LTX-2 i2v on RX 7900 XTX, strange behavior with --disable-pinned-memory

### Custom Node Testing

- [ ] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

consistent behavior:
- if first run works, then second run shouldn't crash
- the higher the resolution, the slower the execution

(in my #11551 ticket, user OrsoEric commented that his crashes on the second run where resolved by using --disable-pinned-memory. This also seems to have an, albeit strange, effect on LTX-2 flows)

### Actual Behavior

second run crashes when not using --disable-pinned-memory
when using --disable-pinned-memory, the second run with 1024x1024 is faster than second run with 512x512

### Steps to Reproduce

- fresh install of ComfyUI_windows_portable_amd.7z version 9.2
- run with these command-line parameters: .\python.exe -s .\ComfyUI\main.py --windows-standalone-build --use-pytorch-cross-attention --disable-smart-memory
- Load Template : LTX-2 Image to Video (Distilled)
- Change model to: ltx-2-19b-distilled-fp8.safetensors
- use a 1024x1024 source image
- set "resize image" also to 1024x1024
- run the flow
- run the flow again

### Debug Logs

```powershell
## ComfyUI-Manager: installing dependencies done.
[2026-01-18 14:08:31.664] ** ComfyUI startup time: 2026-01-18 14:08:31.664
[2026-01-18 14:08:31.664] ** Platform: Windows
[2026-01-18 14:08:31.664] ** Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
[2026-01-18 14:08:31.664] ** Python executable: C:\Comfy\python_embeded\python.exe
[2026-01-18 14:08:31.664] ** ComfyUI Path: C:\Comfy\ComfyUI
[2026-01-18 14:08:31.664] ** ComfyUI Base Folder Path: C:\Comfy\ComfyUI
[2026-01-18 14:08:31.664] ** User directory: C:\Comfy\ComfyUI\user
[2026-01-18 14:08:31.664] ** ComfyUI-Manager config path: C:\Comfy\ComfyUI\user\__manager\config.ini
[2026-01-18 14:08:31.664] ** Log path: C:\Comfy\ComfyUI\user\comfyui.log

Prestartup times for custom nodes:
[2026-01-18 14:08:32.455] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-easy-use
[2026-01-18 14:08:32.455] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\rgthree-comfy
[2026-01-18 14:08:32.455] 1.8 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-manager
[2026-01-18 14:08:32.455]
[2026-01-18 14:08:33.715] Checkpoint files will always be loaded safely.
[2026-01-18 14:08:33.958] Total VRAM 24560 MB, total RAM 64669 MB
[2026-01-18 14:08:33.958] pytorch version: 2.9.0+rocmsdk20251116
[2026-01-18 14:08:33.958] Set: torch.backends.cudnn.enabled = False for better AMD performance.
[2026-01-18 14:08:33.958] AMD arch: gfx1100
[2026-01-18 14:08:33.958] ROCm version: (7, 1)
[2026-01-18 14:08:33.958] Set vram state to: NORMAL_VRAM
[2026-01-18 14:08:33.958] Disabling smart memory management
[2026-01-18 14:08:33.958] Device: cuda:0 AMD Radeon RX 7900 XTX : native
[2026-01-18 14:08:33.968] Using async weight offloading with 2 streams
[2026-01-18 14:08:33.968] Enabled pinned memory 29101.0
[2026-01-18 14:08:34.836] 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']}
[2026-01-18 14:08:34.836] Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
[2026-01-18 14:08:34.836] Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []}
[2026-01-18 14:08:35.089] Using pytorch attention
[2026-01-18 14:08:36.890] Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
[2026-01-18 14:08:36.890] ComfyUI version: 0.9.2
[2026-01-18 14:08:36.906] ComfyUI frontend version: 1.36.14
[2026-01-18 14:08:36.922] [Prompt Server] web root: C:\Comfy\python_embeded\Lib\site-packages\comfyui_frontend_package\static
[2026-01-18 14:08:39.569] [ComfyUI-Easy-Use] server: v1.3.4 Loaded
