Immich Sizing Guide: What a 20,000 Photo Google Takeout Import Really Costs in Disk, RAM and Hours

Description and use

By following the sizing guide, users can provision appropriate disk, RAM, and CPU resources, set critical day‑one settings, and anticipate a 24‑48 hour processing window for a 20,000‑item library, avoiding costly re‑runs and ensuring a smooth Immich deployment.

Problem

Users importing large Google Takeout libraries into Immich face uncertainty about required disk space, RAM, CPU, and processing time, and risk costly re‑runs if settings change.

Outcome

By following the sizing guide, users can provision appropriate disk, RAM, and CPU resources, set critical day‑one settings, and anticipate a 24‑48 hour processing window for a 20,000‑item library, avoiding costly re‑runs and ensuring a smooth Immich deployment.

Workflow

• 1. Estimate original Takeout size (60–120 GB for 20,000 items).
• 2. Allocate disk space: 1.35–1.6× original size (e.g., 200 GB for a 90 GB Takeout).
• 3. Provision RAM: 6 GB for full stack with ML, 4 GB if ML disabled.
• 4. Allocate at least 4 CPU cores; adjust concurrency settings in Immich Administration → Job Settings.
• 5. Set day‑one settings: storage template, machine‑learning model, transcoding policy, and keep them unchanged to avoid full re‑runs.
• 6. Import via immich‑go or CLI; monitor upload speed.
• 7. After upload, monitor background jobs: thumbnail generation, video transcoding, smart search, face detection.
• 8. Verify disk usage with du -sh on upload/, thumbs/, encoded-video/ and Postgres volume.
• 9. If RAM OOM occurs, increase RAM, reduce ML model size, or limit concurrent ML jobs.
• 10. Once all queues finish, library becomes searchable and fully functional.

Evidence

Budget roughly 1.35 to 1.6 times your raw photo size on disk, 6 GB of RAM for the full stack with machine learning enabled, 4 CPU cores, and a full weekend of background processing before a 20,000 item library is fully searchable.

Source

The import itself is the fast part. Thumbnail generation, video transcoding and machine learning jobs are what actually keep the server busy, and three settings decided in the first hour, the storage template, the machine learning model, and the transcoding policy, are the ones that force a full re‑run of every job if you change your mind in month two.

Source

Plan for 6 GB of system RAM for the full stack with machine learning enabled, and 4 GB if you disable smart search.

Source