6GB max RAM · 2 bays · 2.5GbE
AINAS Score 20
No verified price

Traditional NASUGREEN
Why Traditional NAS: Rockchip ARM 8-core (A72+A53, 6 TOPS NPU used for photo recognition), 4GB LPDDR4X fixed memory and no Docker support, so traditional.
Updated
Can the UGREEN DH2300 run local AI?
The UGREEN DH2300 is a Traditional NAS built for file storage and backup rather than local AI. With Rockchip 8-core ARM (A72+A53, RK3576) and a 4GB memory ceiling, it can run light services but not a useful local language model.
We list it so you can compare it with AI-capable systems at a similar price.
The DH2300 suits a buyer who wants a simple two-bay home NAS for storing files and organising photos, not one for running local AI. It has an 8-core Rockchip ARM CPU, a 6 TOPS NPU and 4GB of soldered LPDDR4X memory. The NPU drives AI Album, which the vendor confirms for faces, scenes, places and duplicates. Photo search by people and places, plus OCR text search in images, is vendor confirmed as limited. It has 1GbE networking, runs UGOS and carries a 2 year warranty. The 4GB configuration also includes 32GB of eMMC. The AINAS Score is 21, which reflects a system built for storage and photos rather than model work.
The one thing that rules the DH2300 out is that it cannot run local AI software at all. It has no Docker support and no virtual machines, and the vendor confirms no local LLM. Memory is fixed at 4GB, with no GPU expansion, no M.2 slots and no PCIe slot. Anyone who wants a local LLM, private document search or any self-hosted AI tool is affected. They should look for a system with Docker support, memory that is large and upgradeable, and a GPU path such as a PCIe slot or OCuLink. Faster networking than 1GbE would also help with larger libraries.
Amazon UK
4GB
£159.99£219.99
Verified
In stock
Buy on AmazonProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.co.uk or Amazon.com, as applicable, at the time of purchase will apply to the purchase of this product.
Each capability shows how it was confirmed: Verified by AINAS, Vendor confirmed, Community confirmed or Unverified.
| Capability | Supported | Evidence | Note |
|---|---|---|---|
| Local LLM | No | Vendor confirmed | ARM, 4GB RAM, no Docker. |
| Ollama | No | Vendor confirmed | No Docker support. |
| Open WebUI | No | Vendor confirmed | No Docker support. |
| RAG | No | Unverified | No LLM or knowledge base features documented. |
| AI search | Limited | Vendor confirmed | Photo search by people/places and OCR text search in images. |
| GPU expansion | No | Vendor confirmed | No expansion. |
| Hardware acceleration | Limited | Vendor confirmed | 6 TOPS NPU for AI photo recognition. |
| AI photos | Yes | Vendor confirmed | AI Album: faces, scenes, places, duplicates. |
| AI video | No | Unverified | No video AI documented. |
| Surveillance AI | No | Vendor confirmed | Surveillance Center KB: DH2300 not currently supported. |
Calculated from the DH2300's memory and each model's published size at each quantization. Rows marked Estimated are calculations, not benchmarks. Check any model.
| Model | Best quantization | Runs? | Memory needed | Expected performance | Verification |
|---|---|---|---|---|---|
| Llama 3.2 3BLlama, 3.21B | None fits | No | 11GB at Q4 | Not measured | Estimated |
| Llama 3.1 8BLlama, 8.03B | None fits | No | 14GB at Q4 | Not measured | Estimated |
| Qwen3 8BQwen, 8.19B | None fits | No | 15GB at Q4 | Not measured | Estimated |
| Gemma 3 12BGemma, 11.8B | None fits | No | 18GB at Q4 | Not measured | Estimated |
| Qwen3 14BQwen, 14.8B | None fits | No | 20GB at Q4 | Not measured | Estimated |
| Phi-4 14BPhi, 14.7B | None fits | No | 19GB at Q4 | Not measured | Estimated |
| Mistral Small 3 24BMistral, 23.6B | None fits | No | 25GB at Q4 | Not measured | Estimated |
| Gemma 3 27BGemma, 27B | None fits | No | 29GB at Q4 | Not measured | Estimated |
| Qwen3 30B A3BQwen, 30.5B | None fits | No | 31GB at Q4 | Not measured | Estimated |
| Qwen3 32BQwen, 32.8B | None fits | No | 32GB at Q4 | Not measured | Estimated |
| DeepSeek-R1 32B (Qwen distill)DeepSeek, 32.8B | None fits | No | 32GB at Q4 | Not measured | Estimated |
| Llama 3.3 70BLlama, 70.6B | None fits | No | 60GB at Q4 | Not measured | Estimated |
| Llama 4 Scout 17Bx16ELlama, 109B | None fits | No | 89GB at Q4 | Not measured | Estimated |
| Qwen2.5 72BQwen, 72.7B | None fits | No | 65GB at Q4 | Not measured | Estimated |
| Mistral Large 2 123BMistral, 123B | None fits | No | 96GB at Q4 | Not measured | Estimated |
One verdict per workload, from the DH2300's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
The DH2300 is a poor fit for running a local LLM. The vendor confirms no local LLM support, citing ARM, 4GB of RAM and no Docker, so nothing on the system can run model software.
