64GB max RAM · 4 x M.2 · 10GbE
AINAS Score 48
£619.99£773.99
Amazon UK · Verified
AI ReadyTerraMaster
Why AI Ready: Intel Xeon E-2224G (4 cores, UHD P630 iGPU, no NPU) with two PCIe 3.0 x8 slots and a 64GB memory ceiling, which meets the AI Ready line for Xeon under 128GB. TerraMaster documents the slots for NIC and RAID cards, not GPUs.
Updated
Can the TerraMaster U16-722-2224 run local AI?
Partly. The TerraMaster U16-722-2224 runs local models on Intel Xeon E-2224G alone, and its 64GB ceiling allows 32B class models comfortably, and 70B class models at 4-bit with little headroom (Estimated), but replies will be slow without graphics acceleration.
No graphics card can be fitted, which keeps every model on the CPU.
Local LLM support is not yet verified.
The U16-722-2224 suits a buyer who wants a 16-bay rackmount NAS that can also run containers and virtual machines. It has an Intel Xeon E-2224G with 4 cores, DDR4-3200 non-ECC UDIMM memory that goes up to 64GB, and 8GB fitted. Docker and VM support are both listed, and it has two PCIe 3.0 x8 slots. On memory alone, Llama 3.3 70B fits at Q4 (Estimated), and Qwen3 32B fits at Q8 (Estimated). Inference would be CPU-only, and local LLM support is unconfirmed for this model. It carries an AINAS Score of 48 and a 550W power supply.
The main drawback is the lack of a GPU path. The two PCIe slots are documented for 10GbE and RAID cards, and GPU use is not documented, so buyers who want graphics acceleration for larger models or creative AI should not assume it. The integrated UHD Graphics P630 handles video transcoding only. Network speed is also a constraint, because the system has four 1GbE ports and nothing faster. Anyone who needs a GPU should look for a system with a documented GPU expansion path, an NPU or a supported graphics card slot. Buyers who need faster networking should look for 10GbE or higher built in.
No verified price
We have not verified a listing for the TerraMaster U16-722-2224 at a retailer we track. The specifications below stand on their own; a price appears here as soon as a listing is confirmed.
Product 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 | Unverified | CPU-only inference on a 4-core Xeon E-2224G via Docker. TerraMaster blog covers Ollama on other x86 models; no source for this model. |
| Ollama | Limited | Vendor confirmed | TerraMaster blog describes running Ollama as a container through the TOS Docker manager on x86 NAS generally; this model not named. |
| Open WebUI | Limited | Vendor confirmed | Same blog names Open WebUI as a container for x86 TOS NAS; this model not named. |
| RAG | Limited | Unverified | Possible through Open WebUI or similar containers; no source showing RAG on this model. |
| AI search | No | Unverified | No local AI or semantic file search documented for this model. |
| GPU expansion | Limited | Unverified | Two PCIe 3.0 x8 slots documented for 10GbE and RAID cards; GPU use not documented. |
| Hardware acceleration | Yes | Vendor confirmed | Datasheet lists hardware transcoding (H.265, MPEG-4 Part 2, VC-1, up to 4K30). Video only, no NPU. |
| AI photos | Limited | Unverified | TOS 7 Photos: AI face recognition and 80+ scene categories. Requires TOS 7; no source confirms TOS 7 for this model. |
| AI video | No | Unverified | No AI video feature found in TerraMaster material. |
| Surveillance AI | Limited | Unverified | TOS 7 surveillance app supports up to 32 ONVIF IP cameras. Requires TOS 7; no source confirms TOS 7 for this model. |
Calculated from the U16-722-2224'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 U16-722-2224's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
The U16-722-2224 is an acceptable local LLM host, with caveats. Local LLM support is limited and Unverified for this model, and inference would be CPU-only on a 4-core Xeon through Docker.
Private RAG is workable but not confirmed on this system. RAG is marked limited and Unverified, though Ollama and Open WebUI are vendor confirmed as containers on x86 TOS NAS generally, and Docker is supported here.
AI search is a poor fit for the U16-722-2224. No local AI or semantic file search is documented for this model, and the capability is marked as not supported and Unverified.
We cannot confirm AI photos on the U16-722-2224. The TOS 7 Photos app offers face recognition and scene categories, but no source confirms that this model runs TOS 7.
