Comfyui freezes at KSampler on 4090 64Gb RAM

[Suther1and]

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

Hey! I’m having an issue I can’t figure out. I did a fresh install of ComfyUI (again) . My GPU is an RTX 4090 and 64 GB of RAM.

When I use normal text-to-image, image-to-image, or upscale, everything works fine. It takes about 20 seconds to generate an image and 1–2 minutes for an upscale. GPU load goes to 99–100%, RAM around 70–85%, and VRAM 80–95%.

The problem starts when I try heavier tasks like inpaint or outpaint. Most of the time it gets stuck on the KSampler node at a certain percent. From there it can take dozens of minutes to move forward while pushing my RAM to 80–85%, GPU to 99–100%, and VRAM to 97–99%. For example, an inpaint recently took 20 minutes to finish. Sometimes it takes so long i have to restart the server (most of the times end task button does nothing)

I’ve already tried closing everything else on my PC while running ComfyUI, but it didn’t help. I don’t think it’s supposed to behave like this with my specs, so I’m guessing something is wrong with my setup. Any help would be really appreciated.

I tried portable, installer and git clone versions on Comfui, tried different browsers, all the combinations of everything.

Logs

got prompt
Using pytorch attention in VAE
Using pytorch attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
Requested to load FluxClipModel_
loaded completely 9.5367431640625e+25 9319.23095703125 True
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16
clip missing: ['text_projection.weight']
Requested to load AutoencodingEngine
loaded completely 9346.644035339355 159.87335777282715 True
model weight dtype torch.bfloat16, manual cast: None
model_type FLUX
Requested to load Flux
loaded partially 21274.370919891357 21274.279418945312 0
 15%|████████████▎                                                                     | 3/20 [07:08<40:26, 142.75s/it]

Other

No response

[Suther1and]

update: i did a lot of chatgpt suggestions including installation of studio driver. After Comfyui load i managed to generate 2 images with same workflow: 42 sec and 19 sec. After that i changed the image (same 1024x1024) and its stuck at KSampler 75% once again for 10 minutes

[Suther1and]

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
got prompt
Using pytorch attention in VAE
Using pytorch attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
clip missing: ['text_projection.weight']
Requested to load FluxClipModel_
loaded completely 21472.8 9319.23095703125 True
Requested to load AutoencodingEngine
loaded completely 9346.644035339355 159.87335777282715 True
model weight dtype torch.bfloat16, manual cast: None
model_type FLUX
FETCH ComfyRegistry Data: 40/95
Requested to load Flux
loaded partially 21274.370919891357 21274.279418945312 0
0%| | 0/20 [00:00<?, ?it/s]FETCH ComfyRegistry Data: 45/95
70%|█████████████████████████████████████████████████████████▍ | 14/20 [00:14<00:05, 1.14it/s]FETCH ComfyRegistry Data: 50/95
100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:20<00:00, 1.01s/it]
Requested to load AutoencodingEngine
loaded completely 551.685546875 159.87335777282715 True
Prompt executed in 42.80 seconds
FETCH ComfyRegistry Data: 55/95
FETCH ComfyRegistry Data: 60/95
got prompt
loaded partially 19992.742013641357 19992.119262695312 0
45%|█████████████████████████████████████▎ | 9/20 [00:10<00:09, 1.21it/s]FETCH ComfyRegistry Data: 65/95
100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:18<00:00, 1.06it/s]
Requested to load AutoencodingEngine
loaded completely 266.412109375 159.87335777282715 True
Prompt executed in 19.91 seconds
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.
got prompt
Requested to load FluxClipModel_
loaded completely 12047.889532852172 9319.23095703125 True
0 models unloaded.
loaded partially 21274.370919891357 21274.279418945312 0
0%| | 0/20 [00:00<?, ?it/s]

[JavadZohani]

It's funny how everyone has run into this problem in their own way, and nobody seems willing to actually fix it.

[Suther1and]

update2: i closed every single other program on my pc and it solved my issue. Sometimes i still reach very high VRAM usage but at least it works. I doubt it should load it to 99% with simple workflows but it is what it is.

[JavadZohani]

update2: i closed every single other program on my pc and it solved my issue. Sometimes i still reach very high VRAM usage but at least it works. I doubt it should load it to 99% with simple workflows but it is what it is.

That’s true, but it might work temporarily on your system. Many other systems don’t have enough RAM, and even when they allocate a lot for this program, it might still be better to stick with older versions for more reliable performance.

[comfyanonymous]

Can you post the entire log?

[Suther1and]

Can you post the entire log?

Above is the entire log from that session. I can only generate stuff when i have 0 other apps running in background. That lowers VRAM usage to 96-97% and generation proceed. If i have another app running it goes to 97-99 and causes freeze. Is there a way to lower the usage?

[Suther1and]

If i have a portable version do i need to change anything related to cuda and pytorch?

[JavadZohani]

Can you post the entire log?
This test is with the latest version.

D:\AI\Packages\ComfyUI>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Checkpoint files will always be loaded safely.
Total VRAM 24576 MB, total RAM 32346 MB
pytorch version: 2.8.0+cu129
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 4090 Ti : cudaMallocAsync
Using pytorch attention
Python version: 3.13.6 (tags/v3.13.6:4e66535, Aug 6 2025, 14:36:00) [MSC v.1944 64 bit (AMD64)]
ComfyUI version: 0.3.54
****** User settings have been changed to be stored on the server instead of browser storage. ******
****** For multi-user setups add the --multi-user CLI argument to enable multiple user profiles. ******
ComfyUI frontend version: 1.25.11
[Prompt Server] web root: D:\AI\Packages\ComfyUI\python_embeded\Lib\site-packages\comfyui_frontend_package\static

Import times for custom nodes:
0.0 seconds: D:\AI\Packages\ComfyUI\ComfyUI\custom_nodes\websocket_image_save.py

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
got prompt
Using pytorch attention in VAE
Using pytorch attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
Requested to load WanVAE
0 models unloaded.
loaded partially 128.0 127.9998779296875 0
Using scaled fp8: fp8 matrix mult: False, scale input: False
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
0 models unloaded.
loaded partially 128.0 127.9998779296875 0
Requested to load QwenImageTEModel_
loaded completely 9460.800122070312 7909.737449645996 True
0 models unloaded.
loaded partially 128.0 127.9998779296875 0
Requested to load QwenImageTEModel_
loaded completely 9442.675122070312 7909.737449645996 True

D:\AI\Packages\ComfyUI>pause
Press any key to continue . . .

[JavadZohani]

I’ve been waiting for a few days for this issue to be fixed, and in the meantime I’ve been using other programs to see if the problem gets resolved or not.

[fypovich]

I have a similar problem, it freezes at different stages of generation.

I also have 64 GB of RAM and an RTX 5070 TI.

Has anyone managed to fix this?

[WinPooh32]

@fypovich @JavadZohani @Suther1and
Disabling cache helped to me:

--cache-none --mmap-torch-files

Full example:

python3 main.py --cache-none --mmap-torch-files

[github-actions]

This issue is being marked stale because it has not had any activity for 30 days. Reply below within 7 days if your issue still isn't solved, and it will be left open. Otherwise, the issue will be closed automatically.