Private RAG is a poor fit on the DH2300. Ollama and Open WebUI are both vendor confirmed as unsupported without Docker, and RAG itself is unconfirmed because no knowledge base features are documented.
AI search is a poor fit, though a small part works. The vendor confirms limited support, covering photo search by people and places and OCR text search in images, not searching documents.
AI photos is rated poor overall, even though the feature itself is present. AI Album is vendor confirmed for faces, scenes, places and duplicates, using the 6 TOPS NPU, but 4GB of memory limits the system.
Surveillance is a poor fit. The vendor's Surveillance Center knowledge base states the DH2300 is not currently supported, so camera recording and AI detection cannot be used.
Home lab use is a poor fit on the DH2300. It has no Docker and no virtual machine support, a single 1GbE port, no M.2 slots and no PCIe slot for expansion.
Business use is a poor fit. Memory is soldered at 4GB without ECC, networking is one 1GbE port, and there is no Docker or virtual machine support for running business applications.
Creative AI is a poor fit. GPU expansion is vendor confirmed as none, hardware acceleration is limited to the 6 TOPS NPU for photo recognition, and the local LLM row says no.
The DH2300 ships with 4GB of LPDDR4X (soldered) that cannot be upgraded, so the configuration you buy is the one you keep. For local AI that matters more than anything else on the spec sheet: model size is capped by memory, not storage.
No. The DH2300 has no PCIe slot or OCuLink port for a graphics card. Inference runs on the CPU, with ARM Mali-G52 MC3 handling media tasks, so keep to small models.
The 6 TOPS NPU helps with photo recognition, transcription and vendor AI features. Most open-source runtimes such as Ollama do not use NAS NPUs yet, so do not count it towards language model speed.
Not well. A useful private RAG setup needs more memory or compute than the DH2300 offers.
Networking is 1GbE. Fine for one or two users; indexing a large archive over the network will be slower.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| 4GB | 4GB | None | Diskless, 4GB LPDDR4X, 32GB eMMC. UK bundles with UPS or M.2 enclosure. |
The closest UGREEN model by bay count and AINAS Score. Only the specifications that differ are listed.
The DH2300 scores best for Business (24/100) and Local LLM (21/100). It suits small offices that need shared storage with private AI tools and room to grow.
It is a weaker fit for anyone who wants to run models larger than about 8B parameters and for buyers who expect to add a graphics card later.
Other Traditional NAS systems closest in price.
No. Llama 3.2 3B is the smallest model listed on this page, and it is Estimated as not fitting at Q4 or Q8. It needs 2GB at Q4 and 3.4GB at Q8, against 4GB of soldered memory. More importantly, the DH2300 has no Docker support, so nothing on it can run model software. The vendor confirms no local LLM.
No. The DH2300 has 4GB of LPDDR4X that is soldered to the board, and the maximum is also 4GB. There is no expansion slot. If you need more memory for any workload, you must choose a different system at purchase, because this one cannot be changed later.
No. The vendor states the DH2300 does not support Docker or virtual machines. That rules out containers such as Ollama and Open WebUI, both vendor confirmed as unsupported. It runs UGOS, and its software is limited to the vendor's own apps, such as AI Album for photos.
Yes. The DH2300 has a 6 TOPS NPU. The vendor says it is used for AI photo recognition, and hardware acceleration is marked as limited. AI Album covers faces, scenes, places and duplicates. The NPU is not confirmed for running language models or any other AI workload.
No. The DH2300 has 0 M.2 slots, so there is no NVMe cache or NVMe storage pool. Storage comes from its 2 drive bays. The configuration sold is diskless, with 4GB of memory and 32GB of eMMC, so you supply your own drives.
No. The DH2300 has 4GB of soldered LPDDR4X with no ECC, and the memory cannot be upgraded. That suits a two-bay home system for files and photos, where AI Album uses the 6 TOPS NPU for faces, scenes, places and duplicates. Buyers who need error-correcting memory for business records, or more than 4GB for any self-hosted tool, should look for a system that lists ECC memory and memory that can be expanded.
No. Ollama support is documented by the manufacturer. See the capability table above for the source.
No. Open WebUI support is documented by the manufacturer. See the capability table above for the source.
Yes. AI photos support is documented by the manufacturer. See the capability table above for the source.
No. Surveillance AI support is documented by the manufacturer. See the capability table above for the source.
Partly. Hardware acceleration support is documented by the manufacturer. See the capability table above for the source.
It has 1GbE built in. The fastest port is 1GbE.
Every field records its source. Specifications were last checked on . Spot an error? Send a correction with a manufacturer link.