128GB max RAM · 24 bays · 2 x 10GbE
AINAS Score 57
No verified price

AI ReadyQNAP
Why AI Ready: Intel Core i3-12100 (4C/8T, no NPU) with UHD 730, 64GB memory ceiling and a PCIe Gen4 x16 slot, but the official GPU list tops out at slot-powered cards such as the 6GB RTX 3050 LP, so LLMs are mostly CPU-bound.
Updated
Can the QNAP TVS-h474 run local AI?
Partly. The QNAP TVS-h474 runs local models on Intel Core i3-12100 (4-core/8-thread, up to 4.3 GHz); Pentium Gold G7400 (2C/4T) on PT-8G 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.
A graphics card can be added through its PCIe x16 slot, which is the biggest single speed upgrade for local models.
Local LLM support is documented by the manufacturer.
The TVS-h474 suits a buyer who wants a four-bay NAS that can also host local models in containers. It has an Intel Core i3-12100 (4-core/8-thread), 8GB of DDR4 SODIMM memory expandable to 64GB, two M.2 slots and two 2.5GbE ports. Container Station supports Docker, and Ollama and Open WebUI are vendor confirmed. With 64GB installed, Llama 3.3 70B fits at Q4 (Estimated) and Qwen3 32B fits at both Q4 and Q8 (Estimated). The PCIe Gen4 x16 slot takes a graphics card up to 185mm long. The AINAS Score is 56. QuTS hero h6.1.0 or QTS 6.1.0 runs on it, with Virtualization Station for VMs.
The main catch is the GPU path. The x16 slot is powered by the slot only, and the largest listed card is a 6GB RTX 3050 LP, so local LLM support is marked limited and inference is mostly on the CPU. Anyone who wants large models running on a GPU should look for a system with a full-power PCIe slot, a strong PSU and room for a card with 16GB or more of VRAM. Qsirch AI Mode needs a 16GB VRAM GPU, and none is on this model's list. Buyers who rely on the included 8GB will also need to budget for a memory upgrade before the larger models listed on this page become usable.
Amazon US
TVS-h474-PT-8G
Ships from the US; delivery and import charges shown at checkout
Price awaiting re-verification
Last verified · checked
Out of 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 | QNAP Ollama guide applies (x86, 16GB+ for 7B). Largest listed GPU is 6GB RTX 3050 LP, so mostly CPU inference. |
| Ollama | Yes | Vendor confirmed | Container Station Ollama tutorial for x86 QTS/QuTS hero NAS. |
| Open WebUI | Yes | Vendor confirmed | QNAP Open WebUI tutorial, QTS 5.2 / QuTS hero h5.2+. |
| RAG | Limited | Unverified | Qsirch RAG Search (beta) exists; requirements not stated, not confirmed for this model. |
| AI search | Yes | Vendor confirmed | Qsirch AI Semantic Search/Index: x86, 8GB RAM, x74 series or above. AI Mode needs a 16GB VRAM GPU, none on this model's list. |
| GPU expansion | Limited | Vendor confirmed | PCIe Gen4 x16 slot, 185 mm cards, 250W PSU; product page says only entry-level cards powered by the slot. |
| Hardware acceleration | Yes | Vendor confirmed | UHD Graphics transcoding and OpenVINO; Coral Edge TPU (4 TOPS) and Hailo-8 M.2 (26 TOPS) on AI accelerator list. |
| AI photos | Yes | Vendor confirmed | QuMagie AI-assisted photo management. |
| AI video | Yes | Vendor confirmed | QVR Face and QVR Human listed under QTS. |
| Surveillance AI | Yes | Vendor confirmed | QVR Surveillance 2 free + 6 on registration; QVR Pro 8 free (QTS). |
Calculated from the TVS-h474'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 TVS-h474's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
The TVS-h474 is a good fit for local LLMs. Ollama is vendor confirmed through Container Station, and memory can reach 64GB, though local LLM is marked limited because the largest listed GPU is 6GB, so inference is mostly CPU.
Private RAG is an OK fit. Open WebUI and Ollama are vendor confirmed, but the Qsirch RAG Search beta is Unverified, with requirements not stated and no confirmation for this model.
AI search is an OK fit. Qsirch AI Semantic Search is vendor confirmed on x86 with 8GB of RAM, but AI Mode needs a 16GB VRAM GPU and none is on this model's list.
AI photos is an OK fit. QuMagie provides AI-assisted photo management and is vendor confirmed, and the UHD Graphics 730 supports hardware acceleration through OpenVINO.
