Skip to content

System Requirements

Hardware

CASTLE is designed for NVIDIA GPUs with CUDA 12.x support.

  • Minimum VRAM: [HUMAN TO CONFIRM]
  • Recommended VRAM: [HUMAN TO CONFIRM]

CPU-Only Mode

CASTLE can run without a GPU, but performance will be significantly slower — particularly for tracking (SAM + DeAOT) and feature extraction (DINOv2/v3). GPU-accelerated clustering via cuml will be unavailable.

RAM

  • Minimum: [HUMAN TO CONFIRM]
  • Recommended (large datasets): [HUMAN TO CONFIRM]

Storage

  • ~4 GB for model checkpoints
  • Additional space for project data (videos, extracted features, results)

Software

Requirement Version
Python 3.10+
CUDA 12.x (12.6 tested in Docker)
OS Linux (primary), macOS (limited GPU support), Windows (limited)
Docker (optional) 20.10+ with NVIDIA Container Toolkit

Tested Configurations

Under Construction

The following table will be populated with confirmed working configurations.

Configuration GPU VRAM RAM OS Status
[HUMAN TO CONFIRM]

Notes

  • Linux is the primary development and testing platform
  • macOS has limited support — Apple Silicon GPUs are not supported (no CUDA); CPU-only mode works
  • Windows should work but is less tested; WSL2 with GPU passthrough is recommended