[Detector Support]: Python segmentation fault on NVIDIA Tesla P4

[ConstBur]

Describe the problem you are having

When trying to use the ONNX detector with a Frigate+ YOLOv9 model on a NVIDIA Tesla P4, I get a segmentation fault error, see the logs.

Version

0.16.2

Frigate config file

mqtt:
  enabled: false

detect:
  enabled: true

detectors:
  onnx_0:
    type: onnx

model:
  path: plus://<model_id>

cameras:
  Cabano: # <------ Name the camera
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac
      inputs:
        - path: rtsp://<camera_url>
          input_args: preset-rtsp-generic
          roles:
            - detect
        - path:
            rtsp://<camera_url>
          input_args: preset-rtsp-generic
          roles:
            - record
version: 0.16-0

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: "frigate"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    environment:
      - "PLUS_API_KEY=<api_key>"
      - "TZ=America/Toronto"
      - "NVIDIA_VISIBLE_DEVICES=all"
      - "NVIDIA_DRIVER_CAPABILITIES=all"
    image: "ghcr.io/blakeblackshear/frigate:0.16.2-tensorrt"
    ports:
      - "5000:5000/tcp"
      - "8554:8554/tcp"
      - "8555:8555/tcp"
      - "8555:8555/udp"
    privileged: true
    shm_size: "512mb"
    restart: "unless-stopped"
    volumes:
      - "/etc/localtime:/etc/localtime:ro"
      - "/home/gituser/frigate:/config"
      - "/home/gituser/storage:/media/frigate"

Relevant Frigate log output

2025-10-20 20:39:36.571425861  2025-10-20 20:39:36.571340717 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 439, index: 3, mask: {8, 9, }, error code: 22
error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.
2025-10-20 20:39:36.571653254  [2025-10-20 20:39:36] frigate.detectors.plugins.onnx INFO    : ONNX: loaded onnxruntime module
2025-10-20 20:39:36.572041960  [2025-10-20 20:39:36] frigate.detectors.plugins.onnx INFO    : ONNX: loading /config/model_cache/521342cac553e84ab37e51dd787552ce
2025-10-20 20:39:36.578383355  2025-10-20 20:39:36.572442109 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 438, index: 2, mask: {6, 7, }, error code: 22
error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.
2025-10-20 20:39:36.578385523  2025-10-20 20:39:36.575395593 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 440, index: 4, mask: {10, 11, }, error code: 2
2 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.
2025-10-20 20:39:36.580443987  2025-10-20 20:39:36.578505738 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 441, index: 5, mask: {12, }, error code: 22 er
ror msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.
2025-10-20 20:39:36.628389930  2025-10-20 20:39:36.625413502 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 442, index: 6, mask: {13, }, error code: 22 er
ror msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.
2025-10-20 20:39:36.637384608  2025-10-20 20:39:36.634403040 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 443, index: 7, mask: {14, }, error code: 22 er
ror msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.
2025-10-20 20:39:36.639977699  [2025-10-20 20:39:36] frigate.app                    INFO    : Output process started: 448
2025-10-20 20:39:36.934004975  Fatal Python error: Segmentation fault
2025-10-20 20:39:36.934007950
2025-10-20 20:39:36.934047991  Current thread 0x00007d0670d8f040 (most recent call first):
2025-10-20 20:39:36.934114173    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 537 in _create_inference_session
2025-10-20 20:39:36.934177397    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 465 in __init__
2025-10-20 20:39:36.934235074    File "/opt/frigate/frigate/detectors/plugins/onnx.py", line 53 in __init__
2025-10-20 20:39:36.934297870    File "/opt/frigate/frigate/detectors/__init__.py", line 18 in create_detector
2025-10-20 20:39:36.934394858    File "/opt/frigate/frigate/object_detection/base.py", line 57 in __init__
2025-10-20 20:39:36.934464547    File "/opt/frigate/frigate/object_detection/base.py", line 112 in run_detector
2025-10-20 20:39:36.934533623    File "/usr/lib/python3.11/multiprocessing/process.py", line 108 in run
2025-10-20 20:39:36.934599882    File "/opt/frigate/frigate/util/process.py", line 41 in run_wrapper
2025-10-20 20:39:36.934669336    File "/usr/lib/python3.11/multiprocessing/process.py", line 314 in _bootstrap
2025-10-20 20:39:36.934740717    File "/usr/lib/python3.11/multiprocessing/popen_fork.py", line 71 in _launch
2025-10-20 20:39:36.934814976    File "/usr/lib/python3.11/multiprocessing/popen_fork.py", line 19 in __init__
2025-10-20 20:39:36.934889166    File "/usr/lib/python3.11/multiprocessing/context.py", line 281 in _Popen
2025-10-20 20:39:36.934959653    File "/usr/lib/python3.11/multiprocessing/context.py", line 224 in _Popen
2025-10-20 20:39:36.935027670    File "/usr/lib/python3.11/multiprocessing/process.py", line 121 in start
2025-10-20 20:39:36.935099180    File "/opt/frigate/frigate/util/process.py", line 30 in start
2025-10-20 20:39:36.935170237    File "/opt/frigate/frigate/object_detection/base.py", line 195 in start_or_restart
2025-10-20 20:39:36.935240360    File "/opt/frigate/frigate/object_detection/base.py", line 163 in __init__
2025-10-20 20:39:36.935312694    File "/opt/frigate/frigate/app.py", line 391 in start_detectors
2025-10-20 20:39:36.935390115    File "/opt/frigate/frigate/app.py", line 648 in start
2025-10-20 20:39:36.935469372    File "/opt/frigate/frigate/__main__.py", line 104 in main
2025-10-20 20:39:36.935542358    File "/opt/frigate/frigate/__main__.py", line 108 in <module>
2025-10-20 20:39:36.935613431    File "<frozen runpy>", line 88 in _run_code
2025-10-20 20:39:36.935688415    File "<frozen runpy>", line 198 in _run_module_as_main

