202611021124_ZIT基础文生图
2026-11-02 11:24:20 · sess-20261102-112139 · 工作流 ComfyUI
success冷启omnixpu加速B58016G
总耗时
102.01s
采样耗时
7s
采样步数
8 / 8
每步采样
0.95s/it
模型载入
11.20GB
单次峰值
6.69GB
生成物

模型加载模型文件 / 加载节点来自工作流结构(API 图);载入 MB / 耗时 / MB/s / 路径来源来自执行日志
| 角色 | 模型文件 | 加载节点 | 载入 MB | 耗时 s | MB/s | 路径来源 |
|---|---|---|---|---|---|---|
| 底模 base | ZIT_REDZimageTurbo2.0-INT8-Convrot-Comfyui_redzit222026HD.safetensors | #72 DiffusionModelLoaderKJ Lumina2 | 6850.2 | 25.337 | 270 | policy 相邻 |
| 文本编码器 text_encoder | qwen_3_4b_int8_convrot.safetensors | #73 CLIPLoader ZImageTEModel_ | 4303.5 | 0.077 | 55890 | policy 相邻 |
| VAE vae | ae.safetensors | #74 VAELoaderKJ AutoencodingEngine | 319.8 | 1.542 | 207 | 按角色匹配 |
共 3 行(图中每一处模型引用各占一行,不合并):3 行在执行日志里有载入数据,0 行只有结构引用。
MB/s 用于区分「内存回迁」(GB/s 级)与「磁盘重读」(数百 MB/s)。text_encoder 在某些栈上 device_to=cpu,它的加载耗时接近 0,不代表省了显存搬运。
MB/s 用于区分「内存回迁」(GB/s 级)与「磁盘重读」(数百 MB/s)。text_encoder 在某些栈上 device_to=cpu,它的加载耗时接近 0,不代表省了显存搬运。
工作流结构UI 图含 bypass 节点,API 图是真正执行的子集导出
| 画布节点数(UI 图) | 13 |
|---|---|
| 实际执行节点数(API 图) | 13 |
| 节点状态分布(UI 图 mode) | active 13 · muted 0 · bypassed 0 |
| 被 bypass 的节点 | 无 |
| 被 mute 的节点 | 无 |
节点详情 13 个节点 · 5 个插件
| 节点名 | 插件名 | 插件版本 |
|---|---|---|
CLIPLoader 加载CLIP text_encoders | comfy-core | 0.3.73 |
CLIPTextEncode CLIP文本编码 - 正面文本 | comfy-core | 0.9.2 |
ConditioningZeroOut 条件零化 | comfy-core | 0.11.1 |
KSampler K采样器 | comfy-core | 0.9.2 |
ModelSamplingAuraFlow 采样算法(AuraFlow) | comfy-core | 0.28.0 |
SaveImage 保存图像 | comfy-core | 0.3.48 |
VAEDecodeTiled VAE解码(分块) | comfy-core | 0.12.3 |
LogCollection | ComfyUI-FullLog | — |
DiffusionModelLoaderKJ | comfyui-kjnodes | bc8e4ce4254b… |
VAELoaderKJ | comfyui-kjnodes | 50a0837f9aea… |
CR Aspect Ratio | ComfyUI_Comfyroll_CustomNodes | d78b780ae43f… |
RAMCleanup | comfyui_memory_cleanup | 58de13a6090e… |
VRAMCleanup | comfyui_memory_cleanup | 1.1.3 |
运行环境
| 系统 | Windows-11-10.0.22000-SP0 · AMD64 · build 10.0.22000 |
|---|---|
| 物理内存 | 15.9 GB (16G) |
| GPU | Intel(R) Arc(TM) B580 Graphics (0xE20B, 11946.2 MB, driver 1.17.39894+4) |
| 平台 | Intel(R) oneAPI Unified Runtime over Level-Zero V2 |
| 计算单元 | 160 |
| Python | 3.13.11 |
| torch | 2.15.0+xpu (xpu available: True, devices: 1) |
| ComfyUI | 0.43.1 preview=Latent2RGB |
| 插件数 | 67 |
| 加速栈 | |
| 加速方案 | omnixpu加速 (日志中 omni_xpu_kernel 0.2.0b3+torch215.bmg 生效,且出现 [OmniXPU] 插件标记) |
| OmniXPU 内核 | omni_xpu_kernel 0.2.0b3+torch215.bmg 已生效:sdp, norm, rotary, linear_fp8, int8, layout |
| OmniXPU 插件 | ComfyUI-OmniXPU 15 个 adapter applied、4 个 skipped |
| aimdo | comfy-aimdo 0.6.0 未启用 (disabled by env) |
| kitchen XPU | comfy-kitchen 0.2.40 xpu 后端可用 (adaln, apply_rope, apply_rope1, apply_rope1_… 共 50 项) |
加速方案 = 内核生效行 + 插件
[OmniXPU] 标记,两处都在工作流执行日志里才判「omnixpu加速」;aimdo / kitchen 是可选栈,不影响方案判定,也不进文件名。运行时环境变量 17 项
| 变量名 | 值 |
|---|---|
OCL_ICD_FILENAMES | ;F:\Intel-oneAPI\compiler\latest\bin\intelocl64.dll |
OMNIXPU_DEBUG | 0 |
OMNIXPU_ENABLE | 1 |
ONEAPI_CLANG_VERSION | 22 |
ONEAPI_DEVICE_SELECTOR | level_zero:0 |
ONEAPI_PATH | F:\Intel-oneAPI |
ONEAPI_ROOT | F:\Intel-oneAPI |
SYCL_CACHE_MAX_SIZE | 8192 |
SYCL_CACHE_THRESHOLD | 7 |
SYCL_DEVICE_FILTER | level_zero |
SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS | 0 |
SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY | 1 |
SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS | 1 |
UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS | 1 |
UR_L0_USE_IMMEDIATE_COMMANDLISTS | 1 |
ZE_DEVICE_SLEEP | 0 |
ZE_ENABLE_PCI_ID_DEVICE_ORDER | 1 |
