How to run face indexing on gpu?

[vladfi1]

I'm running on a Linux server with an nvidia gpu. I believe I have enabled every nivida/gpu-related setting possible in the compose.yaml:

• PHOTOPRISM_INIT: "https tensorflow-gpu"
• NVIDIA_VISIBLE_DEVICES: "all"
• Uncommented the deploy section under "NVIDIA GPU Hardware Acceleration"
• Installed the nvidia container toolkit. nvidia-smi inside docker shows my gpu.

However when I run photoprism faces index I see only cpu usage, no gpu usage. Is there anything else I'd need to do to run face embedding on gpu, or is this not really supported?

Note: from my experience using tensorflow in python, you typically need to install the appropriate cuda/cudnn libraries, so my hypothesis is that without these installed tensorflow can't use the gpu.

[HomidWay]

Bump.

with PHOTOPRISM_INIT: gpu tensorflow-gpu

I managed to get

GPU detected: nvidia amdgpu

Installing GPU drivers for AMD64...

NVIDIA Container Toolkit must be installed: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html

Installing AMD VA-API GPU Drivers...

and on start of indexing

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR

W0000 00:00:1770201004.926226    2043 gpu_device.cc:2344] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.

Skipping registering GPU devices...

nvidia-smi detects that photoprism does actually detects GPU but it's not actually used.

Wed Feb  4 13:50:14 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.82.09              Driver Version: 580.95.05      CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA P106-100                Off |   00000000:12:00.0 Off |                  N/A |
|  0%   37C    P8              5W /  120W |      67MiB /   6144MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A          143781      C   /opt/photoprism/bin/photoprism           64MiB |
+-----------------------------------------------------------------------------------------+ 

Would be nice to have solution