Bulk imports queue metadata extraction, thumbnail generation, machine-learning jobs and video processing across many assets. The same server can feel light after processing finishes but become CPU-, RAM- or I/O-bound during the initial ingest.
For users importing tens of thousands of existing photos and videos instead of starting with an empty library.Wymagania sprzętowe serwera Immich: procesor, pamięć RAM, pamięć masowa i uczenie maszynowe
Poznaj wymagania serwera Immich dotyczące pamięci RAM, procesora, przechowywania zdjęć, PostgreSQL, uczenia maszynowego, transkodowania wideo oraz wyboru sprzętu ZimaOS.
Immich server requirements at a glance
Size Immich for imports and background processing, not only idle browsing. Thumbnail generation, video transcoding, Smart Search and facial recognition create the main CPU, memory and storage-I/O demand.
- CPU
- Immich requires at least 2 CPU cores and recommends 4. More CPU headroom helps during large imports, thumbnail generation, software video transcoding and machine-learning jobs.
- RAM
- Immich requires at least 6 GB RAM and recommends 8 GB. Its current requirements page says a 4 GB system can run with machine learning disabled, but 4 GB is below the documented minimum for the normal configuration.
- Storage
- Keep the PostgreSQL database on local SSD storage and never on a network share. Size the original photo/video library separately and allow roughly 10–20% extra capacity on average for generated thumbnails and transcoded video.
- Network
- Immich does not publish one universal network minimum. Local backup speed, remote uploads, library imports and external-library access depend on file sizes, concurrent users and the storage path.
- GPU and machine learning
- A GPU is optional. Immich supports experimental hardware transcoding and experimental hardware-accelerated machine learning on supported backends. Hardware capability alone does not prove the ZimaOS containers can use it; drivers, device exposure and the selected Immich backend must be verified.
- Best Zima starting point
- ZimaBoard 2 832 meets Immich's published 8 GB recommended memory figure and 4-core count, but it should be paired with appropriate local SSD and media storage. Choose ZimaBoard 2 1664 for more memory headroom, or ZimaCube 2 when multi-drive growth and heavier concurrent processing become priorities.
From official requirements to the right setup
Immich sizing starts with its official 6 GB RAM and 2-core minimum, then changes with library size, initial-import workload, machine learning, video transcoding, database placement and the other services sharing ZimaOS.
-
Official requirements
Use Immich's current official baseline first: at least 6 GB RAM and 2 CPU cores, with 8 GB RAM and 4 cores recommended for a smoother experience.
-
Confirm your needs
Separate database and generated data from bulk originals. Keep DB_DATA_LOCATION on local SSD storage, never a network share, and use a filesystem that supports the ownership and permission semantics Immich requires.
-
Leave room to grow
Decide whether Smart Search, facial recognition and video transcoding will run locally. CPU-based processing is supported; hardware transcoding and hardware-accelerated machine learning are separate optional features with different backends and must be validated inside the deployed containers.
-
Run it on ZimaOS
Install Immich from the ZimaOS App Store, map persistent storage, verify database and library paths, then test uploads, background jobs, search, facial recognition and representative videos before scaling the library.
Check every playback client
- Total photo and video library size
- Approximate asset count and expected yearly growth
- Initial bulk import or continuous mobile backup
- Smart Search and facial recognition enabled or disabled
- Video quantity, codec, resolution and transcoding policy
- Local SSD location and compatible filesystem for PostgreSQL
- Original-library location, redundancy and backup plan
- Other ZimaOS apps, users and workloads running at the same time
Official minimum requirements
Immich publishes explicit minimum and recommended CPU and memory requirements. Storage capacity is workload-dependent, but database placement and filesystem behavior have important official requirements.
