收起左侧

相册AI识别调用模型触发GPU hang机 ,log提示clWaitForEvents, error code: -1

2
回复
212
查看
[ 复制链接 ]

0

主题

4

回帖

0

牛值

江湖小虾

2026-6-18 15:21:10 显示全部楼层 阅读模式

如题,AI功能无法调用GPU,自动降级为CPU,一分钟只能识别几张图片。

设备环境:X86物理机,系统版本1.1.3107,CPU J4125 + 8G内存

BUG现象:相册AI识别无法调用GUP

出现频率:必现

日志:

[2026-06-18 14:04:35] face_det INFO: cpu_usage=100.0, using existing affinity on 4 CPU(s)

[2026-06-18 14:04:35] face_det INFO: num_threads_onnx:1
[2026-06-18 14:04:35] face_det INFO: model_format:OPENVINO
[2026-06-18 14:04:35] face_det INFO: det_model:
[2026-06-18 14:04:35] face_det INFO: det_openvino_model:/vol1/@sysappmeta/ai-manager/models/face-model-1.0.1/det.xml
[2026-06-18 14:04:35] face_det INFO: det_openvino_weights:/vol1/@sysappmeta/ai-manager/models/face-model-1.0.1/det.bin
[2026-06-18 14:04:35] face_det INFO: openvino_cache_dir:/vol1/@sysappmeta/ai-manager/var/com.trim.face_det/openvino-cache
[2026-06-18 14:04:35] face_det INFO: det_rknn_model:
[2026-06-18 14:04:35] face_det INFO: rec_model:
[2026-06-18 14:04:35] face_det INFO: rec_openvino_model:/vol1/@sysappmeta/ai-manager/models/face-model-1.0.1/rec.xml
[2026-06-18 14:04:35] face_det INFO: rec_openvino_weights:/vol1/@sysappmeta/ai-manager/models/face-model-1.0.1/rec.bin
[2026-06-18 14:04:35] face_det INFO: rec_rknn_model:
[2026-06-18 14:04:35] face_det INFO: confidence:0.7
[2026-06-18 14:04:35] face_det INFO: resolution:1600
Traceback (most recent call last):
File "/workspace/face_det.py", line 157, in
File "/workspace/face_det.py", line 44, in init
File "/usr/trim/lib/ai_manager/com.trim.face_det/cp311/lib/insight_openvino.py", line 71, in init
File "/vol1/@sysappmeta/ai-manager/python/3_11_x86/lib/python3.11/site-packages/openvino/runtime/ie_api.py", line 543, in compile_model
super().compile_model(model, device_name, {} if config is None else config),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Exception from src/inference/src/cpp/core.cpp:104:
Exception from src/inference/src/dev/plugin.cpp:53:
Exception from src/plugins/intel_gpu/src/runtime/ocl/ocl_stream.cpp:429:
[GPU] clWaitForEvents, error code: -14

[2026-06-18 14:04:46] face_det INFO: cpu_usage=40.0, limited service to 1/4 CPU(s): [0]
[2026-06-18 14:04:46] face_det INFO: num_threads_onnx:1
[2026-06-18 14:04:46] face_det INFO: model_format:ONNX
[2026-06-18 14:04:46] face_det INFO: det_model:/vol1/@sysappmeta/ai-manager/models/face-model-1.0.1/det.onnx
[2026-06-18 14:04:46] face_det INFO: det_openvino_model:
[2026-06-18 14:04:46] face_det INFO: det_openvino_weights:
[2026-06-18 14:04:46] face_det INFO: openvino_cache_dir:
[2026-06-18 14:04:46] face_det INFO: det_rknn_model:
[2026-06-18 14:04:46] face_det INFO: rec_model:/vol1/@sysappmeta/ai-manager/models/face-model-1.0.1/rec.onnx
[2026-06-18 14:04:46] face_det INFO: rec_openvino_model:
[2026-06-18 14:04:46] face_det INFO: rec_openvino_weights:
[2026-06-18 14:04:46] face_det INFO: rec_rknn_model:
[2026-06-18 14:04:46] face_det INFO: confidence:0.7
[2026-06-18 14:04:46] face_det INFO: resolution:1600
Available Providers: TensorrtExecutionProvider, CUDAExecutionProvider, CPUExecutionProvider
Selected provider: CPUExecutionProvider
Available Providers: TensorrtExecutionProvider, CUDAExecutionProvider, CPUExecutionProvider
Selected provider: CPUExecutionProvider
launch_model finish
[2026-06-18 14:04:57] face_det INFO: HTTP RPC socket ready at unix:///run/com.trim.face_det_internal.sock

收藏
送赞
分享

0

主题

4

回帖

0

牛值

江湖小虾

2026-6-18 15:45:05 楼主 显示全部楼层

root@MyNAS:~# root@MyNAS:~# dmesg | tail -5

[246700.183517] i915 0000:00:02.0: [drm] python[935156] context reset due to GPU hang

[246700.186848] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:e757fefe, in python [935156]

[247137.178227] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out

发现执行AI任务时GPU hang机了

406

主题

1万

回帖

0

牛值

管理员

fnOS1.0上线纪念勋章

2026-6-24 15:24:22 显示全部楼层

当前相册 AI 的部分人脸识别任务在 GPU 初始化失败或不支持时,会自动回退到 CPU 执行,所以后续日志里出现 Selected provider: CPUExecutionProvider 是预期的降级保护逻辑,由于 CPU 模式性能会明显低于 GPU,因此会出现识别速度较慢的情况。我们后续看能否进行优化

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则