Slow generation after recent update

[midmind]

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

After update now i get stuck with low vram patches, while before update everything was working completely fine

Logs

Adding extra search path animatediff_models C:\ComfyUI_portable\ComfyUI\models\animatediff_models
Adding extra search path animatediff_motion_lora C:\ComfyUI_portable\ComfyUI\models\animatediff_motion_lora
Adding extra search path animatediff_video_formats C:\ComfyUI_portable\ComfyUI\models\animatediff_video_formats
Adding extra search path audio_encoders C:\ComfyUI_portable\ComfyUI\models\audio_encoders
Adding extra search path blip C:\ComfyUI_portable\ComfyUI\models\blip
Adding extra search path checkpoints C:\ComfyUI_portable\ComfyUI\models\checkpoints
Adding extra search path clip C:\ComfyUI_portable\ComfyUI\models\clip
Adding extra search path clip_vision C:\ComfyUI_portable\ComfyUI\models\clip_vision
Adding extra search path CogVideo C:\ComfyUI_portable\ComfyUI\models\CogVideo
Adding extra search path configs C:\ComfyUI_portable\ComfyUI\models\configs
Adding extra search path controlnet C:\ComfyUI_portable\ComfyUI\models\controlnet
Adding extra search path diffusers C:\ComfyUI_portable\ComfyUI\models\diffusers
Adding extra search path diffusion_models C:\ComfyUI_portable\ComfyUI\models\diffusion_models
Adding extra search path embeddings C:\ComfyUI_portable\ComfyUI\models\embeddings
Adding extra search path facedetection C:\ComfyUI_portable\ComfyUI\models\facedetection
Adding extra search path facerestore_models C:\ComfyUI_portable\ComfyUI\models\facerestore_models
Adding extra search path gligen C:\ComfyUI_portable\ComfyUI\models\gligen
Adding extra search path hypernetworks C:\ComfyUI_portable\ComfyUI\models\hypernetworks
Adding extra search path insightface C:\ComfyUI_portable\ComfyUI\models\insightface
Adding extra search path Joy_caption C:\ComfyUI_portable\ComfyUI\models\Joy_caption
Adding extra search path latent_upscale_models C:\ComfyUI_portable\ComfyUI\models\latent_upscale_models
Adding extra search path layerstyle C:\ComfyUI_portable\ComfyUI\models\layerstyle
Adding extra search path liveportrait C:\ComfyUI_portable\ComfyUI\models\liveportrait
Adding extra search path LLM C:\ComfyUI_portable\ComfyUI\models\LLM
Adding extra search path loras C:\ComfyUI_portable\ComfyUI\models\loras
Adding extra search path luts C:\ComfyUI_portable\ComfyUI\models\luts
Adding extra search path model_patches C:\ComfyUI_portable\ComfyUI\models\model_patches
Adding extra search path motion_models C:\ComfyUI_portable\ComfyUI\models\motion_models
Adding extra search path nsfw_detector C:\ComfyUI_portable\ComfyUI\models\nsfw_detector
Adding extra search path onnx C:\ComfyUI_portable\ComfyUI\models\onnx
Adding extra search path other C:\ComfyUI_portable\ComfyUI\models\other
Adding extra search path photomaker C:\ComfyUI_portable\ComfyUI\models\photomaker
Adding extra search path poses C:\ComfyUI_portable\ComfyUI\models\poses
Adding extra search path reactor C:\ComfyUI_portable\ComfyUI\models\reactor
Adding extra search path sam3 C:\ComfyUI_portable\ComfyUI\models\sam3
Adding extra search path sams C:\ComfyUI_portable\ComfyUI\models\sams
Adding extra search path SEEDVR2 C:\ComfyUI_portable\ComfyUI\models\SEEDVR2
Adding extra search path style_models C:\ComfyUI_portable\ComfyUI\models\style_models
Adding extra search path t5 C:\ComfyUI_portable\ComfyUI\models\t5
Adding extra search path text_encoders C:\ComfyUI_portable\ComfyUI\models\text_encoders
Adding extra search path ultralytics C:\ComfyUI_portable\ComfyUI\models\ultralytics
Adding extra search path unet C:\ComfyUI_portable\ComfyUI\models\unet
Adding extra search path upscale_models C:\ComfyUI_portable\ComfyUI\models\upscale_models
Adding extra search path vae C:\ComfyUI_portable\ComfyUI\models\vae
Adding extra search path vae_approx C:\ComfyUI_portable\ComfyUI\models\vae_approx
Adding extra search path wildcards C:\ComfyUI_portable\ComfyUI\models\wildcards
Adding extra search path xlabs C:\ComfyUI_portable\ComfyUI\models\xlabs
Adding extra search path yolo C:\ComfyUI_portable\ComfyUI\models\yolo
[START] Security scan
[DONE] Security scan
## ComfyUI-Manager: installing dependencies done.
** ComfyUI startup time: 2026-03-11 00:36:55.544
** Platform: Windows
** Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr  8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
** Python executable: C:\ComfyUI-Easy\ComfyUI-Easy-Install\python_embeded\python.exe
** ComfyUI Path: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI
** ComfyUI Base Folder Path: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI
** User directory: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\user
** ComfyUI-Manager config path: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\user\__manager\config.ini
** Log path: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\user\comfyui.log