Use 6 GB RAM and 2 CPU cores as the current documented minimum, with 8 GB and 4 cores recommended. Keep PostgreSQL on local SSD, never a network share, use a compatible filesystem, and budget additional space for thumbnails and transcoded video.
| Requirement | Official minimum | What this supports |
|---|---|---|
| CPU | Minimum 2 cores; recommended 4 cores | On amd64, the Immich v3 machine-learning container requires the x86-64-v2 microarchitecture level or newer. This requirement is specifically documented for the machine-learning container. |
| System memory | Minimum 6 GB RAM; recommended 8 GB | Immich says a 4 GB system can run with machine learning disabled, but its documented minimum remains 6 GB. If Docker resource limits are used, PostgreSQL itself requires at least 2 GB RAM. |
| PostgreSQL storage | Local SSD strongly recommended; network shares are not allowed | Immich says PostgreSQL database files are typically about 1–3 GB. DB_DATA_LOCATION should ideally use local SSD storage and must never be placed on a network share. |
| Photo and video library | Depends on originals and growth plan | Plan capacity from current originals, future uploads, retention, redundancy and backups. The library itself can become the dominant storage requirement. |
| Generated media overhead | Approximately 10–20% additional library size on average | Immich says thumbnails and transcoded video can increase the size of the photo library by about 10–20% on average. |
| Filesystem and platform | Linux or another 64-bit *nix OS recommended | Immich recommends a Unix-compatible filesystem such as EXT4, ZFS, APFS, BTRFS or XFS with support for user/group ownership and permissions. |
When to upgrade your hardware
Upgrade Immich hardware for a defined workload change: a much larger import, more simultaneous background jobs, heavier machine learning or a storage plan that has outgrown a compact two-drive layout.
A much larger initial photo import
Machine learning becomes a daily feature
Storage growth exceeds a simple two-drive setup
Smart Search and facial recognition run through the machine-learning service and can add significant processing demand during initial or repeated jobs. Hardware-accelerated machine learning is optional and experimental; use it only after the exact backend and container device path are verified.
For users who rely on semantic search, face grouping or frequent ingestion of new assets.Originals, thumbnails, encoded video, database growth, redundancy and backups all consume separate capacity. Move to a multi-drive platform when the library needs easier expansion, larger redundancy groups or more independent storage tiers.
For families, photographers and long-term archives that are growing beyond a small personal photo server.Plan hardware growth with confidence
Immich grows in three different directions: original-media capacity, database and generated-data I/O, and background compute. Plan each one separately so a larger library does not force unnecessary upgrades everywhere.
Keep PostgreSQL on local SSD
Immich says PostgreSQL database files are typically about 1–3 GB and that stable, fast database access is critical. Keep DB_DATA_LOCATION on local SSD storage rather than any network share.
Use SSD or NVMe for PostgreSQL and other latency-sensitive application data even when original photos and videos live on larger-capacity HDD storage.Budget for generated media
Thumbnail generation and transcoded video add storage beyond the originals. Immich estimates about 10–20% additional library size on average, although the real percentage depends on the media mix and transcoding settings.
Leave free capacity for generated data instead of filling the storage pool to the size of the original library alone.Separate irreplaceable originals from availability
Storage redundancy can improve availability after a drive failure, but it does not replace a separate backup. Immich's own database backup protects metadata and user data; original images and videos also need an independent file-level backup.
Keep at least one separate copy of irreplaceable originals and make sure the Immich database and application paths needed for recovery are also protected.Leave compute headroom for background jobs
Immich can look idle after the library is processed, but new uploads can trigger thumbnails, machine learning and video encoding again. Other ZimaOS apps compete for the same CPU, RAM and storage I/O.
Choose additional memory and CPU headroom for the peak combination of Immich jobs and the other services that actually run at the same time.Can it run on ZimaOS?
Immich is available in the ZimaOS App Store. The important setup work is persistent storage placement: keep PostgreSQL on suitable local storage, map the photo library deliberately, and treat optional acceleration as unverified until it works inside the running containers.
Install Immich from the ZimaOS App Store
Install Immich from ZimaOS, launch the web interface, create the first account and complete the initial application setup before importing a large library.
Open Immich in the ZimaOS App StorePlan persistent photo and database paths
Before a large import, confirm where uploads, originals and application data will live. Keep PostgreSQL on local SSD storage and make sure its filesystem supports the ownership and permission semantics required by Immich.
Read the ZimaOS Immich guideValidate optional acceleration before relying on it
For video transcoding or machine learning, do not infer application support only from the CPU or GPU specification. Confirm the required host driver, container device exposure, Immich backend selection and actual device use during a real job.
Read Immich hardware acceleration guidanceChoose Zima hardware for your Immich workload
Start with library size, initial-import behavior, machine learning, video processing and storage topology. CPU and RAM determine processing headroom, while the larger purchasing decision is often whether a compact two-drive platform is enough or a multi-drive NAS is needed.
