[HW Accel Support]: How to verify OpenVINO is being used for Intel GPU accelerated Semantic Search?

[ammgws]

Describe the problem you are having

HW: Intel Arc A310 passed through to docker as /dev/dri/renderD128 (already being used for ffmpeg hwaccel in frigate)

I also want to use the GPU for hw accel for semantic search.

First I enabled semantic search with the "small" model, confirmed it was working, then afterwards changed to the "large" model and restarted Frigate. I want to make sure it is using the GPU as expected, and not falling back to the CPU. The docs say that "OpenVINO will automatically be detected and used for Semantic Search in the default Frigate image." but I cannot find anything in the logs to confirm that.

Is there any way to verify OpenVINO is being used for Intel GPU accelerated Semantic Search?

Version

0.15.0-cea210d

Frigate config file

semantic_search:
  enabled: True
  model_size: large
  reindex: false

logger:
    default: info
    logs:
      frigate.embeddings: debug

docker-compose file or Docker CLI command

services:
  frigate:
    devices:
      - /dev/apex_0:/dev/apex_0
      - /dev/dri/renderD128:/dev/dri/renderD128

Relevant Frigate log output

info
2025-03-14 15:17:52
frigate.embeddings.embeddings
Indexing tracked object embeddings...

info
2025-03-14 15:18:32
frigate.embeddings.embeddings
Embedded 1158 thumbnails and 0 descriptions in 40.5 seconds

debug
2025-03-14 16:32:42
frigate.embeddings.functions.onnx
models are already downloaded for jinaai/jina-clip-v1

Relevant go2rtc log output

-

FFprobe output from your camera

-

Install method

Proxmox via Docker

Object Detector

Coral

Network connection

Wired

Camera make and model

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[hawkeye217]

You could run intel_gpu_top during reindexing to verify. Also, you could compare the reindexing time of the small vs. large models one right after the other. That isn't quite an apples to apples comparison but should give you a good idea. Generally though, if there are no errors in the logs, the GPU is being used when you've set large in your config.

[ammgws]

Thanks for the hint - I can see it running now:

CPU: Embedded 1158 thumbnails and 0 descriptions in 40.5 seconds
GPU: Embedded 1163 thumbnails and 0 descriptions in 19.8 seconds

[hawkeye217]

Looks good!