`--reserve-vram` ignored

### Custom Node Testing

- [x] 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

VRAM usage should never exceed (capacity – reserved)

### Actual Behavior

Running with `--reserve-vram 2 --disable-all-custom-nodes` with the default template workflow `video_minimax_h3_t2v`. Pushed resolution to 2.0 megapixels to hit VRAM capacity (31.5 GB - RTX 5090) and ran a single step before stopping execution. VRAM is expected never to exceed 29.5 GB with this setting.

**Pre-run**: VRAM: 2.8 GB - Shared: 0.3 GB - RAM: 9.1 GB
**1st run**: VRAM: 26.6 GB - Shared: 24.2 GB - RAM: 48.9 GB
**2nd run**: VRAM: 31.0 GB - Shared: 24.6 GB - RAM: 48.9 GB
**3rd run**: VRAM: 30.1 GB - Shared: 24.6 GB - RAM: 48.1 GB

After the first run was completed in 2nd and 3rd run, VRAM usage would lower to 23.3 GB, which I understand to mean that `--reserve-vram` is ignored during model load and initialization, but enforced afterward. Moreover, Comfy doesn't seem able to properly handle cache, using more VRAM on a second run than on a first.

### Steps to Reproduce

--

### Debug Logs

```powershell
[INFO] Found comfy_kitchen backend hip: {'available': False, 'disabled': False, 'unavailable_reason': 'PyTorch ROCm/HIP runtime not available', 'capabilities': []}
[INFO] Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'int8_linear', 'quantize_and_rotate_rowwise', 'quantize_int8_rowwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'w4a8_int8_linear']}
[INFO] Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'int8_linear', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']}
[INFO] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope1_', 'apply_rope_', 'apply_rope_split_half', 'apply_rope_split_half1', 'apply_rope_split_half1_', 'apply_rope_split_half_', 'convrot_w4a4_linear', 'dequantize_convrot_w4a4_weight', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_embedding', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'dequantize_w4a8_int8_weight', 'gemv_awq_w4a16', 'int8_linear', 'prepare_int4_weight_for_int8_linear', 'quantize_and_rotate_rowwise', 'quantize_convrot_w4a4_weight', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'quantize_w4a8_int8_weight', 'rms_adaln', 'rms_rope', 'rms_rope1', 'rms_rope1_', 'rms_rope_', 'rms_rope_split_half', 'rms_rope_split_half1', 'rms_rope_split_half1_', 'rms_rope_split_half_', 'rotate_int8_convrot_weight', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8', 'w4a8_int8_linear']}
[INFO] Checkpoint files will always be loaded safely.
[INFO] Total VRAM 32606 MB, total RAM 65417 MB
[INFO] pytorch version: 2.13.0+cu130
[INFO] Set vram state to: NORMAL_VRAM
[INFO] Device: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
[INFO] Using async weight offloading with 2 streams
[INFO] Enabled pinned memory 26166.0
[INFO] Using pytorch attention
[INFO] aimdo: src-win/cuda-detour.c:38:INFO:aimdo_setup_hooks: installing 6 hooks
[INFO] aimdo: src/control.c:262:INFO:comfy-aimdo NVML pressure enabled
[INFO] aimdo: src-win/shmem-detect.c:80:INFO:comfy-aimdo WDDM adapter match: NVIDIA GeForce RTX 5090 runtime_luid=00000000:0000aa1e dxgi_luid=00000000:0000aa1e
[INFO] aimdo: src/control.c:277:INFO:comfy-aimdo inited for GPU: NVIDIA GeForce RTX 5090 (VRAM: 32606 MB)
[INFO] DynamicVRAM support detected and enabled
[INFO] Python version: 3.13.14 (tags/v3.13.14:fd17997, Jun 10 2026, 13:03:48) [MSC v.1944 64 bit (AMD64)]
[INFO] ComfyUI version: 0.30.0
[INFO] comfy-aimdo version: 0.4.13
[INFO] comfy-kitchen version: 0.2.27
[INFO] comfyui-frontend-package version: 1.48.6
[INFO] comfyui-workflow-templates version: 0.11.31
[INFO] comfyui-embedded-docs version: 0.5.9
[INFO] comfy-kitchen version: 0.2.27
[INFO] comfy-aimdo version: 0.4.13
[INFO] [Prompt Server] web root: D:\comfy\python_embeded\Lib\site-packages\comfyui_frontend_package\static
[INFO] Asset seeder disabled
[INFO] No OpenGL_accelerate module loaded: No module named 'OpenGL_accelerate'
[INFO] Skipping loading of custom nodes
[INFO] Context impl SQLiteImpl.
[INFO] Will assume non-transactional DDL.
[INFO] Using RAM pressure cache.
[INFO] Starting server

[INFO] To see the GUI go to: http://127.0.0.1:8188
[INFO] got prompt
[INFO] VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
[INFO] VAE load device: cuda:0, offload device: cpu, dtype: torch.float16
[INFO] Found quantization metadata version 1
[INFO] Using MixedPrecisionOps for text encoder
[INFO] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[INFO] Requested to load MiniMaxH3TEModel_
[INFO] Model MiniMaxH3TEModel_ prepared for dynamic VRAM loading. 14956MB Staged. 0 patches attached. Force pre-loaded 410 weights: 4572 KB.
[INFO] Found quantization metadata version 1
[INFO] Detected mixed precision quantization
[INFO] Using mixed precision operations
[INFO] Native ops: mxfp8, nvfp4, float8_e4m3fn, int8_tensorwise, convrot_w4a4, float8_e5m2
[INFO] model weight dtype torch.bfloat16, manual cast: torch.bfloat16
[INFO] model_type FLOW_AV
[INFO] Requested to load MiniMaxH3
[INFO] 0 models unloaded.
[INFO] Model MiniMaxH3 prepared for dynamic VRAM loading. 19995MB Staged. 0 patches attached. Force pre-loaded 210 weights: 1175 KB.
5%|██▍ | 1/20 [00:59<18:58, 59.93s/it, Model Initialization complete! ]
[INFO] Interrupting prompt 7197f459-5587-476b-95f3-3d8a38f3117b
5%|██▎ | 1/20 [01:43<32:55, 103.99s/it, Model Initialization complete! ]
[INFO] Processing interrupted
[INFO] Prompt executed in 111.84 seconds
[INFO] got prompt
[INFO] 0 models unloaded.
[INFO] Model MiniMaxH3 prepared for dynamic VRAM loading. 19995MB Staged. 0 patches attached. Force pre-loaded 210 weights: 1175 KB.
5%|██▍ | 1/20 [00:52<16:35, 52.40s/it, Model Initialization complete! ]
[INFO] Interrupting prompt b2660a48-5c30-46a1-99c5-0af359020c97
10%|████████▎ | 2/20 [01:45<15:49, 52.75s/it]
[INFO] Processing interrupted
[INFO] Prompt executed in 105.44 seconds
[INFO] got prompt
[INFO] Model MiniMaxH3 prepared for dynamic VRAM loading. 19995MB Staged. 0 patches attached. Force pre-loaded 210 weights: 1175 KB.
5%|██▍ | 1/20 [00:52<16:42, 52.78s/it, Model Initialization complete! ]
[INFO] Interrupting prompt 06d12746-c1cb-4458-9db8-c31df24bb511
5%|██▍ | 1/20 [01:00<19:04, 60.21s/it, Model Initialization complete! ]
[INFO] Processing interrupted
[INFO] Prompt executed in 100.82 seconds
```

### Other

_No response_