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

AI CapableQNAP
Why AI Capable: Ryzen V1500B (2018 embedded) is not a current CPU; its PCIe Gen3 x4 slots take only low-profile GPUs up to 6GB on QNAP's list.
Updated
Can the QNAP TS-473A run local AI?
Yes, the QNAP TS-473A can run local language models. It pairs AMD Ryzen Embedded V1500B with up to 64GB of memory, which suits 32B class models comfortably, and 70B class models at 4-bit with little headroom (Estimated).
It also has a GPU path (PCIe x4 slot), so a graphics card can be added later for much faster inference.
Local LLM support is documented by the manufacturer.
The TS-473A suits a buyer who wants a four-bay QNAP NAS that can also run local AI containers on a budget of effort rather than hardware. It takes up to 64GB of DDR4 SODIMM with ECC supported, ships with 8GB, and runs Docker and virtual machines on QTS or QuTS hero. Two PCIe Gen3 x4 slots accept a low-profile NVIDIA card, and QNAP lists cards up to the RTX 3050 LP 6G. Ollama and Open WebUI are vendor confirmed. On memory alone, Llama 3.3 70B fits at Q4 (Estimated). It also has 4 bays, 2 M.2 slots and 2 x 2.5GbE.
The main limit is the GPU path. The Ryzen Embedded V1500B has no integrated GPU and no NPU, so any acceleration needs an add-in card, and the cards QNAP lists top out at 6GB. Buyers who want fast inference on mid-size or large models should look for a system with a much larger GPU memory ceiling, or a built-in NPU with a higher TOPS figure. Network buyers should also note that the onboard ports are 2.5GbE only, so 10GbE needs a PCIe card. Memory is less of a constraint, since up to 64GB of DDR4 SODIMM holds large models, but any model too big for a 6GB card runs on the 4-core CPU.
Amazon US
TS-473A-8G
Ships from the US; delivery and import charges shown at checkout
$887.84
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 | Yes | Vendor confirmed | QNAP Ollama guide (x86, optional NVIDIA GPU from compatibility list). 64 GB RAM; GPUs up to RTX 3050 6GB, so small models on GPU. |
| Ollama | Yes | Vendor confirmed | Container Station Ollama tutorial for x86_64 QTS/QuTS hero. |
| Open WebUI | Yes | Vendor confirmed | QNAP Open WebUI tutorial, QTS 5.2 / QuTS hero h5.2+. |
| RAG | Limited | Unverified | Qsirch AI Knowledge Search (RAG) uses cloud LLMs; not confirmed for this model. |
| AI search | Yes | Vendor confirmed | Qsirch AI Semantic Search needs 64-bit x86 and 8 GB RAM. |
| GPU expansion | Yes | Vendor confirmed | GPGPU list up to RTX 3050 LP 6G; product page: entry-level NVIDIA cards for video processing or VM passthrough. |
| Hardware acceleration | Limited | Vendor confirmed | No iGPU; add-in GPU for transcoding, QAI-M100/U100 (3 TOPS) and Coral M.2/USB Edge TPU on compatibility list. |
| AI photos | Yes | Vendor confirmed | QuMagie AI-assisted photo management (people, things, places). |
| AI video | Yes | Unverified | Software spec page summary did not confirm QVR Face for this model (it is listed for sibling TS-873A). |
| Surveillance AI | Yes | Vendor confirmed | QVR Pro 8 free channels (QTS); QVR Surveillance 2 free + 6 on registration. |
Calculated from the TS-473A'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 TS-473A's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
The TS-473A is an acceptable local LLM host, not a fast one. Ollama is vendor confirmed through Container Station, and QNAP's notes say small models on GPU because listed cards stop at 6GB.
Private RAG is workable here, with one caveat. Ollama and Open WebUI are vendor confirmed, but the Qsirch RAG feature uses cloud LLMs and is unconfirmed for this model.
AI search is a reasonable fit. Qsirch AI Semantic Search is vendor confirmed and needs 64-bit x86 and 8GB of RAM, which the 8GB included memory meets.
AI photo management is covered. QuMagie is vendor confirmed for people, things and places, and runs on the x86 CPU without needing a graphics card.