[2026-01-18 14:08:39.569] [ComfyUI-Easy-Use] web root: C:\Comfy\ComfyUI\custom_nodes\comfyui-easy-use\web_version/v2 Loaded
[2026-01-18 14:08:39.578] ComfyUI-GGUF: Allowing full torch compile
[2026-01-18 14:08:39.863] ### Loading: ComfyUI-Manager (V3.39.2)
[2026-01-18 14:08:39.863] [ComfyUI-Manager] network_mode: offline
[2026-01-18 14:08:39.863] [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'.
[2026-01-18 14:08:39.927] ### ComfyUI Revision: 150 [8f40b43e] *DETACHED | Released on '2026-01-15'
[2026-01-18 14:08:39.943] [ComfyUI-Manager] All startup tasks have been completed.
[2026-01-18 14:08:39.943] [C:\Comfy\ComfyUI\custom_nodes\comfyui_controlnet_aux] | INFO -> Using ckpts path: C:\Comfy\ComfyUI\custom_nodes\comfyui_controlnet_aux\ckpts
[2026-01-18 14:08:39.943] [C:\Comfy\ComfyUI\custom_nodes\comfyui_controlnet_aux] | INFO -> Using symlinks: False
[2026-01-18 14:08:39.943] [C:\Comfy\ComfyUI\custom_nodes\comfyui_controlnet_aux] | INFO -> Using ort providers: ['CUDAExecutionProvider', 'DirectMLExecutionProvider', 'OpenVINOExecutionProvider', 'ROCMExecutionProvider', 'CPUExecutionProvider', 'CoreMLExecutionProvider']
[2026-01-18 14:08:40.164] C:\Comfy\ComfyUI\custom_nodes\comfyui_controlnet_aux\node_wrappers\dwpose.py:26: UserWarning: DWPose: Onnxruntime not found or doesn't come with acceleration providers, switch to OpenCV with CPU device. DWPose might run very slowly
warnings.warn("DWPose: Onnxruntime not found or doesn't come with acceleration providers, switch to OpenCV with CPU device. DWPose might run very slowly")
[2026-01-18 14:08:40.212]
[2026-01-18 14:08:40.212] [rgthree-comfy] Loaded 48 fantastic nodes. 🎉
[2026-01-18 14:08:40.212]
[2026-01-18 14:08:40.212] [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.

[2026-01-18 14:08:40.212]
Import times for custom nodes:
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\websocket_image_save.py
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\seedvarianceenhancer
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-mxtoolkit
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\random-int-divisor-node
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\ComfyUI-GGUF
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui_essentials
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-frame-interpolation
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-kjnodes
[2026-01-18 14:08:40.212] 0.0 seconds: C:\Comfy\ComfyUI\custom_nodes\rgthree-comfy
[2026-01-18 14:08:40.212] 0.1 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-manager
[2026-01-18 14:08:40.212] 0.2 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui_controlnet_aux
[2026-01-18 14:08:40.212] 0.3 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-itools
[2026-01-18 14:08:40.212] 1.7 seconds: C:\Comfy\ComfyUI\custom_nodes\comfyui-easy-use
[2026-01-18 14:08:40.212]
[2026-01-18 14:08:40.228] Context impl SQLiteImpl.
[2026-01-18 14:08:40.228] Will assume non-transactional DDL.
[2026-01-18 14:08:40.275] Assets scan(roots=['models']) completed in 0.057s (created=0, skipped_existing=349, total_seen=349)
[2026-01-18 14:08:40.323] Starting server

[2026-01-18 14:08:40.323] To see the GUI go to: http://127.0.0.1:8188
[2026-01-18 14:08:42.731] [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.
[2026-01-18 14:08:42.732] [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.
[2026-01-18 14:08:42.733] [DEPRECATION WARNING] Detected import of deprecated legacy API: /extensions/core/widgetInputs.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.
[2026-01-18 14:08:43.272] [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.
