO:comfy-aimdo NVML pressure enabled
[INFO] aimdo: src-win/shmem-detect.c:80:INFO:comfy-aimdo WDDM adapter match: NVIDIA GeForce RTX 5070 Ti Laptop GPU runtime_luid=00000000:0001065d dxgi_luid=00000000:0001065d
[INFO] aimdo: src/control.c:277:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5070 Ti Laptop GPU (VRAM: 12226 MB)
[INFO] DynamicVRAM support detected and enabled
[INFO] Python version: 3.13.14 (tags/v3.13.14:fd17997, Jun 10 2026, 13:03:48) [MSC v.1944 64 bit (AMD64)]
[INFO] ComfyUI version: 0.31.0
[INFO] comfy-aimdo version: 0.4.13
[INFO] comfy-kitchen version: 0.2.30
[INFO] comfyui-frontend-package version: 1.48.7
[INFO] comfyui-workflow-templates version: 0.11.37
[INFO] comfyui-embedded-docs version: 0.5.9
[INFO] comfy-kitchen version: 0.2.30
[INFO] comfy-aimdo version: 0.4.13
[INFO] [Prompt Server] web root: C:\Users\<USER>\AppData\Local\Programs\ComfyUI\python_embeded\Lib\site-packages\comfyui_frontend_package\static
[INFO] Asset seeder disabled
[INFO] No OpenGL_accelerate module loaded: No module named 'OpenGL_accelerate'
[INFO]
Import times for custom nodes:
[INFO]    0.0 seconds: C:\Users\<USER>\AppData\Local\Programs\ComfyUI\ComfyUI\custom_nodes\websocket_image_save.py
[INFO]
[ERROR] Failed to initialize database. Please ensure you have installed the latest requirements. If the error persists, please report this as in future the database will be required: (sqlite3.OperationalError) unable to open database file
(Background on this error at: https://sqlalche.me/e/20/e3q8)
[INFO] Using RAM pressure cache.
[INFO] Starting server

[INFO] To see the GUI go to: http://127.0.0.1:8188

Other

System

• Laptop: MECHREVO CANGLONG 16 Ultra
• BIOS: N.1.12MRO16
• OS: Windows 11 Pro, build 26200
• CPU: AMD Ryzen 9 9955HX3D, 16 cores / 32 threads
• RAM: 32 GB
• GPU: NVIDIA GeForce RTX 5070 Ti Laptop GPU, 12 GB VRAM, Blackwell
• GPU VBIOS: 98.05.31.00.70
• Graphics mode: hybrid graphics
• NVIDIA driver: 610.88 / 32.0.16.1088, clean-installed
• NVIDIA Overlay: disabled

Workflow File

MiniMax-H3-Ref-Official.json

The database initialization error is present at startup, before the generation begins. The server nevertheless starts successfully; there is no evidence in this transcript that this message is related to the later NVIDIA kernel timeout.

The clean-install environment was verified from the installed files:

ComfyUI: 0.31.0
Python: 3.13.14
PyTorch: 2.13.0+cu130
comfy-aimdo: 0.4.13
comfy-kitchen: 0.2.30
comfyui-frontend-package: 1.48.7
comfyui-workflow-templates: 0.11.37
comfyui_workflow_templates_json: 0.1.37
SageAttention: not installed
Triton: not installed
Third-party custom nodes: none

Windows recorded the following for the primary occurrence:

2026-08-11 17:02:39  Microsoft-Windows-Kernel-Power  Event 41
BugcheckCode=307 (0x133)
BugcheckParameter1=0x0
BugcheckParameter2=0x500
BugcheckParameter3=0x500
BugcheckParameter4=0xfffff80072dc53c8
LongPowerButtonPressDetected=false
WHEABootErrorCount=0

2026-08-11 17:02:52  Microsoft-Windows-WER-SystemErrorReporting  Event 1001
Bugcheck: 0x00000133 (0x0, 0x500, 0x500, 0xfffff80072dc53c8)
Dump: C:\Windows\Minidump\081126-17718-01.dmp

[CHKE9]

same issue on my 5090

[JerryWn12]

Update: It didn't crash by using the ck-attention + w4a8 model + 8b text encoder + int8 video vae + KJNodes Low VRAM Attention + Spectrum.
Same resolution & steps, but it introduce a bit quality reduction.
The updated workflow: MiniMax-H3-Ref-V2.json