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

AI ReadyQNAP
Why AI Ready: AMD EPYC 7302P or 7232P/7252 (no NPU, no iGPU) with PCIe Gen4 expansion and a 1TB DDR4 ECC ceiling, but no vendor GPU path: the two x16 slots are low-profile in a 1U chassis and QNAP lists no GPU, only 4 TOPS Coral accelerators, so ready rather than native.
Updated
Can the QNAP TS-h1090FU run local AI?
The QNAP TS-h1090FU can run small local models, with limits. Its AMD EPYC 7302P (16-core/32-thread, up to 3.3 GHz); AMD EPYC 7232P/7252 (8C/16T, up to 3.2 GHz) on 7232P models and 1024GB memory ceiling suit 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit (Estimated), so treat it as a storage-first NAS that can host light AI tools.
There is no way to add a discrete GPU, so all inference runs on the CPU.
Local LLM support is documented by the manufacturer.
The TS-h1090FU suits a business or home lab that wants an all-flash rackmount NAS and is happy to run AI on the CPU. It has 10 bays taking U.2 NVMe or SATA drives, an AMD EPYC 7302P with 16 cores, and 256GB of DDR4 ECC RDIMM as shipped, expandable to 1024GB. Two 25GbE ports and two 2.5GbE ports handle fast network access. Docker and virtual machines are supported, and QNAP documents Ollama and Open WebUI on it. Estimated: the largest model listed on this page that fits at Q4 is Qwen3 235B A22B. It carries a 5 year warranty and scores 56 in the AINAS Score.
The main limit is the GPU path. The TS-h1090FU is a 1U chassis with two low-profile PCIe Gen4 x16 slots, and QNAP lists no supported GPU for it. Anyone planning GPU inference, Qsirch AI Mode (which needs a 16GB VRAM GPU) or on-premises RAG with a GPU should look for a system with full-height PCIe slots and a vendor-listed GPU, or an integrated GPU or NPU. The only AI accelerators QNAP lists here are Coral USB and M.2 Edge TPU units at 4 TOPS each. Inference on this system is CPU only.
Amazon US
TS-h1090FU-7302P-128G
Ships from the US; delivery and import charges shown at checkout
$13,999.00
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 | QNAP Ollama guide covers x86 QTS/QuTS hero but names no model. Strong CPU and RAM; no GPU listed, so CPU inference only. |
| Ollama | Yes | Vendor confirmed | Container Station Ollama tutorial for x86 QTS/QuTS hero NAS, CPU method applies. Model not named. |
| Open WebUI | Yes | Vendor confirmed | QNAP tutorial needs QTS 5.2 / QuTS hero h5.2 or later and Container Station 3. Model not named. |
| RAG | Limited | Vendor confirmed | Qsirch RAG Search needs 64-bit x86, QTS/QuTS hero 5.0.1+. Cloud RAG needs an OpenAI-compatible API; on-prem needs a compatibility-tested GPU and none is listed. Model not named. QNAP lists this model as raw data storage for AI: https://www.qnap.com/en/solution/optimize-ai-with-qnap-storage |
| AI search | Limited | Vendor confirmed | AI Semantic Search needs 64-bit x86 and 8GB RAM. AI Semantic Index needs x74 or above; x90 not stated. AI Mode needs a 16GB VRAM GPU and none is listed for this model. |
| GPU expansion | No | Vendor confirmed | No GPGPU category on the compatibility list. Two low-profile PCIe Gen4 x16 slots exist in a 1U chassis, with no supported GPU. |
| Hardware acceleration | Limited | Vendor confirmed | No iGPU and no listed GPU. AI accelerators listed: Coral USB and M.2 Edge TPU, 4 TOPS each. |
| AI photos | Yes | Vendor confirmed | QuMagie AI-assisted photo management. |
| AI video | Yes | Vendor confirmed | QVR Face listed under QTS only. |
| Surveillance AI | Yes | Vendor confirmed | QVR Surveillance 2 free channels plus 6 on registration; QVR Pro 8 free (QTS only). |
Calculated from the TS-h1090FU'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 | Q8 | Yes | 98GB | Not measured | Estimated |
| Llama 4 Scout 17Bx16ELlama, 109B | Q8 | Yes | 149GB | Not measured | Estimated |
| Qwen2.5 72BQwen, 72.7B | Q8 | Yes | 101GB | Not measured | Estimated |
| Mistral Large 2 123BMistral, 123B | Q8 | Yes | 164GB | Not measured | Estimated |
One verdict per workload, from the TS-h1090FU's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.
Local LLM work is workable but limited, and runs on the CPU only. QNAP marks local LLM as limited (Vendor confirmed), names no model in its Ollama guide, and lists no GPU, though 256GB of RAM is included.