Prestartup times for custom nodes:
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\rgthree-comfy
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Easy-Use
   3.1 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-manager

Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Checkpoint files will always be loaded safely.
Total VRAM 12288 MB, total RAM 16310 MB
pytorch version: 2.9.1+cu130
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce RTX 3060 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 7339.0
working around nvidia conv3d memory bug.
Using sage attention
Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr  8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
ComfyUI version: 0.16.4
comfy-aimdo version: 0.2.9
comfy-kitchen version: 0.2.7
ComfyUI frontend version: 1.39.19
[Prompt Server] web root: C:\ComfyUI-Easy\ComfyUI-Easy-Install\python_embeded\Lib\site-packages\comfyui_frontend_package\static
Asset seeder disabled
[Crystools INFO] Crystools version: 1.27.4
[Crystools INFO] Platform release: 11
[Crystools INFO] JETSON: Not detected.
[Crystools INFO] CPU: AMD Ryzen 7 5800X 8-Core Processor - Arch: AMD64 - OS: Windows 11
[Crystools INFO] pynvml (NVIDIA) initialized.
[Crystools INFO] GPU/s:
[Crystools INFO] 0) NVIDIA GeForce RTX 3060
[Crystools INFO] NVIDIA Driver: 595.79
[ComfyUI-Easy-Use] server: v1.3.7 Loaded
[ComfyUI-Easy-Use] web root: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Easy-Use\web_version/v2 Loaded
ComfyUI-GGUF: Allowing full torch compile
### Loading: ComfyUI-Impact-Pack (V8.28.2)
[Impact Pack] custom_wildcards path not found: C:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Impact-Pack\custom_wildcards. Using default path.
[Impact Pack] Wildcard total size (0.00 MB) is within cache limit (50.00 MB). Using full cache mode.
[Impact Pack] Wildcards loading done.
### Loading: ComfyUI-Impact-Subpack (V1.3.5)
[Impact Pack/Subpack] Using folder_paths to determine whitelist path: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\user\default\ComfyUI-Impact-Subpack\model-whitelist.txt
[Impact Pack/Subpack] Ensured whitelist directory exists: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\user\default\ComfyUI-Impact-Subpack
[Impact Pack/Subpack] Loaded 0 model(s) from whitelist: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\user\default\ComfyUI-Impact-Subpack\model-whitelist.txt
[Impact Subpack] ultralytics_bbox: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\models\ultralytics\bbox
[Impact Subpack] ultralytics_segm: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\models\ultralytics\segm
### Loading: ComfyUI-Manager (V3.39.2)
[ComfyUI-Manager] network_mode: public
[ComfyUI-Manager] ComfyUI per-queue preview override detected (PR #11261). Manager's preview method feature is disabled. Use ComfyUI's --preview-method CLI option or 'Settings > Execution > Live preview method'.
### ComfyUI Version: v0.16.4-10-g8086468d | Released on '2026-03-10'
### Loading: ComfyUI-RBG-SmartSeedVariance ###
[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
[SuperNodes] Classes initialized.
[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json

__     ______   ____                      ____              ____ _      _
\ \   / /  _ \ / ___| __ _ _ __ ___   ___|  _ \  _____   __/ ___(_)_ __| |
 \ \ / /| |_) | |  _ / _` | '_ ` _ \ / _ \ | | |/ _ \ \ / / |  _| | '__| |
  \ V / |  _ <| |_| | (_| | | | | | |  __/ |_| |  __/\ V /| |_| | | |  | |
   \_/  |_| \_\\____|\__,_|_| |_| |_|\___|____/ \___| \_/  \____|_|_|  |_|

