AINAS
QNAP TS-h686 6-bay NAS

QNAP TS-h686

AI CapableQNAPDiscontinued

Why AI Capable: Xeon D-1602 (2 cores, 2019) is not a current CPU, and QNAP's GPU list for its PCIe slot stops at a 4GB GTX 1650.

Updated

Can the QNAP TS-h686 run local AI?

Only lightly. Expect 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit (Estimated) on the QNAP TS-h686: Intel Xeon D-1602 (2-core/4-thread, up to 3.2 GHz) does the work with up to 128GB of memory, which is fine for experiments rather than a daily assistant.

Its PCIe x8 slot leaves room to add a GPU when you outgrow CPU inference.

Local LLM support is documented by the manufacturer.

The short answer

The TS-h686 suits a buyer who wants a six-bay QNAP that can also host local models and has room to grow. It takes up to 128GB of DDR4 ECC UDIMM memory, and the memory is expandable. It has 4 x 2.5GbE, 2 M.2 NVMe slots and two PCIe Gen3 x8 slots. Container Station covers Docker, and Ollama and Open WebUI are both vendor confirmed. At 128GB, Mistral Large 2 123B fits at Q4 (Estimated). Qwen3 32B fits at both Q4 and Q8 (Estimated). The AINAS Score is 53. Surveillance and QuMagie photo management are vendor confirmed, and the warranty is 3 years.

The main issue is processing power. The TS-h686 has a 2-core Xeon D-1602 and no NPU or integrated graphics. QNAP's official GPU list for the PCIe slots stops at a 4GB GTX 1650, and the vendor notes describe CPU inference as slow despite the memory. Buyers who want responsive large models should look for a system with a current multi-core CPU, a GPU path with much more video memory, or an NPU with a stated TOPS figure. QNAP has also discontinued the TS-h686, and its only variant is marked end of life, which matters to anyone who wants years of vendor support.

Current offers

No verified price

We have not verified a listing for the QNAP TS-h686 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.

Key specifications

CPU
Intel Xeon D-1602 (2-core/4-thread, up to 3.2 GHz)
Source: Manufacturer
Included RAM
8GB DDR4 ECC UDIMM
Source: Manufacturer
Maximum RAM
128GB ECC
Source: Manufacturer
Networking
4 x 2.5GbE
Source: Manufacturer
PCIe
Slot 1 PCIe Gen3 x8 (260 x 111.15 mm) + Slot 2 PCIe Gen3 x8 (150 x 68.9 mm)
Source: Manufacturer
GPU expansion
PCIe x8 slot
Source: Official docs
Power supply
250W
Source: Manufacturer
OS
QuTS hero h6.0.2 / QTS 5.2.10
Source: Manufacturer
Docker
Yes
VM support
Yes
Released
2020
Segment
Desktop

Can the TS-h686 run local LLMs?

Each capability shows how it was confirmed: Verified by AINAS, Vendor confirmed, Community confirmed or Unverified.

CapabilitySupportedEvidenceNote
Local LLMLimitedVendor confirmedQNAP Ollama guide applies (x86). 2-core Xeon D and 4GB max listed GPU: slow CPU inference despite 128GB RAM.
Ollama YesVendor confirmedContainer Station Ollama tutorial for x86 QTS/QuTS hero NAS.
Open WebUI YesVendor confirmedQNAP Open WebUI tutorial, QTS 5.2 / QuTS hero h5.2+.
RAGLimitedUnverifiedQsirch RAG Search (beta) exists; not confirmed for this model.
AI searchLimitedUnverifiedx86 with 8GB meets AI Semantic Search minimum; whether x86 series counts as x74 or above for AI Semantic Index is not stated.
GPU expansionLimitedVendor confirmedTwo PCIe Gen3 x8 slots, 250W PSU; official list tops out at GTX 1650 4GB.
Hardware accelerationLimitedVendor confirmedNo iGPU; hardware transcoding only via optional PCIe GPU. No AI accelerator category on compatibility list.
AI photos YesVendor confirmedQuMagie AI-assisted photo management.
AI video YesVendor confirmedQVR Face and QVR Human listed under QTS.
Surveillance AI YesVendor confirmedQVR Surveillance 2 free + 6 on registration; QVR Pro 8 free (QTS).

What AI models can the TS-h686 run?

Calculated from the TS-h686's memory and each model's published size at each quantization. Rows marked Estimated are calculations, not benchmarks. Check any model.