Private RAG is a good fit, with some conditions. Qsirch RAG Search is marked limited (Vendor confirmed), cloud RAG needs an OpenAI-compatible API, and Open WebUI and Ollama are both Vendor confirmed through Container Station.
AI search is acceptable rather than complete. Qsirch AI Semantic Search needs 8GB of RAM and is marked limited (Vendor confirmed), while AI Mode needs a 16GB VRAM GPU and none is listed.
AI photo management is a good fit. QuMagie, QNAP's AI-assisted photo app, is listed as supported on the TS-h1090FU and the entry is Vendor confirmed.
Surveillance is a good fit. QVR Surveillance gives 2 free channels plus 6 on registration, and QVR Pro gives 8 free channels (QTS only), all Vendor confirmed.
Home lab use is a good fit. It runs Docker and virtual machines, takes up to 1024GB of ECC memory, and has two PCIe Gen4 x16 slots for low-profile cards.
Business use is a good fit. It combines 10 NVMe or SATA bays, ECC memory, two redundant 550W supplies, 25GbE networking and a 5 year warranty in a 1U rackmount chassis.
Creative AI is only acceptable, because there is no GPU route. GPU expansion is marked no (Vendor confirmed), hardware acceleration is limited to 4 TOPS Coral accelerators, and the 16-core CPU does the work.
The TS-h1090FU ships with 256GB of DDR4 ECC RDIMM and supports up to 1024GB with ECC. Upgrading to the maximum lifts the model ceiling from 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit to 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit (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, 1024GB leaves space for a model of roughly 1693B parameters at most (Estimated).
ECC support protects against silent memory errors on a box that runs around the clock.
No. The TS-h1090FU has no PCIe slot or OCuLink port for a graphics card. Inference runs on the CPU, so keep to small models.
With limits. The TS-h1090FU 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.
Networking is 2 x 25GbE + 2 x 2.5GbE. That is ample for indexing large libraries and for several people querying at once.
| Configuration | RAM | SSD | Notes |
|---|---|---|---|
| TS-h1090FU-7232P-32G | 32GB | None | EPYC 7232P/7252 8C/16T. |
| TS-h1090FU-7302P-64G | 64GB | None | |
| TS-h1090FU-7232P-64G | 64GB | None | EPYC 7232P/7252 8C/16T. |
| TS-h1090FU-7302P-128G | 128GB | None | |
| TS-h1090FU-7302P-256G | 256GB | None |
The closest QNAP model by bay count and AINAS Score. Only the specifications that differ are listed.
The TS-h1090FU scores best for Home Lab (67/100) and Business (63/100). It suits home lab owners who want one box for storage, containers, media and a few AI services.
It is a weaker fit for buyers who expect to add a graphics card later.
Other AI Ready systems closest in price.
Buying guides that list it
Workloads it suits
Yes, on memory alone. Estimated: Qwen3 235B A22B needs 142GB at Q4 and 250GB at Q8, and both show as fitting. That is the largest model listed on this page for this system. It is a calculation, not a benchmark, and inference runs on the CPU because no GPU is listed.
Up to 1024GB. The 7302P-256G configuration includes 256GB, and the memory is DDR4 ECC RDIMM. It is expandable, and the spec page describes 12 x 128GB. The 1024GB ceiling is rated medium confidence, because QNAP's launch release says up to 1TB.
No. GPU expansion is marked no (Vendor confirmed). It has two PCIe Gen4 x16 slots, but they take low-profile cards only, in a 1U chassis, and QNAP's compatibility list has no GPU category. AI accelerators listed are Coral USB and M.2 Edge TPU units at 4 TOPS each.
No. It has no built-in M.2 slots. NVMe drives go in the 10 U.2 bays, which also take SATA drives. If you want M.2, QNAP offers QDA adapters and QM2 cards, which use the PCIe slots.
Two. QNAP lists 550W x2 on AC 100-240 V for the TS-h1090FU, which suits a 1U rack that runs around the clock. That power feeds the 16-core EPYC 7302P, up to 1024GB of DDR4 ECC RDIMM, 10 U.2 NVMe or SATA bays and two low-profile PCIe Gen4 x16 slots. None of it goes to a GPU, since QNAP lists no supported card, so the supplies serve CPU, memory and flash work rather than graphics.
5 years. QNAP gives the TS-h1090FU a 5 year warranty. It was released in 2022 and is not marked end of life. Five configurations are sold, with 32GB to 256GB of memory, using either the EPYC 7302P or the 7232P.
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 25GbE + 2 x 2.5GbE built in, and the 2x PCIe Gen4 x16, low-profile cards only: Slot 1 up to 198 x 68.9 x 18.76 mm, Slot 2 up to 250 x 68.9 x 18.76 mm can take a faster network card. The fastest port is 25GbE.