             🎮 VRGameDevGirl custom nodes loaded successfully! 🎞️

[ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
------------------------------------------
Comfyroll Studio v1.76 :  175 Nodes Loaded
------------------------------------------
** For changes, please see patch notes at https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes/blob/main/Patch_Notes.md
** For help, please see the wiki at https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes/wiki
------------------------------------------
[CRT-Nodes __init__] Importing node classes...
Loading crt-nodes module

================================================================================
🚨 IMPORTANT: **comfyui_controlnet_aux** is not installed or not detectable!
   Install it via ComfyUI Manager for full preprocessor support.
   Fallback: Only 'none' and 'canny' preprocessors available.
   GitHub: https://github.com/Fannovel16/comfyui_controlnet_aux
================================================================================

[CRT-Nodes __init__] Registering custom model paths...
... 'ultralytics_bbox' path registered.
... 'ultralytics_segm' path registered.
[CRT-Nodes __init__] INITIALIZATION COMPLETE.
Using sage attention
(RES4LYF) Init
(RES4LYF) Importing beta samplers.
(RES4LYF) Importing legacy samplers.

[rgthree-comfy] Loaded 48 extraordinary nodes. 🎉

[rgthree-comfy] ComfyUI's new Node 2.0 rendering may be incompatible with some rgthree-comfy nodes and features, breaking some rendering as well as losing the ability to access a node's properties (a vital part of many nodes). It also appears to run MUCH more slowly spiking CPU usage and causing jankiness and unresponsiveness, especially with large workflows. Personally I am not planning to use the new Nodes 2.0 and, unfortunately, am not able to invest the time to investigate and overhaul rgthree-comfy where needed. If you have issues when Nodes 2.0 is enabled, I'd urge you to switch it off as well and join me in hoping ComfyUI is not planning to deprecate the existing, stable canvas rendering all together.

⚡ SeedVR2 optimizations check: SageAttention ✅ | Flash Attention ✅ | Triton ✅
🔧 Conv3d workaround active: PyTorch 2.9.1, cuDNN 91200 (fixing VAE 3x memory bug)
📊 Initial CUDA memory: 10.93GB free / 12.00GB total
Logger already set up. Skipping setup_logger().
[⚡ZI_POWER INFO] Version: 1.0.0
[⚡ZI_POWER INFO] This package includes 9 nodes and 4 deprecated ones.
[⚡ZI_POWER INFO] It also features 123 predefined styles.

Import times for custom nodes:
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\websocket_image_save.py
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_CameraAngleSelector
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\quick-connections
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-ZImageTurboProgressiveLockedUpscale
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\Comfyui-Resolution-Master
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-simple-prompt-batcher
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-inpaint-cropandstitch
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\cg-use-everywhere
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-GGUF
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-TiledDiffusion
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-RBG-SmartSeedVariance
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui_lg_samplingutils
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\LanPaint
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyMath
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui_essentials
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Chibi-Nodes
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-apex-artist
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-kjnodes
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\basic_data_handling
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\rgthree-comfy
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\z-image-power-nodes
   0.0 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI_Comfyroll_CustomNodes
   0.1 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-impact-subpack
   0.2 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\crt-nodes
   0.3 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\seedvr2_videoupscaler
   0.4 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-supernodes
   0.4 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Crystools
   0.4 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\RES4LYF
   0.4 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Impact-Pack
   0.8 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-manager
   2.6 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\comfyui-vrgamedevgirl
   5.2 seconds: C:\ComfyUI-Easy\ComfyUI-Easy-Install\ComfyUI\custom_nodes\ComfyUI-Easy-Use

