64GB max RAM · 12 bays · 2 x 10GbE
AINAS Score 47
No verified price

AI CapableUGREEN
Why AI Capable: AMD Ryzen Embedded R2514 (4C/8T Zen+, no NPU) with Radeon Vega 8, no GPU expansion path and 64GB max DDR4: older Ryzen embedded with no expansion is capable at best.
Updated
Can the UGREEN DXP4800 GT run local AI?
Only lightly. Expect 32B class models comfortably, and 70B class models at 4-bit with little headroom (Estimated) on the UGREEN DXP4800 GT: AMD Ryzen Embedded R2514 does the work with up to 64GB of memory, which is fine for experiments rather than a daily assistant.
Without any GPU option, the processor carries the whole inference load.
Local LLM support is documented by the manufacturer.
The DXP4800 GT suits a buyer who wants a four-bay home or small-office NAS that can also run local AI on the CPU. It has 2 x 10GbE, two M.2 NVMe slots and takes up to 64GB of DDR4 SO-DIMM 2666, which is upgradeable and supports ECC when you fit ECC modules. It ships with 8GB. Docker and virtual machines are supported on UGOS Pro. UGREEN documents Ollama and Open WebUI guides for it. Llama 3.3 70B fits at Q4 (Estimated) once the memory is upgraded. The AINAS Score is 46.
The main limit is the GPU path. The DXP4800 GT has no PCIe slot, no Thunderbolt, no USB4 and no OCuLink, so a graphics card cannot be added later. It also has no NPU. Local LLM support is limited to CPU-only inference, according to UGREEN. Anyone planning heavy model use, creative AI work or faster inference should look for a system with a PCIe slot or OCuLink for an external GPU, or an NPU with a published TOPS figure. Buyers who mainly want storage, photo albums, surveillance and small models will not be affected.
Amazon UK
8GB
£449.99£529.99
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 | Limited | Vendor confirmed | CPU-only inference via Ollama in Docker per UGREEN KB; no NPU. |
| Ollama | Yes | Vendor confirmed | UGREEN KB Docker Compose guide for Ollama/Open WebUI on UGOS Pro. |
| Open WebUI | Yes | Vendor confirmed | Official UGOS Pro Open WebUI guide. |
| RAG | Limited | Vendor confirmed | RAG via Open WebUI document chat only. |
| AI search | Limited | Vendor confirmed | Photos description and OCR search via Model Manager; document semantic search applicability per model not stated. |
| GPU expansion | No | Vendor confirmed | No PCIe slot, Thunderbolt or OCuLink. |
| Hardware acceleration | Limited | Unverified | Radeon Vega 8 iGPU; UGREEN does not document hardware transcoding or AI acceleration for GT; no NPU. |
| AI photos | Yes | Vendor confirmed | AI photo album: faces, places, objects, duplicates. |
| AI video | Limited | Vendor confirmed | Surveillance Center local AI detection (people, vehicle, package, pet); features vary by model. |
| Surveillance AI | Yes | Vendor confirmed | Surveillance Center supports DXP series. |
Calculated from the DXP4800 GT'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 | Q8 | Yes | 13GB | Not measured | Estimated |
| Llama 3.1 8BLlama, 8.03B | Q8 | Yes | 19GB | Not measured | Estimated |
| Qwen3 8BQwen, 8.19B | Q8 | Yes | 19GB | Not measured | Estimated |
| Gemma 3 12BGemma, 11.8B | Q8 | Yes | 24GB | Not measured | Estimated |
| Qwen3 14BQwen, 14.8B | Q8 | Yes | 28GB | Not measured | Estimated |
| Phi-4 14BPhi, 14.7B | Q8 | Yes | 28GB | Not measured | Estimated |
| Mistral Small 3 24BMistral, 23.6B | Q8 | Yes | 38GB | Not measured | Estimated |
| Gemma 3 27BGemma, 27B | Q8 | Yes | 44GB | Not measured | Estimated |
| Qwen3 30B A3BQwen, 30.5B | Q8 | Yes | 47GB | Not measured | Estimated |
| Qwen3 32BQwen, 32.8B | Q8 | Yes | 50GB | Not measured | Estimated |
| DeepSeek-R1 32B (Qwen distill)DeepSeek, 32.8B | Q8 | Yes | 50GB | Not measured | Estimated |
| Llama 3.3 70BLlama, 70.6B | Q4 | Yes | 60GB | Not measured | Estimated |
| Llama 4 Scout 17Bx16ELlama, 109B | None fits | No | 89GB at Q4 | Not measured | Estimated |
| Qwen2.5 72BQwen, 72.7B | Q3 | Yes | 54GB | Not measured | Estimated |
| Mistral Large 2 123BMistral, 123B | Q2 | Yes | 62GB | Not measured | Estimated |
One verdict per workload, from the DXP4800 GT's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
The DXP4800 GT is an acceptable local LLM host, with limits. UGREEN confirms CPU-only inference through Ollama in Docker, and the capability is marked limited because there is no NPU.
