Can My NAS Run AI?

Which NAS Can Run Which AI Job?

Each NAS at the memory it ships with. Language model columns use Ollama's download size, Ollama's default context of 4096 tokens, one request, NTK's 15% context allowance and 2 GB kept free for the NAS. Immich, Synology Photos and Whisper use their makers' published requirements.

"Needs a memory upgrade" means the job fits within the maker's maximum memory but not in what the NAS ships with. Synology Photos is built into DSM, so it applies to Synology NAS only. No column is a speed estimate.

How the Checker Works

Language models. Ollama publishes each model's download size at a stated quantisation; the whole model has to sit in memory. Ollama's FAQ says it uses a context window of 4,096 tokens by default. (Ollama) Ollama's documentation says a larger context length increases the memory needed to run a model, and that tasks needing a large context, such as agents and coding tools, should be set to at least 64,000 tokens. (Ollama) Ollama's FAQ says parallel requests to one model multiply its context size, and the memory needed scales with the number of parallel requests times the context length. (Ollama) Ollama publishes no per model figure for that, so NTK adds 15% of the model size for every 4,096 tokens held in memory, and keeps 2 GB free for the NAS. Both are NTK planning figures; the AI Hardware Requirements Calculator lets you change them or enter what ollama ps shows.

Apps. Immich's requirements are a minimum of 6 GB of RAM (8 GB recommended) and a minimum of 2 CPU cores (4 recommended); with only 4 GB of RAM it can run with machine learning turned off. (Immich) Immich runs on amd64 and arm64 processors; since version 3 its machine learning container on amd64 needs the x86-64-v2 microarchitecture level or newer, and Immich needs Docker with the Docker Compose plugin. (Immich) Immich documents GPU acceleration for its machine learning through ARM NN (Mali), CUDA (NVIDIA), ROCm (AMD), OpenVINO (Intel GPUs such as Iris Xe and Arc) and RKNN (Rockchip), and calls the feature experimental. (Immich) PhotoPrism asks for at least 2 CPU cores and 3 GB of memory, with the amount of RAM matching the number of CPU cores, and at least 4 GB of swap. (PhotoPrism) The whisper.cpp README gives the memory each Whisper model uses: tiny about 273 MB, base about 388 MB, small about 852 MB, medium about 2.1 GB and large about 3.9 GB. (whisper.cpp (ggml-org)) Synology lists the NAS models whose Synology Photos supports face and object recognition. (Synology) For models that ship with 2 GB of memory, Synology says to upgrade to at least 4 GB to get object recognition. (Synology)

Processors, GPUs and NPUs. Ollama's hardware page lists NVIDIA GPUs (compute capability 5.0 and newer), AMD GPUs through ROCm, Apple GPUs through Metal, and further GPUs on Windows and Linux through Vulkan, which on most Linux distributions needs extra components installed. It does not list NPUs. (Ollama) NTK found no NAS maker documenting Ollama GPU acceleration on its NAS, so the checker treats Ollama on a NAS as running on the processor and gives no speed figure. A dedicated computer with a GPU on Ollama's list is the other route when AI is the main job; the NAS then keeps storing the files.

Remote access. Using a home AI from outside depends on your NBN connection, mainly upload speed and CGNAT. The NBN Remote Access Checker and the NBN Plan Finder cover both.

Buying. Your first step for a faulty product is usually the retailer that sold it. The ACCC explains your rights under the Australian Consumer Law at accc.gov.au.

Assumptions NTK made

Method last reviewed: 2026-10-14. Model sizes and app requirements last checked at the source: 2026-10-18. NAS specifications: each maker's page, date in the table.

Frequently Asked Questions