ModelBest quantizationRuns?Memory neededExpected performanceVerification
Llama 3.2 3BLlama, 3.21BQ8Yes13GBNot measuredEstimated
Llama 3.1 8BLlama, 8.03BQ8Yes19GBNot measuredEstimated
Qwen3 8BQwen, 8.19BQ8Yes19GBNot measuredEstimated
Gemma 3 12BGemma, 11.8BQ8Yes24GBNot measuredEstimated
Qwen3 14BQwen, 14.8BQ8Yes28GBNot measuredEstimated
Phi-4 14BPhi, 14.7BQ8Yes28GBNot measuredEstimated
Mistral Small 3 24BMistral, 23.6BQ8Yes38GBNot measuredEstimated
Gemma 3 27BGemma, 27BQ8Yes44GBNot measuredEstimated
Qwen3 30B A3BQwen, 30.5BQ8Yes47GBNot measuredEstimated
Qwen3 32BQwen, 32.8BQ8Yes50GBNot measuredEstimated
DeepSeek-R1 32B (Qwen distill)DeepSeek, 32.8BQ8Yes50GBNot measuredEstimated
Llama 3.3 70BLlama, 70.6BQ8Yes98GBNot measuredEstimated
Llama 4 Scout 17Bx16ELlama, 109BQ4Yes89GBNot measuredEstimated
Qwen2.5 72BQwen, 72.7BQ8Yes101GBNot measuredEstimated
Mistral Large 2 123BMistral, 123BQ5Yes112GBNot measuredEstimated

Workload fit

One verdict per workload, from the TS-h686's use-case scores and capability rows. Unknown means we cannot confirm it, not that it probably works.

Local LLMOK

The TS-h686 is an acceptable local LLM host, with caveats. Local LLM support is marked limited and vendor confirmed, and the notes say a 2-core Xeon D and a 4GB maximum listed GPU mean slow CPU inference.

Private RAGOK

Private RAG is workable on the TS-h686, with a gap. Ollama and Open WebUI are vendor confirmed, but Qsirch RAG Search (beta) is limited and Unverified, so it is not confirmed for this model.

AI searchUnknown

We cannot confirm AI search on the TS-h686. The capability is limited and Unverified, and it is not stated whether this x86 series meets the requirement for the AI Semantic Index, though 8GB meets the minimum for AI Semantic Search.

AI photosOK

AI photo management is a reasonable fit for the TS-h686. QuMagie, QNAP's AI-assisted photo manager, is listed for this model and is vendor confirmed.

SurveillanceOK

Surveillance is a reasonable use for the TS-h686. QVR Surveillance includes 2 free licences plus 6 on registration, QVR Pro includes 8 free under QTS, and QVR Face and QVR Human are listed. All are vendor confirmed.

Home labOK

The TS-h686 works as a home lab server. It supports Docker and virtual machines, takes up to 128GB of ECC memory, and has 6 bays, 2 M.2 slots and 4 x 2.5GbE ports.

BusinessOK

The TS-h686 is an acceptable business file server. It has ECC memory, QuTS hero, 6 drive bays and a 3 year warranty, but it is discontinued and has only a 2-core CPU.

Creative AIOK

Creative AI is possible on the TS-h686 but limited by the GPU path. GPU expansion is limited and vendor confirmed, with two PCIe Gen3 x8 slots and an official list that tops out at a GTX 1650 4GB.

How much RAM does the TS-h686 support?

Plan on upgrading. 8GB is fitted as standard, while the TS-h686 accepts 128GB of DDR4 ECC UDIMM, ECC included, which is the difference between small models up to around 7B parameters and 70B class models at higher-quality quantization, or large mixture-of-experts models at 4-bit (Estimated).

Our estimate for the biggest model that fits in 128GB is about 200B parameters at 4-bit, after setting aside 8GB for the system. Long context windows reduce that figure (Estimated).

ECC memory support is a plus for always-on systems that also hold important files.

Can you add a GPU to the TS-h686?

Yes, through a Slot 1 PCIe Gen3 x8 (260 x 111.15 mm) + Slot 2 PCIe Gen3 x8 (150 x 68.9 mm). That is the single biggest upgrade for local AI on the TS-h686: model layers move into fast graphics memory and replies speed up dramatically. Card size and power are the constraints (250W supply).

Is the TS-h686 suitable for RAG?

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 TS-h686, while the answering model needs more compute than it has.

For network access it has 4 x 2.5GbE. Enough for light use; large imports will take a while.

Limitations

  • The CPU is a 2-core Xeon D-1602 with no NPU or integrated graphics. The vendor notes describe CPU inference as slow despite the 128GB memory ceiling.
  • The official GPU list for the PCIe slots tops out at a GTX 1650 with 4GB. GPU expansion and hardware acceleration are both marked limited.
  • The TS-h686 is discontinued, and the only variant, the TS-h686-D1602-8G, is marked end of life on QNAP's spec page.
  • Only 8GB is included in the TS-h686-D1602-8G, so the large Estimated fits on this page, such as Mistral Large 2 123B at Q4, need the memory raised towards 128GB first.
  • Qsirch RAG Search and AI Semantic Search are Unverified for this model, so neither should be counted on.

Which TS-h686 configurations are sold?

ConfigurationRAMSSDNotes
TS-h686-D1602-8G8GBNone

How is the TS-h686 different from the TVS-h674T?

The closest QNAP model by bay count and AINAS Score. Only the specifications that differ are listed.