Install method

Docker Compose

Object Detector

TensorRT

Screenshots of the Frigate UI's System metrics pages

Not available because Frigate crashes at startup.

Any other information that may be helpful

The same config does work on an NVIDIA RTX 3070, so maybe there's a compute capability limitation?

[inkeep-ai-assistant[bot]]

The segmentation fault you're experiencing with the NVIDIA Tesla P4 appears to be related to compute capability limitations. The TensorRT detector has specific minimum hardware requirements that may not be met by your GPU(1).

For Nvidia GPU support, the documentation specifies that TensorRT requires GPUs with a Compute Capability of 5.0 or greater, which generally correlates to Maxwell-era GPUs or newer(1). The Tesla P4, being a Pascal-based GPU, should technically meet this requirement, but older Tesla cards can have compatibility issues.

Since you're using the ONNX detector with a Frigate+ model, you have a few options:

1. Switch to the standard Frigate image: Instead of using ghcr.io/blakeblackshear/frigate:0.16.2-tensorrt, try the default image which will automatically use OpenVINO for Intel hardware or fall back to CPU processing(2).

2. Use explicit ONNX configuration: The ONNX detector automatically detects and uses TensorRT in the -tensorrt Frigate image when a supported ONNX model is configured(2). However, if your Tesla P4 isn't fully compatible, you might need to force CPU execution.

3. Consider hardware limitations: The documentation notes that 12.x series CUDA libraries require minimum driver version >=545 and GPU Compute Capability of 5.0 or greater(3). While the Tesla P4 should meet these requirements, older Tesla cards may have additional compatibility constraints.

Based on similar segmentation fault issues reported with TensorRT configurations, the problem often relates to model configuration mismatches or hardware compatibility(4). Since your configuration works on an RTX 3070, this suggests the Tesla P4 may not be fully supported by the current TensorRT implementation.

I recommend trying the default Frigate image first, which should automatically fall back to a compatible detector for your hardware.

📚 Sources:

1. Supported Hardware
2. Supported Hardware
3. Recommended hardware
4. [Detector Support]: Fatal Python error: Segmentation fault

---

Have another question? Just tag @inkeep.