Does CompreFace 1.2.0 works with NVIDIA Tesla T4?

[davimury]

I’m currently evaluating CompreFace v1.2.0 for deployment on a system equipped with an NVIDIA Tesla T4 GPU. Before proceeding with the setup, I’d like to confirm whether this version of CompreFace is fully compatible with the Tesla T4 in terms of:

GPU acceleration support

CUDA and cuDNN compatibility

Any known issues or performance limitations

Questions
Has anyone successfully run CompreFace 1.2.0 on a Tesla T4 with GPU acceleration enabled?

Are there any known compatibility or performance issues with this GPU?

Is there any additional configuration required for optimal performance on this hardware?

Any insights, documentation references, or confirmation from the maintainers would be greatly appreciated. Thanks!

[gribouk]

Yes. Make this amendment to docker-compose and it will use GPU:

compreface-core:
restart: always
image: exadel/compreface-core:1.2.0-arcface-r100-gpu
container_name: "compreface-core"
runtime: nvidia
environment:
- ML_PORT=3000
- IMG_LENGTH_LIMIT=${max_detect_size}
- UWSGI_PROCESSES=${uwsgi_processes:-2}
- UWSGI_THREADS=${uwsgi_threads:-1}
- NVIDIA_VISIBLE_DEVICES=all
healthcheck:
test: curl --fail http://localhost:3000/healthcheck || exit 1
interval: 10s
retries: 0
start_period: 0s
timeout: 1s

Welcome!