Surveillance is an OK fit. QVR Surveillance offers 2 free channels plus 6 on registration, and QVR Pro offers 8 free, all vendor confirmed, with QVR Face and QVR Human listed under QTS.
Home lab use is an OK fit. Docker and virtual machines are both supported, with Virtualization Station suggesting 4 VMs, and two 2.5GbE ports plus two M.2 slots cover networking and fast storage.
Business use is an OK fit. It has four bays, a 3-year warranty and a PCIe Gen3 x2 slot for network cards, though ECC memory is not offered and networking is 2.5GbE as standard.
Creative AI is a good fit. The PCIe Gen4 x16 slot takes cards up to 185mm, and hardware acceleration is vendor confirmed, including OpenVINO, Coral Edge TPU and Hailo-8 M.2, although GPU choice is limited to slot-powered cards.
Out of the box the TVS-h474 has 8GB; the official ceiling is 64GB of DDR4 SODIMM. 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).
It can. A Slot 1 PCIe Gen4 x16 + Slot 2 PCIe Gen3 x2; cards up to 185 x 68.9 x 18.76 mm is the route, and a card with 16GB or more of its own memory changes what the TVS-h474 can run. Measure the space and check the 250W power budget before ordering.
For a personal library, yes. A compact embedding model and a small chat model will run, but the TVS-h474 is better as the document store behind a RAG system whose model runs on stronger hardware.
The manufacturer also provides AI search across your files (documented by the manufacturer), which covers simple document questions without setting up your own stack.
Built-in networking: 2 x 2.5GbE. Adequate for a household, slower for bulk indexing.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| TVS-h474-i3-8G | 8GB | None | Core i3-12100 4C/8T, UHD 730. |
| TVS-h474-PT-8G | 8GB | None | Pentium Gold G7400 2C/4T, UHD 710. |
The closest QNAP model by bay count and AINAS Score. Only the specifications that differ are listed.
Its strongest AINAS profiles are Local LLM at 60 and RAG at 57, which points to people who want a private chat assistant or coding model at home without sending data to a cloud service.
Look elsewhere if you are teams moving large files over the network every day.
Other AI Ready systems closest in overall score.
Workloads it suits
Yes, at Q4 only, and only with memory upgraded. Estimated: Llama 3.3 70B needs 43GB at Q4, which fits within the 64GB ceiling, but 75GB at Q8 does not fit. Qwen2.5 72B is marginal at Q4 (Estimated). Inference would be mostly on the CPU because the largest listed GPU is a 6GB card.
It takes up to 64GB, as 2 x 32GB DDR4 SODIMM. Each configuration ships with 8GB, so you would need to add memory for the larger models. The memory is expandable rather than soldered, and it is not ECC.
Yes, but only entry-level cards. The PCIe Gen4 x16 slot takes cards up to 185 x 68.9 x 18.76mm, powered by the slot. Listed cards include the GT 1030, GTX 1650 LP and the 6GB RTX 3050 LP. The GPU expansion support is marked limited and vendor confirmed.
Yes, it has 2 M.2 2280 slots running at PCIe Gen3 x2. The Hailo-8 M.2 (26 TOPS) is also on the AI accelerator list. There is no NPU built in, so any extra AI acceleration comes from add-in hardware.
Two. The TVS-h474-i3-8G has an Intel Core i3-12100 with 4 cores and 8 threads and UHD Graphics 730, and the TVS-h474-PT-8G has a Pentium Gold G7400 with 2 cores and 4 threads and UHD 710. Both come with 8GB of DDR4 SODIMM and take up to 64GB, so the Estimated fits on this page apply to either once memory is upgraded. Because inference here runs mostly on the CPU, the i3 version's extra cores matter more than the GPU slot does.
Yes. Container Station supports Docker and LXC/LXD, and QNAP's Ollama and Open WebUI tutorials are vendor confirmed. It also supports virtual machines through Virtualization Station, which suggests 4 VMs. The operating system is QuTS hero h6.1.0 or QTS 6.1.0.
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.
Yes. Hardware acceleration support is documented by the manufacturer. See the capability table above for the source.
It has 2 x 2.5GbE built in, and the Slot 1 PCIe Gen4 x16 + Slot 2 PCIe Gen3 x2; cards up to 185 x 68.9 x 18.76 mm can take a faster network card. The fastest port is 2 x 2.5GbE.
Every field records its source. Specifications were last checked on . Spot an error? Send a correction with a manufacturer link.