Skip to content

NVIDIA GPU Acceleration

If your computer has an NVIDIA graphics card, PhotoMapAI can use it to speed up the initial indexing of your photo collection by roughly 10x. (After the index is built, the GPU offers only a modest speedup for text- and image-similarity searches.)

You do not need to install CUDA

This is the part that surprises most people: you do not need to download or install the CUDA Toolkit from NVIDIA. PhotoMapAI installs its own copy of PyTorch, and the GPU build of PyTorch already bundles every CUDA runtime library it needs (the CUDA runtime, cuDNN, cuBLAS, and so on). This is true on both Windows and Linux — there is no difference between the two platforms here.

The one thing PhotoMapAI cannot bundle is the NVIDIA graphics driver, because that talks directly to your hardware. So the only GPU prerequisite is a reasonably recent NVIDIA driver. If you can play modern games or already use your card for anything graphics-intensive, you almost certainly have it.

macOS: CUDA is neither available nor required. PhotoMapAI automatically uses the built-in GPU acceleration on Apple M-series chips. You can ignore this page.

Check whether your system is ready

Open a command window — PowerShell or Command Prompt on Windows, Terminal on Linux — and run:

nvidia-smi

If it prints a table like this, you're ready — there is nothing else to install:

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.230.02             Driver Version: 535.230.02   CUDA Version: 13.0     |
|-----------------------------------------+----------------------+----------------------+
(more information follows)

Two things to notice:

  • nvidia-smi ships with the driver, not with the CUDA Toolkit. The fact that it runs at all means the driver is installed and working — which is exactly (and only) what PhotoMapAI needs.
  • The "CUDA Version" in the top-right is the highest CUDA version your driver can support — not a toolkit you have to install. Make sure it reads 12.x or newer (current PyTorch builds target CUDA 12 and 13). Anything in that range works.

If nvidia-smi runs and shows your card, skip the rest of this page.

If nvidia-smi is not found

That means the NVIDIA driver isn't installed (you still do not need the CUDA Toolkit). Install just the driver:

Windows

Most Windows machines with an NVIDIA card already have the driver via Windows Update or GeForce Experience. If not, download the latest Game Ready or Studio driver from nvidia.com/drivers, install it, and run nvidia-smi again.

Linux (Ubuntu / Mint and similar)

A fresh Linux install boots with the open-source nouveau driver, so nvidia-smi won't exist until you install NVIDIA's proprietary driver. On Ubuntu/Mint, use any one of:

# easiest: let the distro pick the right driver
sudo ubuntu-drivers autoinstall

# or install a specific version
sudo apt install nvidia-driver-550        # use the version offered by your distro

Mint and Ubuntu also expose this through the "Additional Drivers" / "Driver Manager" GUI. Reboot, then run nvidia-smi to confirm. Again — this installs the driver, not the CUDA Toolkit.

Telling PhotoMapAI to use the GPU

  • Desktop installer: nothing to do. On first launch PhotoMapAI auto-detects the GPU and installs the GPU build of PyTorch for you. If you add a card later (or want to force a re-detect), launch with the --gpu flag — see GPU acceleration in the main installation guide.
  • PyPI / uv: uv tool install photomapai --torch-backend auto picks the GPU or CPU build of PyTorch automatically based on what nvidia-smi reports.

To confirm GPU support is active, watch for a console message about GPU acceleration when PhotoMapAI starts up.