Will the library fit comfortably on a compact two-drive plan with moderate import workloads?
Yes — compact storage and moderate processing are enough
Meet Immich's memory and core-count guidance, add local SSD storage for PostgreSQL/application data, and size the attached media storage for originals, generated media and backups.
- Personal library, 8 GB recommended baseline, added SSD and media drivesZimaBoard 2 832
- Larger imports, more apps or extra memory headroom in the same compact platformZimaBoard 2 1664
No — multi-drive growth or heavier concurrent processing is expected
Use ZimaCube 2 when integrated multi-drive capacity, more application storage, heavier imports, more simultaneous services or a clearly validated dedicated-GPU path are part of the requirement.
- Growing multi-drive photo library with the official 8 GB recommended memory baselineZimaCube 2 Standard
- Large imports, more simultaneous services and additional CPU/RAM/network headroomZimaCube 2 Pro
- Dedicated NVIDIA acceleration plus local AI, creator or heavy VM workloadsZimaCube 2 Creator Pack
This is a workload guide, not a guaranteed import-time, asset-count or GPU-acceleration benchmark. Results depend on asset count, RAW/video mix, ML models, job concurrency, transcoding settings, storage layout, ZimaOS version, container configuration and the exact acceleration backend.
| Zima hardware | Best for | Example workload | Core configuration | Recommended boundary | Next step |
|---|---|---|---|---|---|
| ZimaBoard 2 832 | A compact personal Immich server whose 8 GB RAM and 4-core Intel N150 meet Immich's published recommended memory and core-count figures, with added SSD/HDD storage for the real library. | Personal or small-family mobile backup, ordinary browsing, moderate initial imports and light additional ZimaOS services. |
|
Matching the published 8 GB and 4-core figures is not a performance guarantee. Large first imports, CPU-based machine learning and other simultaneous services can use most of the available headroom. | Get Now |
| ZimaBoard 2 1664 | A compact Immich server with more memory headroom for larger imports, machine-learning jobs and other ZimaOS applications. | Family photo backup, larger import batches, Smart Search and facial recognition with additional containers running on the same server. |
|
Choose it for memory headroom and compact expansion, not because it doubles transcoding or machine-learning performance over the 832. | Get Now |
| ZimaCube 2 Standard | A growing Immich library that benefits from integrated multi-drive capacity and a stronger 12th-gen Intel CPU while remaining at Immich's 8 GB recommended memory figure. | Multi-user photo backup, larger HDD libraries, regular imports, Smart Search, facial recognition and other moderate self-hosted services. |
|
The 8 GB memory configuration matches Immich's recommendation but leaves less headroom for a large first import, machine-learning jobs and several other memory-intensive services running at the same time. | Get Now |
| ZimaCube 2 Pro | Large Immich libraries, heavier imports and multiple always-on services that benefit from more CPU, memory, faster SSD expansion and 10GbE connectivity. | Large family or creator libraries, sustained imports, background machine learning, multi-service ZimaOS deployments and faster local ingest workflows. |
|
More CPU, RAM and network headroom does not remove the need for correct PostgreSQL placement, independent backups or end-to-end verification of optional hardware acceleration. | Get Now |
| ZimaCube 2 Creator Pack | Immich combined with a genuine dedicated-GPU requirement, very large initial processing workloads, local AI, creator applications or heavy virtual machines. | A mixed photo and compute server where Immich is one part of a larger GPU-accelerated, AI or creator workflow. |
|
Usually unnecessary for a normal personal photo library. Choose it when the dedicated GPU, 64 GB memory, AI, CUDA or heavy VM/creator workload is itself part of the requirement. | Get Now |
What the Press Says
Highlights from trusted reviewers worldwide.
“ZimaCube 2: Not just another NAS, tested with 25TB storage, local AI agents, 4K transcoding, and real homelab workflows.”Read full review
“The ZimaBoard 2 is a compact x86 server board that can be turned into a mini NAS, home server, media box, or self-hosting hub.”Read full review
“ZimaCube 2: A modern, high-performance NAS with plenty of room to grow—built for users who want more than basic storage.”Read full review
“Coverage focused on ZimaCube 2's open hardware foundation, no monthly fee, and self-hosting flexibility.”Read full review
Loved by the Community
Stories and reviews from people who build with Zima every day.