Private RAG is workable but narrow. The capability is limited to document chat through Open WebUI, which UGREEN documents with an official UGOS Pro guide, and it is vendor confirmed.
AI search is acceptable, with partial coverage. Photo description and OCR search work through Model Manager, but whether document semantic search applies per model is not stated by UGREEN.
AI photos are a confirmed strength. UGREEN lists an AI photo album covering faces, places, objects and duplicates, and the capability is vendor confirmed as supported.
Surveillance is acceptable and vendor confirmed. Surveillance Center supports the DXP series, and local AI detection for people, vehicles, packages and pets is limited, with features varying by model.
The DXP4800 GT is an acceptable home lab system. It supports Docker and virtual machines on UGOS Pro, takes up to 64GB of memory, and has 2 x 10GbE and two M.2 slots.
Business use is acceptable. ECC memory is supported once you add ECC modules, there are four bays and 2 x 10GbE networking, and the warranty is 2 years.
Creative AI is acceptable only for light work. GPU expansion is not supported, hardware acceleration is unconfirmed, and local LLM is limited to CPU-only inference.
Plan on upgrading. 8GB is fitted as standard, while the DXP4800 GT accepts 64GB of DDR4 SO-DIMM 2666, ECC included, which is the difference between small models up to around 7B parameters and 32B class models comfortably, and 70B class models at 4-bit with little headroom (Estimated).
Our estimate for the biggest model that fits in 64GB is about 93B parameters at 4-bit, after setting aside 8GB for the system. Long context windows reduce that figure (Estimated).
ECC support protects against silent memory errors on a box that runs around the clock.
A graphics card is not an option on the DXP4800 GT, which has neither a PCIe slot nor OCuLink. Inference runs on the CPU, with AMD Radeon Vega 8 handling media tasks, so keep to small models.
As the storage half of a RAG setup it works well; as the model host it is limited. Small embedding models run fine on the DXP4800 GT, while the answering model needs more compute than it has.
For network access it has 2 x 10GbE. Fast enough that the network will not be the bottleneck.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| 8GB | 8GB | None | Diskless, 8GB DDR4, 64GB eMMC system storage. |
The closest UGREEN model by bay count and AINAS Score. Only the specifications that differ are listed.
Best fit: small offices that need shared storage with private AI tools and room to grow. That is where the DXP4800 GT scores highest (Business 56/100), followed by Home Lab (55/100).
Less suited to buyers who expect to add a graphics card later.
Other AI Capable systems closest in price.
Buying guides that list it
Yes, at Q4 only, and only after a memory upgrade. Llama 3.3 70B needs 43GB at Q4, and the DXP4800 GT takes up to 64GB, so it fits at Q4 (Estimated). Q8 does not fit (Estimated). It ships with 8GB, and inference is CPU-only, so treat this as a calculation of memory, not a benchmark.
Yes. It ships with 8GB of DDR4 SO-DIMM 2666 and takes up to 64GB. The memory sits in slots rather than being soldered. The platform supports ECC, but the included module is non-ECC, so you need to add ECC modules if you want it.
No. UGREEN confirms there is no PCIe slot, Thunderbolt or OCuLink, and there are no USB4 ports. The only graphics hardware is the integrated Radeon Vega 8. Buyers who want a GPU should look for a PCIe slot or OCuLink.
Yes, it has 2 M.2 NVMe slots. It also has 4 drive bays, and bays 1 and 2 accept U.2 NVMe SSDs as well. It is sold diskless, with 64GB of eMMC system storage.
Yes, with the right modules. The DDR4 SO-DIMM 2666 platform in the DXP4800 GT supports ECC, but the 8GB module it ships with is non-ECC, so ECC protection means buying and fitting ECC modules. Memory goes up to 64GB, and at that ceiling Llama 3.3 70B fits at Q4 (Estimated). For a four-bay system with 2 x 10GbE holding business files, fitting ECC modules during the upgrade adds error correction at the same time.
Yes. Docker is supported on UGOS Pro, and virtual machines are supported too. UGREEN publishes a Docker Compose guide for Ollama and Open WebUI, which is how local LLM inference is run on this system.
Yes. Ollama support is documented by the manufacturer. See the capability table above for the source.
Yes. 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.
Yes. Surveillance AI support is documented by the manufacturer. See the capability table above for the source.
Partly. Hardware acceleration support is not yet verified. See the capability table above for the source.
It has 2 x 10GbE built in. The fastest port is 2 x 10GbE.
Every field records its source. Specifications were last checked on . Spot an error? Send a correction with a manufacturer link.