Surveillance works acceptably. QVR Pro offers 8 free channels on QTS and QVR Surveillance offers 2 free plus 6 on registration, both vendor confirmed, but QVR Face for this model is unconfirmed.
The TS-473A is an adequate home lab box. Docker and virtual machines are both supported, with GPU pass-through listed as working, and memory expands to 64GB across 2 SODIMM slots.
Business use is acceptable. ECC memory is supported, there are 4 bays, a 250W power supply and a 3 year warranty, though the onboard networking is 2.5GbE only.
Creative AI is possible but limited. GPU expansion is vendor confirmed up to the RTX 3050 LP 6G, while hardware acceleration is marked limited because there is no iGPU and cards must be added.
The TS-473A ships with 8GB of DDR4 SODIMM and supports up to 64GB with ECC. Upgrading to the maximum lifts the model ceiling from small models up to around 7B parameters to 32B class models comfortably, and 70B class models at 4-bit with little headroom (both Estimated).
As a rule of thumb, a 4-bit model needs about 0.6GB per billion parameters plus room for context. After leaving around 8GB for the NAS operating system, 64GB leaves space for a model of roughly 93B parameters at most (Estimated).
ECC support protects against silent memory errors on a box that runs around the clock.
Only a small one. The 2x PCIe Gen3 x4 in the TS-473A is sized for network and storage cards; low-profile GPUs are the limit.
With limits. The TS-473A can host a small RAG stack such as AnythingLLM or Open WebUI with a compact embedding model, but memory and compute restrict how large the answering model can be. It suits a personal document library more than a team knowledge base.
The manufacturer also provides AI search across your files (documented by the manufacturer), which covers simple document questions without setting up your own stack.
Networking is 2 x 2.5GbE. Fine for one or two users; indexing a large archive over the network will be slower.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| TS-473A-8G | 8GB | None |
The closest QNAP model by bay count and AINAS Score. Only the specifications that differ are listed.
The TS-473A scores best for Local LLM (51/100) and RAG (50/100). It suits people who want a private chat assistant or coding model at home without sending data to a cloud service.
It is a weaker fit for teams moving large files over the network every day.
Other AI Capable systems closest in price.
Buying guides that list it
Yes, on memory alone. Llama 3.3 70B needs 43GB at Q4 and the TS-473A takes up to 64GB, so it fits at Q4 (Estimated). At Q8 it needs 75GB and does not fit (Estimated). Qwen2.5 72B is marginal at Q4 (Estimated). This is a memory calculation, not a benchmark, and the 6GB GPU limit means large models would run on the CPU.
Up to 64GB. It ships with 8GB of DDR4 SODIMM and has 2 slots, so the maximum is 2x 32GB. The memory sits in two slots rather than being soldered. ECC memory is supported. Added memory is what decides which models fit, so a larger kit widens the models listed on this page.
Yes. It has 2 PCIe Gen3 x4 slots and GPU expansion is vendor confirmed. QNAP's list includes the GTX 1650 LP, RTX 3050 LP 6G, Quadro P620 and P1000, and RTX A400 4GB. The cards are low-profile, and the largest listed has 6GB of memory.
Yes, 2. They are M.2 2280 slots running at PCIe Gen3 x1. They sit alongside 4 SATA 3.5in bays. The x1 link suits caching or small fast volumes, and the slots do not use the PCIe expansion slots.
Yes, on memory. Qwen3 32B needs 20GB at Q4 and 35GB at Q8, and both fit within the 64GB ceiling (Estimated). Neither fits on the 6GB cards QNAP lists for the PCIe slots, so the model would run on the 4-core Ryzen Embedded V1500B rather than the card. The TS-473A-8G comes with 8GB, so a memory upgrade comes first. Ollama and Open WebUI are vendor confirmed through Container Station. These are calculations, not benchmarks.
Yes. Docker is supported through Container Station on QTS 6.1.0 and QuTS hero h6.1.0. Virtual machines are also supported. QNAP's Ollama and Open WebUI guides are vendor confirmed for this x86 platform, so containers are the route for local AI software.
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 documented by the manufacturer. See the capability table above for the source.
It has 2 x 2.5GbE built in, and the 2x PCIe Gen3 x4 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.