时间轴
| 时刻 | 相对起点 s | 增量 s | 事件 |
|---|---|---|---|
| 11:24:20.294 | 0.07 | 0.07 | 下发存储策略 · ae.safetensors |
| 11:24:21.812 | 1.59 | 1.52 | 下发存储策略 · qwen_3_4b_int8_convrot.safetensors |
| 11:24:24.133 | 3.91 | 2.32 | 请求加载 ZImageTEModel_ |
| 11:24:24.210 | 3.99 | 0.08 | qwen_3_4b_int8_convrot.safetensors 就绪(4303.51 MB) |
| 11:24:51.509 | 31.29 | 27.30 | 下发存储策略 · ZIT_REDZimageTurbo2.0-INT8-Convrot-Comfyui_redzit222026HD.safetensors |
| 11:24:51.766 | 31.54 | 0.25 | 请求加载 Lumina2 |
| 11:25:17.103 | 56.88 | 25.34 | ZIT_REDZimageTurbo2.0-INT8-Convrot-Comfyui_redzit222026HD.safetensors 就绪(6850.19 MB) |
| 11:25:34.173 | 73.95 | 17.07 | 请求加载 AutoencodingEngine |
| 11:25:35.715 | 75.49 | 1.54 | ae.safetensors 就绪(319.75 MB) |
| 11:26:02.248 | 102.02 | 26.53 | 执行完成 |
原始日志内嵌,不依赖 user/comfyui.log
来源
user/comfyui.log · 32 行 · 日志自报耗时 102.01 s原始日志全文
[2026-11-02 11:24:20.224] got prompt[2026-11-02 11:24:20.294] Model storage policy: fast_disk=False paths=['F:\\ComfyUI-aki-v3\\ComfyUI\\models\\vae\\ae.safetensors'][2026-11-02 11:24:20.301] Using pytorch attention in VAE[2026-11-02 11:24:20.302] Using pytorch attention in VAE[2026-11-02 11:24:21.194] VAE load device: xpu:0, offload device: cpu, dtype: torch.float32[2026-11-02 11:24:21.421] Found quantization metadata version 1[2026-11-02 11:24:21.422] Using MixedPrecisionOps for text encoder[2026-11-02 11:24:21.812] Model storage policy: fast_disk=False paths=['F:\\ComfyUI-aki-v3\\ComfyUI\\models\\text_encoders\\qwen_3_4b_int8_convrot.safetensors'][2026-11-02 11:24:24.133] Requested to load ZImageTEModel_[2026-11-02 11:24:24.210] loaded completely; 4303.51 MB loaded, full load: True[2026-11-02 11:24:24.229] CLIP/text encoder model load device: cpu, offload device: cpu, current: cpu, dtype: torch.float16[2026-11-02 11:24:51.423] Found quantization metadata version 1[2026-11-02 11:24:51.425] Detected mixed precision quantization[2026-11-02 11:24:51.447] Using mixed precision operations[2026-11-02 11:24:51.449] Native ops: int8_tensorwise , emulated ops: float8_e5m2, nvfp4, w6a8_int8, asym_w4a8_int8, float8_e4m3fn, convrot_w4a4, mxfp8[2026-11-02 11:24:51.498] model weight dtype torch.bfloat16, manual cast: torch.bfloat16[2026-11-02 11:24:51.507] model_type FLOW[2026-11-02 11:24:51.509] Model storage policy: fast_disk=False paths=['F:\\ComfyUI-aki-v3\\ComfyUI\\models\\diffusion_models\\ZIT_REDZimageTurbo2.0-INT8-Convrot-Comfyui_redzit222026HD.safetensors'][2026-11-02 11:24:51.766] Requested to load Lumina2[2026-11-02 11:25:17.103] loaded completely; 10019.03 MB usable, 6850.19 MB loaded, full load: True[2026-11-02 11:25:20.118] [OmniXPU] norm first use: RMSNorm shape=torch.Size([1, 403, 2560])[2026-11-02 11:25:20.518] [OmniXPU] attention CUTE #1: heads=30 seq=416 dtype=torch.bfloat16[2026-11-02 11:25:20.715] [OmniXPU] attention CUTE #2: heads=30 seq=416 dtype=torch.bfloat16[2026-11-02 11:25:20.843] [OmniXPU] attention CUTE #3: heads=30 seq=4032 dtype=torch.bfloat16[2026-11-02 11:25:27.070] 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:07<00:00, 1.41it/s]100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:07<00:00, 1.05it/s][2026-11-02 11:25:34.162] VRAM清理完成 [卸载模型: False, 清空缓存: True][2026-11-02 11:25:34.173] Requested to load AutoencodingEngine[2026-11-02 11:25:35.715] loaded completely; 3194.16 MB usable, 319.75 MB loaded, full load: True[2026-11-02 11:25:55.167] RAM清理完成 [71.5% → 46.0%, 释放: 4153MB][2026-11-02 11:26:02.248] Prompt executed in 102.01 seconds