[2026-01-18 14:08:43.276] [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.
[2026-01-18 14:09:27.689] got prompt
[2026-01-18 14:09:28.735] Found quantization metadata version 1
[2026-01-18 14:09:28.735] Detected mixed precision quantization
[2026-01-18 14:09:28.735] Using mixed precision operations
[2026-01-18 14:09:28.796] model weight dtype torch.bfloat16, manual cast: torch.bfloat16
[2026-01-18 14:09:28.798] model_type FLUX
[2026-01-18 14:09:28.922] unet unexpected: ['audio_embeddings_connector.learnable_registers', 'audio_embeddings_connector.transformer_1d_blocks.0.attn1.k_norm.weight',

'video_embeddings_connector.transformer_1d_blocks.0.attn1.k_norm.weight',

'video_embeddings_connector.transformer_1d_blocks.1.ff.net.2.weight']
[2026-01-18 14:09:29.988] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[2026-01-18 14:09:29.990] no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded.
[2026-01-18 14:09:29.992] Requested to load VideoVAE
[2026-01-18 14:09:30.429] loaded completely; 22891.42 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:09:31.927] Found quantization metadata version 1
[2026-01-18 14:09:31.927] Using MixedPrecisionOps for text encoder
[2026-01-18 14:09:32.948] Missing weight for layer gemma3_12b.transformer.model.layers.0.self_attn.q_proj

[2026-01-18 14:09:32.980] Missing weight for layer gemma3_12b.transformer.vision_model.encoder.layers.26.mlp.fc2
[2026-01-18 14:09:32.982] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[2026-01-18 14:09:32.987] Requested to load LTXAVTEModel_
[2026-01-18 14:09:44.005] loaded completely; 22788.42 MB usable, 11725.34 MB loaded, full load: True
[2026-01-18 14:09:46.259] Requested to load LTXAV
[2026-01-18 14:10:14.090] loaded completely; 22466.25 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:17:10.577]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [06:56<00:00, 57.86s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [06:56<00:00, 52.04s/it]
[2026-01-18 14:18:18.517] Requested to load VideoVAE
[2026-01-18 14:18:19.100] loaded completely; 21982.69 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:18:19.570] Requested to load LTXAV
[2026-01-18 14:18:25.996] loaded partially; 19999.10 MB usable, 19975.09 MB loaded, 566.18 MB offloaded, 24.01 MB buffer reserved, lowvram patches: 0
[2026-01-18 14:25:17.136]
100%|███████████████████████████████████████████████████████████████████████████████████| 3/3 [06:51<00:00, 138.38s/it]
100%|███████████████████████████████████████████████████████████████████████████████████| 3/3 [06:51<00:00, 137.05s/it]
[2026-01-18 14:25:21.998] Requested to load AudioVAE
[2026-01-18 14:25:22.156] loaded completely; 22421.86 MB usable, 415.20 MB loaded, full load: True
[2026-01-18 14:25:22.775] Requested to load VideoVAE
[2026-01-18 14:25:23.364] loaded completely; 9571.55 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:25:37.218] Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding.
[2026-01-18 14:25:55.364] Prompt executed in 00:16:27
[2026-01-18 14:26:08.482] got prompt
[2026-01-18 14:26:08.551] Requested to load LTXAV
[2026-01-18 14:26:08.551] 0 models unloaded.
[2026-01-18 14:26:11.749] loaded partially; 0.00 MB usable, 0.00 MB loaded, 20541.27 MB offloaded, 448.07 MB buffer reserved, lowvram patches: 0
[2026-01-18 14:26:16.393]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:04<00:00, 1.72it/s]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:04<00:00, 1.72it/s]

:0:C:\develop\TheRock\rocm-systems\projects\clr\rocclr\device\device.cpp:379 : 246126728560 us: Memobj map does not have ptr: 0xccd38000
```

