Windows fatal exception: stack overflow.

[BigStationW]

Hello,

I can't run the CLIPLoaderDisTorch2MultiGPU node anymore.

[MultiGPU DisTorch V2] Full allocation string:
Requested to load ZImageTEModel_
Windows fatal exception: stack overflow

Stack (most recent call first):
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 743 in _load_list
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 1545 in load
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 1707 in partially_load
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 612 in model_use_more_vram
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 581 in model_load
  File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-MultiGPU\distorch_2.py", line 203 in patched_load_models_gpu
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 422 in load_model
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 375 in encode_from_tokens
  File "D:\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 313 in encode_from_tokens_scheduled
  File "D:\ComfyUI_windows_portable\ComfyUI\nodes.py", line 80 in encode
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 296 in process_inputs
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 308 in _async_map_node_over_list
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 334 in get_output_data
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 525 in execute
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 758 in execute_async
  File "asyncio\events.py", line 88 in _run
  File "asyncio\base_events.py", line 1999 in _run_once
  File "asyncio\base_events.py", line 645 in run_forever
  File "asyncio\windows_events.py", line 322 in run_forever
  File "asyncio\base_events.py", line 678 in run_until_complete
  File "asyncio\runners.py", line 118 in run
  File "asyncio\runners.py", line 195 in run
  File "D:\ComfyUI_windows_portable\ComfyUI\execution.py", line 702 in execute
  File "D:\ComfyUI_windows_portable\ComfyUI\main.py", line 276 in prompt_worker
  File "threading.py", line 1012 in run
  File "threading.py", line 1075 in _bootstrap_inner
  File "threading.py", line 1032 in _bootstrap

I know you tried to fix it on PR#183 but it still doesn't work for me.

[beeteam368]

Yes, I just updated it too, and it's still the same!

Please check! Thanks, Regards!

[Bigfield77]

same here.

Once installed it breaks every workflow even if no MultiGPU node is being used in that workflow.

if i only expose one cuda device, the problem goes away (set CUDA_VISIBLE_DEVICES=0)

[Mr-Hazem-Eng-Artist]

Yeah same problem for me here :/
once I deleted the MultiGPU nodes & restarted Comfy, everything went back to work normally, before that, Comfy 0.17 workflow stops at Text Clip Encoders with errors, I updated to this latest March build of the nodes & the same errors here in this thread started happening & Comfy doesn't open.

ComfyUi 0.17 torch 2.8 cuda 12.8 python 3.12.10

[Bigfield77]

starting the server with --disable-dynamic-vram also fixes the issue for me.

[Mr-Hazem-Eng-Artist]

--disable-dynamic-vram

I confirm it worked but with errors elsewhere, that is what happened:
I used the flag of "--disable-dynamic-vram" , ComfyUi loaded & opened, MultiGPU nodes are loaded,
a workflow with Clip & Vae multigpu on 2nd card cuda:1 , for WAN2.2 worked & rendered the video just before it reached a Tensorrt node & it gave an error,
I use Rife Tensorrt nodes for interpolation & upscaling from user ( yuvraj108c ) , Before ComfyUi 0.17 everything worked well & no issues :\ ,, here is the error regarding Tensorrt :

[E] [defaultAllocator.cpp::nvinfer1::internal::DefaultAllocator::allocate::31] Error Code 1: Cuda Runtime (out of memory)
[E] [executionContext.cpp::nvinfer1::rt::ExecutionContext::ExecutionContext::568] Error Code 2: OutOfMemory (Requested size was 7549747200 bytes.)
ERROR 03-24 09:01:22 [execution.py:612] !!! Exception during processing !!! 'NoneType' object has no attribute 'set_input_shape'

=====

EDIT: I had to remove MultiGPU nodes & also removed 2 flags from the run_nvidia_gpu.bet ( --disable-pinned-memory --disable-dynamic-vram ) ,, & since then, my usual older workflows work well again & some times benefit more from the dynamic VRAM features, no conflicts or errors with clip encoders nodes & no issues launching Comfy ..

so I hope the author will find a solution for MultiGPU in a new upcoming update to make it work normally again with full features without issues like before with the current ComfyUi 0.18.xx versions

[tecknight]

This issue is happening again to me, even when using no multigpu nodes.

[Mr-Hazem-Eng-Artist]

So recently I was on Comfy 0.17 & 0.18.2 ,, but I reverted back to ComfyUi v.0.14.1 (version 0.14.2 is also the very same with just a single hotfix for nano banana api node or though) & I used a version of MultiGPU from month of February ( 2.5.10 ) ,

& Everything went back to normal (No OOMs , No lots of error messages , more stable as usual as before) , I also use the following flags in .bat startupfile of Comfy : --disable-pinned-memory --reserve-vram 1.5

I will stick to those versions of Comfy & MultiGPU for now, I am tired of issues coming with version 0.16+ all the time , even with other things like broken Subgraphs & ... etc

[cptcrash80]

So recently I was on Comfy 0.17 & 0.18.2 ,, but I reverted back to ComfyUi v.0.14.1 (version 0.14.2 is also the very same with just a single hotfix for nano banana api node or though) & I used a version of MultiGPU from month of February ( 2.5.10 ) ,

& Everything went back to normal (No OOMs , No lots of error messages , more stable as usual as before) , I also use the following flags in .bat startupfile of Comfy : --disable-pinned-memory --reserve-vram 1.5

I will stick to those versions of Comfy & MultiGPU for now, I am tired of issues coming with version 0.16+ all the time , even with other things like broken Subgraphs & ... etc

@Mr-Hazem-Eng-Artist Which version of Accelerate do you have installed? I can't get the nodes to load due to it not being able to find the module. I manually installed it but the error persists.