★★★★★
Zima Blade Little yet Powerful
Maybe I am not digital natives but I live with PCs since 12 years old in 1984 when IBM PC clone come to my home. Many years have passed and many operating system I've tried. For me Zima blade and CasaOS was a quantum leap for home PC enthusiast and server lab machine to make me stay curious and relevant for this era.
★★★★★
Very good!!
I use ZimaCube Pro as 5th Proxmox cluster node. It runs several VMs and containers, including a VM with GPU passthrough to run a self-hosted LLM. A specific LXC container runs a Samba server for NAS capabilities using four of six RAID 6 SATA HDDs with ZFS.
★★★★★
Great innovation for mini server!
It is very useful and makes a powerful mini server for many purposes, including university and college students in engineering and electronics. Thank you so much for making this server.
★★★★★
Avaliação ZimaBoard 2
Construí um servidor de uso pessoal. O desempenho está muito bom e funciona perfeitamente onde quer que eu esteja. A surpresa é não dependermos de grandes estruturas para termos nosso próprio servidor de dados. Como iniciante, estou gostando bastante do ZimaOS, pois ele é simples e eficiente.
Frequently asked questions
These answers cover sizing, storage, machine learning and migration questions that are not fully addressed in the sections above.
How much RAM does Immich need?
Immich currently documents 6 GB RAM as the minimum and recommends 8 GB. Its requirements page says a 4 GB system can run with machine learning disabled, but 4 GB is below the documented normal minimum. Add more RAM when large imports, machine learning and other ZimaOS services will overlap.
Is 8 GB RAM enough for Immich?
Yes for many personal and family deployments because 8 GB is Immich's current recommended memory figure. It is not a promise of identical performance on every CPU: very large first imports, more job concurrency or several memory-heavy services can justify 16 GB or more.
Does Immich need an SSD?
Immich does not require the entire photo library to live on SSD, but PostgreSQL should ideally be on local SSD storage and must never be placed on a network share. SSD or NVMe is also useful for latency-sensitive application data, while larger HDD storage can hold the bulk photo/video library.
How much extra storage does Immich use beyond my original photos?
Immich says thumbnail generation and transcoded video can increase the photo library size by roughly 10–20% on average. Actual overhead varies with the proportion of videos, thumbnail settings, transcoding policy and future library growth.
Does Immich need a GPU for facial recognition and Smart Search?
No. Smart Search and facial recognition can run without a GPU. Immich also offers experimental hardware-accelerated machine learning through backends including CUDA, ROCm, OpenVINO, ARM NN and RKNN. Use acceleration only after confirming that the exact device, drivers and container configuration are supported.
Can ZimaBoard 2 run Immich?
Yes. ZimaBoard 2 uses a 4-core Intel N150, and the 832 model has 8 GB RAM, so it matches Immich's published recommended core-count and memory figures. That does not guarantee a particular import or machine-learning speed. Add suitable local SSD storage for PostgreSQL/application data and separate storage for a growing library.
When should I choose ZimaCube 2 instead of ZimaBoard 2 for Immich?
Choose ZimaCube 2 when storage topology or concurrent workload becomes the bigger requirement: a growing multi-drive archive, larger redundancy plan, heavier imports or several always-on services. Pro adds more CPU, RAM and network headroom. Creator Pack makes sense only when a dedicated GPU, 64 GB memory or another AI/creator/VM workload is genuinely required.
Can I migrate an existing photo library into Immich on ZimaOS?
Yes. For a pre-existing archive that you want Immich to index without moving into its managed upload library, Immich officially supports External Libraries. Map the existing path into the container, create the External Library in Immich and scan it. Preserve an independent backup first; if migrating an existing Immich instance, also preserve the database and the required application data paths.
What sources and further reading informed this Immich hardware guide?
Immich's current requirements page is the authority for the 6 GB minimum, 8 GB recommendation, 2/4-core guidance, PostgreSQL placement and filesystem rules. The SSD Nodes article reflects the current 2026 figures and adds practical sizing context. The Industrial Monitor Direct and Medium articles contain older 4 GB/6 GB requirement figures, so they are useful only as historical low-resource or homelab examples. The GitHub #23684 discussion is useful as a large-library community signal, but its opening hardware proposal is LLM-generated and a maintainer/community reply explicitly flags outdated advice such as Typesense, so it must not be treated as an official sizing specification.