  • Processor: Intel Xeon D-1602 (2-core/4-thread, up to 3.2 GHz) on the TS-h686, Intel Core i5-12400 on the TVS-h674T
  • CPU cores: 2 on the TS-h686, 6 on the TVS-h674T
  • Integrated graphics: none on the TS-h686, Intel UHD Graphics 730 on the TVS-h674T
  • Memory fitted: 8GB on the TS-h686, 32GB on the TVS-h674T
  • Memory ceiling: 128GB on the TS-h686, 64GB on the TVS-h674T
  • ECC: yes on the TS-h686, no on the TVS-h674T
  • Networking: 4 x 2.5GbE on the TS-h686, 2 x 2.5GbE on the TVS-h674T
  • GPU path: PCIe x8 slot on the TS-h686, PCIe x16 slot on the TVS-h674T
  • Status: discontinued on the TS-h686, tracked on the TVS-h674T
  • AI class: AI Capable on the TS-h686, AI Ready on the TVS-h674T
  • AINAS Score: 53 on the TS-h686, 54 on the TVS-h674T

See the full QNAP TVS-h674T page

Who should buy the TS-h686?

Best fit: people who want a private chat assistant or coding model at home without sending data to a cloud service. That is where the TS-h686 scores highest (Local LLM 59/100), followed by RAG (59/100).

Less suited to teams moving large files over the network every day and buyers who want a model with a long support life ahead.

What are the alternatives to the TS-h686?

Other AI Capable systems closest in overall score.

UGREEN DXP4800 GT

UGREEN DXP4800 GT

UGREEN

AI Capable

64GB max RAM · 4 bays · 2 x 10GbE

AINAS Score 47

£449.99£529.99

Amazon UK · Verified

View

Common questions

Can the TS-h686 run Mistral Large 2 123B?

Yes at Q4, as an estimate. Mistral Large 2 123B needs 73GB at Q4, which fits within the 128GB maximum (Estimated). At Q8 it needs 130GB, so it does not fit (Estimated). The vendor notes warn of slow CPU inference, because the processor is a 2-core Xeon D-1602. The next model up, Qwen3 235B A22B, does not fit at Q4 or Q8 (Estimated). This is a memory calculation, not a benchmark.

Can you upgrade the memory in the TS-h686?

Yes. The TS-h686 ships with 8GB and takes up to 128GB of DDR4 ECC UDIMM, as 4 x 32GB modules. The memory sits in slots rather than being soldered, so you can expand it. Because the type is ECC UDIMM, buy compatible ECC modules.

Can the TS-h686 take a graphics card?

Yes, with limits. It has two PCIe Gen3 x8 slots and a 250W power supply. QNAP's official list tops out at a GTX 1650 with 4GB, and also includes the GT 1030 and GTX 1050 Ti. There is no integrated GPU, so hardware transcoding needs an optional PCIe card.

Does the TS-h686 run Docker?

Yes. Container Station covers Docker on QuTS hero h6.0.2 and QTS 5.2.10, and virtual machines are supported. QNAP's Ollama tutorial for x86 QTS and QuTS hero is vendor confirmed, and the Open WebUI tutorial needs QTS 5.2 or QuTS hero h5.2 or later. The 2-core Xeon D-1602 is the real constraint: the vendor notes describe CPU inference as slow even with 128GB fitted, so containers for file, photo and camera work suit it better than serving a model to several users.

Can the TS-h686 run Llama 3.3 70B?

Yes, at Q4 and Q8 (Estimated). Llama 3.3 70B needs 43GB at Q4 and 75GB at Q8, and both fit within the 128GB maximum (Estimated). The TS-h686-D1602-8G comes with 8GB, so the memory has to be raised towards 128GB first. The model would run on the 2-core Xeon D-1602, because the official GPU list stops at a 4GB GTX 1650, and the vendor notes describe CPU inference as slow. These are memory calculations, not benchmarks.

Does the TS-h686 have M.2 slots?

Yes, it has 2 M.2 slots that take NVMe SSDs in 22110 or 2280 sizes, at PCIe Gen3 x4. They sit alongside 6 drive bays, made up of 4 x 3.5-inch and 2 x 2.5-inch SATA. QuTS hero or QTS 5.2.10 is the operating system.

Can the TS-h686 run Ollama?

Yes. Ollama support is documented by the manufacturer. See the capability table above for the source.

Does Open WebUI work on the TS-h686?

Yes. Open WebUI support is documented by the manufacturer. See the capability table above for the source.

Does the TS-h686 do AI photo recognition?

Yes. AI photos support is documented by the manufacturer. See the capability table above for the source.

Can the TS-h686 run AI surveillance?

Yes. Surveillance AI support is documented by the manufacturer. See the capability table above for the source.

Does the TS-h686 have hardware AI acceleration?

Partly. Hardware acceleration support is documented by the manufacturer. See the capability table above for the source.

How fast is the TS-h686's network?

It has 4 x 2.5GbE built in, and the Slot 1 PCIe Gen3 x8 (260 x 111.15 mm) + Slot 2 PCIe Gen3 x8 (150 x 68.9 mm) can take a faster network card. The fastest port is 4 x 2.5GbE.

Where do these specifications come from?

Every field records its source. Specifications were last checked on . Spot an error? Send a correction with a manufacturer link.