Context impl SQLiteImpl.
Will assume non-transactional DDL.
Starting server

To see the GUI go to: http://127.0.0.1:8188
FETCH ComfyRegistry Data: 5/129
[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/buttonGroup.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[DEPRECATION WARNING] Detected import of deprecated legacy API: /extensions/core/clipspace.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[DEPRECATION WARNING] Detected import of deprecated legacy API: /extensions/core/groupNode.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/button.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
FETCH ComfyRegistry Data: 10/129
FETCH ComfyRegistry Data: 15/129
FETCH ComfyRegistry Data: 20/129
FETCH ComfyRegistry Data: 25/129
FETCH ComfyRegistry Data: 30/129
FETCH ComfyRegistry Data: 35/129
FETCH ComfyRegistry Data: 40/129
FETCH ComfyRegistry Data: 45/129
FETCH ComfyRegistry Data: 50/129
FETCH ComfyRegistry Data: 55/129
FETCH ComfyRegistry Data: 60/129
FETCH ComfyRegistry Data: 65/129
FETCH ComfyRegistry Data: 70/129
FETCH ComfyRegistry Data: 75/129
FETCH ComfyRegistry Data: 80/129
got prompt
Using pytorch attention in VAE
Using pytorch attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
FETCH ComfyRegistry Data: 85/129
gguf qtypes: F32 (145), Q8_0 (253)
Dequantizing token_embd.weight to prevent runtime OOM.
FETCH ComfyRegistry Data: 90/129
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
gguf qtypes: F32 (245), F16 (24), Q8_0 (180), BF16 (4)
model weight dtype torch.bfloat16, manual cast: None
model_type FLOW
Using sage attention mode: auto
[ModelSamplingZImage] Applied: shift=3.0, multiplier=1.0
[ModelSamplingZImage] sigma_min=0.002994, sigma_max=1.000000
[ModelSamplingZImage] sigmas[0:5]=[0.002994012087583542, 0.005976095795631409, 0.008946320973336697, 0.011904762126505375, 0.014851485379040241]
[ModelSamplingZImage] Patch verification: patched shift=3.0, multiplier=1.0
[ModelSamplingZImage] Patched sigmas[0:5]=[0.002994012087583542, 0.005976095795631409, 0.008946320973336697, 0.011904762126505375, 0.014851485379040241]
Requested to load ZImageTEModel_
FETCH ComfyRegistry Data: 95/129
FETCH ComfyRegistry Data: 100/129
FETCH ComfyRegistry Data: 105/129
loaded completely; 9781.52 MB usable, 4424.19 MB loaded, full load: True
Requested to load Lumina2
FETCH ComfyRegistry Data: 110/129
FETCH ComfyRegistry Data: 115/129
FETCH ComfyRegistry Data: 120/129
FETCH ComfyRegistry Data: 125/129
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.
loaded completely; 7875.09 MB usable, 6973.32 MB loaded, full load: True
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = True
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = False
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = True
(RES4LYF) rk_type: res_2s
100%|████████████████████████████████████████████████████████████████████████████████████| 4/4 [00:30<00:00,  7.66s/it]
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = False
Requested to load AutoencodingEngine
Unloaded partially: 1020.78 MB freed, 5952.56 MB remains loaded, 225.00 MB buffer reserved, lowvram patches: 60
loaded completely; 424.34 MB usable, 159.87 MB loaded, full load: True
Requested to load Lumina2
loaded completely; 8630.49 MB usable, 6973.32 MB loaded, full load: True
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = True
(RES4LYF) rk_type: res_2s
100%|████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:36<00:00, 18.15s/it]
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = False
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = True
100%|████████████████████████████████████████████████████████████████████████████████████| 4/4 [00:28<00:00,  7.10s/it]
Patching torch settings: torch.backends.cuda.matmul.allow_fp16_accumulation = False
Requested to load AutoencodingEngine
Unloaded partially: 6445.97 MB freed, 527.36 MB remains loaded, 288.28 MB buffer reserved, lowvram patches: 232
loaded completely; 487.39 MB usable, 159.87 MB loaded, full load: True
Prompt executed in 159.98 seconds

Other

[ledeni00]

I am getting this, too. Together with periodically getting stuck on clip text encode and ksampler.
With 4090.
Just to add. Once the generation gets "stuck". Switching to another browser tab automatically continues the generation. Never seen that before.

[rattus128]

I cant strictly tell if you are slow at the first or second "partially unload", but it's usually a shared memory spill when this happens. This can stop the tiled VAE fallback from working for you. Check your nvidia control panel setting and make sure that shared memory fallback is switched off.

Im guessing these are reasonably high res images. From there the best thing to do in manually tile your VAE. Select a reasonable tile size (say 768x768) using the VAE Decode (tiled) node instead of the simple VAE decode.