### Other

Remark that the second execution (see debug log above), just before the crash, takes 4 seconds for the 8 steps (and is expressed in it/s instead of s/it). I can only dream that my hardware was that powerfull :-)

When I do the exact same steps as above, but run with an additional command line parameter: --disable-pinned-memory
Then the second run does NOT crash. And not only does it work correctly, it does so 9 times faster!
(and if you look at just the first sampler stage, then you can see the second run is 15(!) times faster)
This is the relevant log for the first and second run when using that extra option:
```
[2026-01-18 13:40:11.793] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[2026-01-18 13:40:11.799] Requested to load LTXAVTEModel_
[2026-01-18 13:40:22.823] loaded completely; 22788.42 MB usable, 11725.34 MB loaded, full load: True
[2026-01-18 13:40:25.066] Requested to load LTXAV
[2026-01-18 13:40:46.892] loaded completely; 22466.25 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 13:47:46.462]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [06:59<00:00, 58.32s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [06:59<00:00, 52.42s/it]
[2026-01-18 13:48:53.880] Requested to load VideoVAE
[2026-01-18 13:48:54.484] loaded completely; 21982.69 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 13:48:54.714] Requested to load LTXAV
[2026-01-18 13:49:00.618] loaded partially; 19999.10 MB usable, 19975.09 MB loaded, 566.18 MB offloaded, 24.01 MB buffer reserved, lowvram patches: 0
[2026-01-18 13:55:55.276]
100%|███████████████████████████████████████████████████████████████████████████████████| 3/3 [06:54<00:00, 139.49s/it]
100%|███████████████████████████████████████████████████████████████████████████████████| 3/3 [06:54<00:00, 138.22s/it]
[2026-01-18 13:56:00.075] Requested to load AudioVAE
[2026-01-18 13:56:00.242] loaded completely; 22421.86 MB usable, 415.20 MB loaded, full load: True
[2026-01-18 13:56:00.868] Requested to load VideoVAE
[2026-01-18 13:56:01.543] loaded completely; 9571.55 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 13:56:15.520] Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding.
[2026-01-18 13:56:33.938] Prompt executed in 00:16:27
[2026-01-18 13:59:23.704] got prompt
[2026-01-18 13:59:23.788] Requested to load LTXAV
[2026-01-18 13:59:23.788] 0 models unloaded.
[2026-01-18 13:59:23.911] loaded partially; 0.00 MB usable, 0.00 MB loaded, 20541.27 MB offloaded, 448.07 MB buffer reserved, lowvram patches: 0
[2026-01-18 13:59:51.056]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:27<00:00, 3.34s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:27<00:00, 3.39s/it]
[2026-01-18 13:59:51.541] Requested to load VideoVAE
[2026-01-18 13:59:51.572] 0 models unloaded.
[2026-01-18 13:59:51.572] loaded partially; 0.00 MB usable, 0.00 MB loaded, 2378.23 MB offloaded, 648.02 MB buffer reserved, lowvram patches: 0
[2026-01-18 13:59:51.858] Requested to load LTXAV
[2026-01-18 13:59:51.873] 0 models unloaded.
[2026-01-18 13:59:51.955] loaded partially; 0.00 MB usable, 0.00 MB loaded, 20541.27 MB offloaded, 448.07 MB buffer reserved, lowvram patches: 0
[2026-01-18 14:00:32.324]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:40<00:00, 13.45s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:40<00:00, 13.46s/it]
[2026-01-18 14:00:32.401] Requested to load AudioVAE
[2026-01-18 14:00:32.403] 0 models unloaded.
[2026-01-18 14:00:32.403] loaded partially; 0.00 MB usable, 0.00 MB loaded, 415.20 MB offloaded, 54.01 MB buffer reserved, lowvram patches: 0
[2026-01-18 14:00:32.526] Requested to load VideoVAE
[2026-01-18 14:00:32.531] 0 models unloaded.
[2026-01-18 14:00:32.531] loaded partially; 0.00 MB usable, 0.00 MB loaded, 2378.23 MB offloaded, 648.02 MB buffer reserved, lowvram patches: 0
[2026-01-18 14:01:01.024] Warning: Ran out of memory when regular VAE decoding, retrying with tiled VAE decoding.
[2026-01-18 14:01:17.940] Prompt executed in 114.22 seconds
```

But when I run the exact same flow, but with a 512x512 resolution, then I don't see this weird speed-up behavior.

**ComfyUI.log for a 512x512 run without the --disable-pinned-memory flag**
Observe that the first and second run execute in about the same time (174.97 vs 177.52):
```
[2026-01-18 14:33:02.167] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[2026-01-18 14:33:02.175] Requested to load LTXAVTEModel_
[2026-01-18 14:33:13.250] loaded completely; 22788.42 MB usable, 11725.34 MB loaded, full load: True
[2026-01-18 14:33:15.547] Requested to load LTXAV
[2026-01-18 14:33:37.384] loaded completely; 22421.81 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:35:07.787]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [01:30<00:00, 12.50s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [01:30<00:00, 11.28s/it]
[2026-01-18 14:35:28.718] Requested to load VideoVAE
[2026-01-18 14:35:29.215] loaded completely; 22179.49 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:35:29.334] Requested to load LTXAV
[2026-01-18 14:35:35.419] loaded completely; 22369.31 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:35:42.678]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:07<00:00, 2.39s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:07<00:00, 2.39s/it]
[2026-01-18 14:35:46.747] Requested to load AudioVAE
[2026-01-18 14:35:46.889] loaded completely; 22383.49 MB usable, 415.20 MB loaded, full load: True
[2026-01-18 14:35:47.190] Requested to load VideoVAE
[2026-01-18 14:35:47.647] loaded completely; 18390.50 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:35:51.974] Prompt executed in 174.97 seconds
[2026-01-18 14:36:10.962] got prompt
[2026-01-18 14:36:11.005] Requested to load LTXAV
[2026-01-18 14:36:16.690] loaded completely; 22370.18 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:37:49.164]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [01:32<00:00, 12.86s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [01:32<00:00, 11.55s/it]
[2026-01-18 14:38:06.526] Requested to load VideoVAE
[2026-01-18 14:38:06.934] loaded completely; 22207.30 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:38:07.291] Requested to load LTXAV
[2026-01-18 14:38:12.729] loaded completely; 22357.13 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:39:00.198]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:47<00:00, 15.86s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:47<00:00, 15.80s/it]
[2026-01-18 14:39:03.825] Requested to load AudioVAE
[2026-01-18 14:39:03.951] loaded completely; 22371.30 MB usable, 415.20 MB loaded, full load: True
[2026-01-18 14:39:04.046] Requested to load VideoVAE
[2026-01-18 14:39:04.585] loaded completely; 18390.50 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:39:08.485] Prompt executed in 177.52 seconds
```

