Immich – Recommended Hardware Specs
Immich Hardware Docker GPU Self‑hosted
These recommendations are for running Immich with Docker, Redis, Postgres (pgvector), and optional GPU‑accelerated machine learning. They are meant as practical, real‑world guidelines rather than strict minimums.
1. Overview – Three Tiers
| Tier | Use Case | CPU | RAM | Storage | GPU |
|---|---|---|---|---|---|
| Minimum | Single user, small library (< 20k photos) | 4 cores / 8 threads | 8–12 GB | SSD, 250 GB+ | Optional |
| Recommended | 1–5 users, medium library (20k–100k) | 6–8 cores / 12–16 threads | 16–32 GB | NVMe SSD, 1 TB+ | Dedicated GPU (CUDA) |
| High‑End | Multi‑user, large library (100k+) | 12+ cores / 24+ threads | 32–64 GB | NVMe SSD + HDD tiering | Dedicated GPU (CUDA, 8 GB+ VRAM) |
2. CPU Recommendations
2.1 Minimum (works, but not ideal)
- 4 cores / 8 threads
- Examples: older Intel i5/i7, Ryzen 3/5
- Suitable for: single user, small library, no heavy video transcoding
2.2 Recommended (sweet spot)
- 6–8 cores / 12–16 threads
- Examples: modern Intel i5/i7, Ryzen 5/7
- Suitable for: multiple users, active uploads, smooth browsing while indexing
2.3 High‑End (for heavy parallel workloads)
- 12+ cores / 24+ threads
- Examples: Intel i7/i9, Ryzen 7/9, Xeon/Epyc small configs
- Suitable for: very large libraries, many concurrent users, running other services alongside Immich
With GPU acceleration enabled, CPU is mostly responsible for metadata extraction, database queries, and API handling. More cores improve responsiveness under load, but Immich does not require server‑grade CPUs to run well.
3. RAM Recommendations
- Minimum: 8 GB (only for very small setups)
- Recommended: 16–32 GB
- High‑End: 32–64 GB for large libraries and multi‑service hosts
Immich itself is not extremely memory‑hungry, but:
- Postgres benefits from extra RAM for caching
- Redis keeps data in memory
- Docker overhead adds up with more containers
4. Storage Recommendations
4.1 Type
- Strongly recommended: SSD or NVMe for the database and thumbnails
- Optional: HDD for cold storage of original media (if mounted separately)
4.2 Capacity
- Minimum: 250 GB total
- Recommended: 1 TB+ (especially for photo/video libraries)
- High‑End: NVMe for DB + cache, large HDD/array for originals
Immich stores thumbnails, metadata, and database files in addition to your original media. Plan for at least 20–30% extra space beyond your raw photo/video size.
5. GPU Recommendations (Optional but Highly Recommended)
5.1 Why a GPU helps
- Accelerates face recognition
- Speeds up embeddings and search indexing
- Reduces CPU load during ML tasks
- Makes large library indexing much faster
5.2 Minimum GPU
- NVIDIA GPU with CUDA support
- 4 GB VRAM (absolute minimum)
- Examples: older GTX 10‑series, RTX 20‑series entry models
5.3 Recommended GPU
- 6–8 GB VRAM
- Examples: RTX 2060/3060/4060 or better
- Suitable for: medium to large libraries, fast ML indexing
5.4 High‑End GPU
- 8–12+ GB VRAM
- Examples: RTX 3070/4070 and above, workstation cards
- Suitable for: very large libraries, frequent re‑indexing, multi‑user ML workloads
Immich can run without a GPU, but indexing and ML features will be significantly slower. For serious use, a CUDA‑capable NVIDIA GPU is strongly recommended.
6. Network & Platform
- OS: Modern Linux distribution (Ubuntu, Debian, etc.)
- Network: Gigabit LAN recommended for fast uploads and browsing
- Access: Docker + Docker Compose for deployment
7. Quick Reference – Recommended Baseline
For a smooth, future‑proof Immich experience:
- CPU: 6–8 cores / 12–16 threads
- RAM: 16–32 GB
- Storage: 1 TB+ SSD/NVMe
- GPU: NVIDIA CUDA card with 6–8 GB VRAM
- Network: Gigabit LAN
These specs are intentionally conservative and aimed at users who want a responsive, reliable Immich instance without having to constantly fight performance bottlenecks.
没有找到评论