[rattus128]

I am getting this, too. Together with periodically getting stuck on clip text encode and ksampler. With 4090. Just to add. Once the generation gets "stuck". Switching to another browser tab automatically continues the generation. Never seen that before.

That's bizarre and unlikely to be anything like what OP is experiencing.

I would love a video of that if you are up for it (Win + Shift + R), if you can get the terminal, task manager (GPU page) and the flaky browser behavior side by side.

[midmind]

I am writing here cause before an update it was perfectly fine, same exact settings,worrkflow, etc.
Now even smaller gguf version of zimage doesnt seem to do the trick, i am writing as before i've always resolved issues with help of google ai studio, but now i'm a little desperate tbh
Btw in the control panel cuda settings set to: prefer no sysmem fallback

[Kitaiwish]

I'm having the same issue, too. Using Z-Image Turbo (Q8 GGUF) running on RTX 4060 Ti 16 GB vram.

Below are what I observed to be different from before update:

1. Each time the prompt is changed, generation starts with a message, "Model ZImageTEModel_ prepared for dynamic VRAM loading. 3317MB Staged. 0 patches attached. Force pre-loaded 145 weights: 383 KB", which takes additional ~20 sec at max.

2. Sampler node starts with a message "Model Initializaing ... ". It takes another ~5 sec.
   0%| | 0/5 [00:00<?, ?it/s, Model Initializing ... ]

After turning off dynamic VRAM mode, everything works fine.

[comfyanonymous]

Try using native models instead of GGUF.

[kellyling80]

I have similar spec as OP, and I am also experiencing slow generation after upgrading and loading dynamic vram. Using native , Q6, or gguf models doesn't change the slowness and delays when it hits ksampler. Tried changing ksamplers/clown/simple...all similar timings.

[rattus128]

I have similar spec as OP, and I am also experiencing slow generation after upgrading and loading dynamic vram. Using native , Q6, or gguf models doesn't change the slowness and delays when it hits ksampler. Tried changing ksamplers/clown/simple...all similar timings.

Can I get your log workflow, and the before and after numbers?

[mindAddiction]

As mentioned in another issue, I'm experiencing the same massive slowdown since the previous updates.
Specifically the upscaling and VAE decode step are tremendously slowed down. Single image upscaling that used to take about a minute now takes about 6 to 8 minutes.

I have no idea what happened, but it is very, very slow in workflows and with prompts that used to be pretty fast before the update.

EDIT:

This is ridiculous, frankly:

[midmind]

I'm having the same issue, too. Using Z-Image Turbo (Q8 GGUF) running on RTX 4060 Ti 16 GB vram.

Below are what I observed to be different from before update:

1. Each time the prompt is changed, generation starts with a message, "Model ZImageTEModel_ prepared for dynamic VRAM loading. 3317MB Staged. 0 patches attached. Force pre-loaded 145 weights: 383 KB", which takes additional ~20 sec at max.
2. Sampler node starts with a message "Model Initializaing ... ". It takes another ~5 sec.
   0%| | 0/5 [00:00<?, ?it/s, Model Initializing ... ]

After turning off dynamic VRAM mode, everything works fine.

I did try disabling dynamic vram and it did not help in my case
also what i've noticed is that for some reason it load lot more than before, on the top of that with each new generation it stacks more and more which causes more lowvram patches(before the update i had 0 lowvram patches)

1st generation: Requested to load Lumina2
loaded completely; 7875.09 MB usable, 6973.32 MB loaded, full load: True

2nd generation:Requested to load Lumina2
loaded completely; 9092.22 MB usable, 6973.32 MB loaded, full load: True

[kingp0dd]

The VRAM issues you're describing seem related to my observations

#12934

[x3style]

I used to have fast generation around 60-70 seconds. After the update the generations were 380-780 seconds. I turned On the "Disable smart memory management" in Settings -> Server-Config -> about half way -> (which technically turns the feature off) and now I get 30-40 seconds generation.

[pinkmelonhere-dev]

===========================================================
Sampler node starts with a message "Model Initializaing ... ". It takes another ~5 sec.
0%| | 0/5 [00:00<?, ?it/s, Model Initializing ... ]

same here, also During the infernece step, gpu usage goes 1 or 2percent most of time and goes up to 90% and back to 1 or 2 very often, old comfy version just keeps around 90% . let me try "disable smart memory management