**ComfyUI.log for a 512x512 run with the --disable-pinned-memory flag**
Observe that the first and second run execute in about the same time (133.39 vs 134.71)
(in the 1024x1024 run, the second run was much much faster)
Also observe that the first Sampler in this second 512x512 run is now slower (92 sec) then the second run in the 1024x1024 first Sampler (27 sec)
```
[2026-01-18 14:42:56.340] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[2026-01-18 14:42:56.346] Requested to load LTXAVTEModel_
[2026-01-18 14:43:07.566] loaded completely; 22788.42 MB usable, 11725.34 MB loaded, full load: True
[2026-01-18 14:43:09.822] Requested to load LTXAV
[2026-01-18 14:43:29.080] loaded completely; 22421.81 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:44:19.051]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:49<00:00, 6.89s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:49<00:00, 6.22s/it]
[2026-01-18 14:44:34.225] Requested to load VideoVAE
[2026-01-18 14:44:34.713] loaded completely; 22179.49 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:44:34.832] Requested to load LTXAV
[2026-01-18 14:44:40.847] loaded completely; 22369.31 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:44:48.125]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:07<00:00, 2.39s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:07<00:00, 2.39s/it]
[2026-01-18 14:44:52.080] Requested to load AudioVAE
[2026-01-18 14:44:52.223] loaded completely; 22383.49 MB usable, 415.20 MB loaded, full load: True
[2026-01-18 14:44:52.525] Requested to load VideoVAE
[2026-01-18 14:44:53.013] loaded completely; 18390.50 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:44:57.073] Prompt executed in 125.84 seconds
[2026-01-18 14:45:26.262] got prompt
[2026-01-18 14:45:26.313] Requested to load LTXAV
[2026-01-18 14:45:31.688] loaded completely; 22370.18 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:47:02.660]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [01:30<00:00, 12.59s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [01:30<00:00, 11.37s/it]
[2026-01-18 14:47:19.665] Requested to load VideoVAE
[2026-01-18 14:47:20.140] loaded completely; 22207.30 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:47:20.260] Requested to load LTXAV
[2026-01-18 14:47:24.480] loaded completely; 22357.13 MB usable, 20541.27 MB loaded, full load: True
[2026-01-18 14:47:31.767]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:07<00:00, 2.41s/it]
100%|████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:07<00:00, 2.41s/it]
[2026-01-18 14:47:35.341] Requested to load AudioVAE
[2026-01-18 14:47:35.452] loaded completely; 22371.30 MB usable, 415.20 MB loaded, full load: True
[2026-01-18 14:47:35.546] Requested to load VideoVAE
[2026-01-18 14:47:35.972] loaded completely; 18390.50 MB usable, 2378.23 MB loaded, full load: True
[2026-01-18 14:47:39.660] Prompt executed in 133.39 seconds
```

my hardware:
CPU : AMD Ryzen 9 7950X (16c/32t)
GPU : AMD Radeon RX 7900 XTX (24 GB VRAM)
System Memory : 64 GB
OS : Windows 11
installed driver : https://www.amd.com/en/resources/support-articles/release-notes/RN-AMDGPU-WINDOWS-PYTORCH-7-1-1.html