What is an AI NAS?

Three very different things share the label

1. AI-first appliances

Boxes designed and marketed around AI. UGREEN’s iDX6011 and iDX6011 Pro put Intel Core Ultra chips and soldered LPDDR5X memory in a six-bay chassis; the Pro is sold on a “96 TOPS AI Compute” figure. Zettlab’s D4 and D6 use a Rockchip RK3588 with a 6 TOPS NPU, and Zettlab’s own comparison table lists their chatbot feature as “AI Q&A (via PC Computing Power)”: the heavy lifting runs on your computer, not the NAS. The Minisforum N5 MAX belongs here too: a Ryzen AI Max+ 395 with 64 GB of 256-bit memory, the one box in this class fast enough to make a 70B model usable (3.5 to 5.8 tok/s for Llama 3.3 70B at Ollama's default 4k context).

2. Powerful NAS units you run AI on

General-purpose x86 NAS hardware where you install the AI yourself (Ollama and Open WebUI for chat, Immich for photos, Frigate for cameras). The Minisforum N5 Pro and N5 AIR pair Radeon graphics with two DDR5 slots; UGREEN’s DXP8800 Plus and DXP6800 Ultra use Intel Core chips; QNAP’s TVS-AIh1688ATX is a 16-bay Core Ultra machine. None of these ships with an AI experience in the box, but most of them run models faster than the appliances, because what matters is memory, and they have more of it to give. An N5 Pro runs an 8B model at 4.9 to 8.2 tok/s as shipped.

3. Mainstream NAS with AI features

Ordinary NAS units whose software has gained AI features. Synology’s DSM 7.4 adds AI Console and AI-powered search in Synology Drive. Read the fine print: Synology states its GPU-powered local models (including the embedding model behind Drive’s AI search) are available only on Synology systems with a GPU, and AI Console’s supported providers are cloud services: Amazon Bedrock, Azure OpenAI, Baidu AI Cloud, Google AI Studio, Google Vertex AI, OpenAI, and other providers with OpenAI-compatible APIs. On a DS925+ that means your text goes to a cloud provider unless you point AI Console at an OpenAI-compatible server you run yourself.

Why TOPS mislead

TOPS (trillions of operations per second) measure how much arithmetic an NPU or GPU can do. A chatbot is not limited by arithmetic. To write each word, the processor reads every active weight of the model out of RAM: 4.9 GB for Llama 3.1 8B at the default quantization. Ten words a second means moving about 50 GB a second. So the number that predicts chat speed is memory bandwidth, and Ollama does not use NPUs at all (its hardware documentation lists NVIDIA CUDA, AMD ROCm, Apple Metal and Vulkan). The table makes the point with the boxes’ own figures.

NPUs do have a job. Immich can use a Rockchip NPU for face and object search, and Frigate can run camera detection on Intel and Rockchip NPUs. If those are your AI workloads, the NPU matters. For a chatbot it sits idle.

Local or cloud: know which one you are buying

• Minisforum N5 series: general x86 hardware. Anything you install (Ollama, Immich, Frigate) runs locally.
• UGREEN iDX6011 / Pro: sold on on-board AI compute (up to 96 TOPS on the Pro). Not listed on UGREEN’s Canadian stores when we checked.
• Zettlab D4 / D6: AI Q&A runs “via PC Computing Power”, per Zettlab. Not shipped to Canada.
• Synology (DSM 7.4): AI Console connects to cloud providers; local GPU models only on GPU-equipped Synology systems.
• Any x86 NAS with Docker: you can run Ollama locally; how fast depends on its memory, as the capability matrix shows.

Who should buy which

You want a private ChatGPT at home and nothing else changes: keep your NAS, and run Ollama on the fastest memory you can afford. The Minisforum N5 AIR is the cheapest box here with a Radeon iGPU and two DDR5 slots. You need a new NAS anyway: buy one where AI is a bonus, such as the Minisforum N5 Pro (fill both RAM slots) or UGREEN’s DXP8800 Plus. You want big models (32B to 70B): only the N5 MAX’s 256-bit memory makes those usable. You mostly want AI photo search: any x86 NAS with 8 GB of RAM runs Immich; a Rockchip NPU or Intel Iris Xe/Arc graphics speeds it up. Our picks, with Canadian prices, are in the best AI NAS in Canada; to check your own box, use the AI NAS calculator.

Frequently asked questions