We cannot confirm surveillance use on this system. The TOS 7 surveillance app supports up to 32 ONVIF IP cameras, but TOS 7 is not confirmed for the U16-722-2224, so the capability is Unverified.
Home lab use is an acceptable fit. Docker and virtual machines are both listed, memory is expandable up to 64GB, and 16 bays leave plenty of room for storage.
Business use is an acceptable fit. It is a 16-bay rackmount with a 2-year warranty and a 550W supply, with an optional redundant supply, though memory is non-ECC.
Creative AI is rated acceptable, with limits. GPU expansion is limited and Unverified because the PCIe slots are documented for network and RAID cards, and hardware acceleration covers video transcoding only.
Out of the box the U16-722-2224 has 8GB; the official ceiling is 64GB of DDR4-3200 non-ECC UDIMM. A memory upgrade is the cheapest way to make it more useful for AI.
Working backwards from 64GB: reserve about 8GB for the operating system and storage services, and at roughly 0.6GB per billion parameters at 4-bit, the largest model that fits is around 93B (Estimated).
There is no supported way to add a GPU to the U16-722-2224: no PCIe slot and no OCuLink. Inference runs on the CPU, with Intel UHD Graphics P630 handling media tasks, so keep to small models.
For a personal library, yes. A compact embedding model and a small chat model will run, but the U16-722-2224 is better as the document store behind a RAG system whose model runs on stronger hardware.
Built-in networking: 4 x 1GbE. Adequate for a household, slower for bulk indexing.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| 8GB | 8GB | None | Diskless; 1x 8GB DDR4 pre-installed, 4 slots. |
The closest TerraMaster model by bay count and AINAS Score. Only the specifications that differ are listed.
Its strongest AINAS profiles are Home Lab at 52 and Business at 50, which points to home lab owners who want one box for storage, containers, media and a few AI services.
Look elsewhere if you are buyers who expect to add a graphics card later, or teams moving large files over the network every day.
Other AI Ready systems closest in overall score.
Yes on memory alone, at Q4 only. Llama 3.3 70B needs 43GB at Q4 and fits within the 64GB ceiling (Estimated). At Q8 it needs 75GB, which does not fit (Estimated). This is a calculation, not a benchmark. Inference would be CPU-only through Docker, and local LLM support is Unverified for this model.
64GB is the maximum, as 4 x 16GB. The system ships with 8GB, a single 8GB module, and has 4 memory slots. It uses DDR4-3200 non-ECC UDIMM. Memory is expandable, so you can add modules later to reach the ceiling.
No. It uses DDR4-3200 non-ECC UDIMM, and ECC is not supported. Buyers who need error-correcting memory for business data should look for a system with ECC memory in its specification. The maximum memory here is 64GB.
Not as a documented feature. It has one PCIe x16 slot and one PCIe x8 slot, both running at PCIe 3.0 x8. TerraMaster documents them for 10GbE and RAID cards. GPU use is not documented and is Unverified, so do not plan around it.
Two cards. The U16-722-2224 has one PCIe x16 slot running at PCIe 3.0 x8 and one PCIe x8 slot, also at PCIe 3.0 x8. TerraMaster documents them for 10GbE and RAID cards, and graphics card use is Unverified, so they are best treated as the route to faster networking beyond the four 1GbE ports. Anyone who fits a 10GbE card in one slot keeps the other free for a RAID card.
Yes. Docker is listed, and virtual machines are listed too. TerraMaster describes running Ollama and Open WebUI as containers through the TOS Docker manager on x86 NAS generally, but this model is not named. The system runs TOS.
Partly. Ollama support is documented by the manufacturer. See the capability table above for the source.
Partly. Open WebUI support is documented by the manufacturer. See the capability table above for the source.
Partly. AI photos support is not yet verified. See the capability table above for the source.
Partly. Surveillance AI support is not yet verified. See the capability table above for the source.
Yes. Hardware acceleration support is documented by the manufacturer. See the capability table above for the source.
It has 4 x 1GbE built in, and the 1x PCIe x16 slot (PCIe 3.0 x8), 1x PCIe x8 slot (PCIe 3.0 x8) can take a faster network card. 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.