SDXL Stuck at KSampler

[xhy2008]

Your question

Hi,I'm working with SDXL basic txt2img workflow on ZLUDA.
My comfyui was installed this morning.
The problem is that comfyui stucked at KSampler node for about twenty minutes before it started the sampling steps,and the only output is "Compiling in progress. Please wait...".
After the first image generation,the second one still stucks at the same place for about ten minutes.
I don't know what caused that.
Could anyone help me?

Logs

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

Adding extra search path checkpoints E:/sdnext\models/Stable-diffusion
Adding extra search path configs E:/sdnext\configs
Adding extra search path vae E:/sdnext\models/VAE
Adding extra search path loras E:/sdnext\models/Lora
Adding extra search path loras E:/sdnext\models/LyCORIS
Adding extra search path upscale_models E:/sdnext\models/ESRGAN
Adding extra search path upscale_models E:/sdnext\models/RealESRGAN
Adding extra search path upscale_models E:/sdnext\models/SwinIR
Adding extra search path embeddings E:/sdnext\models/embeddings
Adding extra search path hypernetworks E:/sdnext\models/hypernetworks
Adding extra search path controlnet E:/sdnext\models/control/controlnet
[START] Security scan
[DONE] Security scan
## ComfyUI-Manager: installing dependencies done.
** ComfyUI startup time: 2025-02-10 10:05:11.984
** Platform: Windows
** Python version: 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
** Python executable: E:\ComfyUI-Zluda\venv\Scripts\python.exe
** ComfyUI Path: E:\ComfyUI-Zluda
** ComfyUI Base Folder Path: E:\ComfyUI-Zluda
** User directory: E:\ComfyUI-Zluda\user
** ComfyUI-Manager config path: E:\ComfyUI-Zluda\user\default\ComfyUI-Manager\config.ini
** Log path: E:\ComfyUI-Zluda\user\comfyui.log

Prestartup times for custom nodes:
   5.8 seconds: E:\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager

Warning, you are using an old pytorch version and some ckpt/pt files might be loaded unsafely. Upgrading to 2.4 or above is recommended.

***----------------------ZLUDA-----------------------------***
  ::  ZLUDA detected, disabling non-supported functions.
  ::  CuDNN, flash_sdp, math_sdp, mem_efficient_sdp disabled)
***--------------------------------------------------------***
  ::  Device: AMD Radeon(TM) Vega 8 Graphics [ZLUDA]

Total VRAM 4116 MB, total RAM 6030 MB
pytorch version: 2.3.0+cu118
Set vram state to: NO_VRAM
Disabling smart memory management
Device: cuda:0 AMD Radeon(TM) Vega 8 Graphics [ZLUDA] : native
Using sub quadratic optimization for attention, if you have memory or speed issues try using: --use-split-cross-attention
ComfyUI version: 0.3.14
[Prompt Server] web root: E:\ComfyUI-Zluda\web
### Loading: ComfyUI-Manager (V3.18.1)
### ComfyUI Revision: 3560 [9561b03c] | Released on '2025-02-09'

Import times for custom nodes:
   0.0 seconds: E:\ComfyUI-Zluda\custom_nodes\ComfyUI-deepcache
   0.0 seconds: E:\ComfyUI-Zluda\custom_nodes\websocket_image_save.py
   1.1 seconds: E:\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager

Starting server

To see the GUI go to: http://127.0.0.1:8188
[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/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
FETCH DATA from: E:\ComfyUI-Zluda\custom_nodes\ComfyUI-Manager\extension-node-map.json [DONE]
FETCH ComfyRegistry Data: 5/33
FETCH ComfyRegistry Data: 10/33
FETCH ComfyRegistry Data: 15/33
FETCH ComfyRegistry Data: 20/33
FETCH ComfyRegistry Data: 25/33
got prompt
Failed to validate prompt for output 21:
* KSampler 22:
  - Value -1 smaller than min of 0: seed
Output will be ignored
invalid prompt: {'type': 'prompt_outputs_failed_validation', 'message': 'Prompt outputs failed validation', 'details': '', 'extra_info': {}}
FETCH ComfyRegistry Data: 30/33
FETCH ComfyRegistry Data [DONE]
[ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
nightly_channel: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/remote
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.jsongot prompt
 [DONE]
[ComfyUI-Manager] All startup tasks have been completed.
model weight dtype torch.float16, manual cast: None
model_type EPS
Using split attention in VAE
Using split attention in VAE
VAE load device: cpu, offload device: cpu, dtype: torch.float32
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 SDXL
loaded partially 9.5367431640625e-08 0.0 0
  0%|                                                                                                                                                                | 0/20 [00:00<?, ?it/s]Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [32:21<00:00, 97.09s/it]
Requested to load AutoencoderKL
loaded completely 9.5367431640625e+25 319.11416244506836 True
Prompt executed in 2044.94 seconds
got prompt
Requested to load SDXL
loaded partially 9.5367431640625e-08 0.0 0
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
Compiling in progress. Please wait...
  0%|                                                                                                                                                                | 0/20 [00:00<?, ?it/s]Compiling in progress. Please wait...
E:\ComfyUI-Zluda\venv\lib\site-packages\torchsde\_brownian\brownian_interval.py:608: UserWarning: Should have tb<=t1 but got tb=14.614643096923828 and t1=14.614643.
  warnings.warn(f"Should have {tb_name}<=t1 but got {tb_name}={tb} and t1={self._end}.")
Compiling in progress. Please wait...

Other

COMMANDLINE_ARGS=--use-quad-cross-attention --auto-launch --novram --disable-smart-memory --cpu-vae --fp32-vae
I have 4GB of VRAM,but comfyui only used 2GB.
Besides,the GPU usage was lower than 50% most of the time.

[patientx]

Can you see the phrase "Compiling in progress. Please wait..."
That means zluda is still building its database as written on readme, this is a one time thing (for any new type of model, for example first time you use a sd1.5 , sdxl ,flux, ltx , upscalers .. ) will do this FOR ONE TIME ONLY. GPU isn't actually being used here, I am not sure but cpu might be more important at this stage only. As for why it takes too much it might be the low vram, weak gpu or cpu in your case. Just wait. Also you can see it passing the first part. ALSO for the first try , change the resolution to 512x512 , change the step count to 1. If it produces something then the system is working. From there you can determine what your system can do.