An NPU is purpose-built AI acceleration hardware that may be integrated into a processor or SoC, or supplied as a separate coprocessor. Unlike a GPU, which handles general-purpose parallel computing, an NPU is optimised specifically for the matrix multiplication operations that underpin neural networks. The result is efficient, low-power inference for targeted AI tasks, without the heat, power draw, or cost of a discrete GPU.
In short: An NPU accelerates specific AI tasks (photo recognition, object detection, voice processing, small model inference) efficiently and cheaply. An NPU is not a general substitute for a GPU; usable LLM size and speed depend on the accelerator, memory, quantization, runtime, and model support rather than a universal parameter-count cutoff. The cited Intel Celeron J/N processors have no integrated NPU, but ARM-based NAS hardware varies: several QNAP ARM models include built-in NPUs. QNAP sells both entry-level ARM NAS models with built-in NPUs and the high-end Core Ultra TVS-AIh1688ATX in Australia.
What an NPU Is and How It Differs from CPU and GPU
The three processor types each have different strengths for AI workloads:
- CPU: General-purpose, sequential-first processing. Good at logic, branching, and low-latency single tasks. Slow at the massive parallel matrix operations AI models require. Available on every computer.
- GPU: Thousands of parallel cores designed originally for graphics. Excellent at the matrix multiplications used in AI training and inference. Discrete-GPU power draw varies widely by model and workload. Expensive. Not present in most NAS.
- NPU: A dedicated accelerator for specific AI inference operations. NPUs are designed for power-efficient AI inference, but actual power draw varies by implementation and workload. Faster than CPU for targeted tasks. Cannot handle large general-purpose models. Integrated into modern CPUs.
An NPU is not a replacement for a GPU. Do not assume an NPU can run a 70B model usefully; feasibility depends on memory, quantization, runtime support, and the specific accelerator. What it can do is handle the AI features that vendors market as "AI NAS" features, such as photo face recognition, smart album generation, document OCR, and keyword search, far more efficiently than routing that work through the CPU.
How NPU Performance Is Measured: TOPS Explained
NPU performance is measured in TOPS: Tera Operations Per Second. One TOPS means the processor can perform one trillion AI operations per second. This metric specifically refers to 8-bit integer (INT8) operations, which is the standard precision for efficient inference.
To put the numbers in context:
- Microsoft requires an NPU capable of more than 40 TOPS for Copilot+ PCs. Outside that certification threshold, TOPS alone does not determine which workloads or model sizes run well because software support, precision, memory bandwidth, and model architecture also matter.
Whether a NAS feature runs locally or in the cloud depends on the vendor's software design and supported execution backend; TOPS alone does not establish application compatibility or performance.
Which CPUs Have NPUs in 2026
NPU support in mainstream CPUs is recent. Here are the relevant families:
CPU Families with Integrated NPU
| CPU Family | NPU TOPS | Typical Device | NAS Use | |
|---|---|---|---|---|
| Intel Core Ultra 100 (Meteor Lake) | ~11 TOPS | ~11 TOPS | High-end laptops; QNAP's TVS-hx74 series does not use Core Ultra processors. | QNAP TVS-H674/H874 |
| Intel Core Ultra 200 (Arrow Lake) | ~40+ TOPS | ~40+ TOPS | 2025 desktops/laptops | Not yet in NAS retail |
| AMD Ryzen AI 300 (Strix Point) | ~50 TOPS | ~50 TOPS | Laptops, mini-PCs | Some mini-PCs (Minisforum, Beelink) |
| Qualcomm Snapdragon X Elite | ~45 TOPS | ~45 TOPS | ARM laptops (Surface, MacBook competitors) | Not in NAS |
| Intel Celeron J/N series | N/A | N/A | Most consumer NAS (Synology, QNAP budget) | No NPU |
| ARM Cortex-A (various) | None or <2 TOPS | None or <2 TOPS | Entry NAS, IoT | Negligible for AI |
The key takeaway: the CPUs that power most consumer NAS devices (Intel Celeron in the QNAP TS-464, AMD Ryzen in the Synology DS925+, Intel Pentium Gold in the UGREEN DXP4800 Plus) have no NPU. AI features on these devices run on the CPU general cores, which is slower and less efficient than a dedicated NPU would be.
Which NAS Devices Have NPU Hardware in Australia
NPU-equipped NAS sold in Australia include several QNAP ARM models and the Core Ultra TVS-AIh1688ATX; the TVS-hx74 series is not NPU-equipped.
| NPU-equipped models (AU retail) | The QNAP TVS-h474, TVS-h674, TVS-h874 and TVS-h874X are not NPU-equipped models. |
|---|---|
| CPU family | 12th-generation Intel Core or Pentium; no integrated NPU |
| NPU TOPS | ~11 TOPS (Intel AI Boost) |
| AU price range | $2,400-$9,000+ (varies by model and drive config) |
| AU retailer | Scorptec, Device Deal |
| Synology NPU NAS | None currently (as of April 2026) |
| UGREEN NPU NAS | UGREEN's DXP4800 uses Intel N100 and the DXP4800 Plus uses Pentium Gold 8505; the newer iDX series uses Intel Core Ultra processors with integrated NPUs and is offered for shipment to Australia. |
| Asustor NPU NAS | None currently |
Synology has not announced a consumer NAS with an integrated NPU as of early 2026. Synology says Photos recognition is performed locally; its mobile app can also use the phone's NPU for object recognition before returning results to the NAS. Synology's enterprise units use higher-end Intel Xeon processors, which also lack integrated NPUs.
UGREEN documents local AI features in UGOS Pro, while optional services may use external model providers; its newer iDX hardware includes an integrated Core Ultra NPU.
What an NPU Actually Accelerates on a NAS
QNAP documents NPU acceleration for QNAP AI Core image recognition on supported NPU-equipped models. Support for third-party OCR, LLM, transcription, and surveillance workloads must be verified separately for each application and runtime.
An 11-TOPS rating alone does not establish interactive 7B-model performance; throughput depends on the model, quantization, context, memory system, runtime, and operator support. It will be noticeably faster than the CPU cores for these tasks, but a 7B Ollama model on an 11 TOPS NPU still generates tokens slowly. The Ryzen AI 300 (50+ TOPS) is a more capable tier for LLM inference.
What an NPU Cannot Do
Vendor marketing blurs the line between what an NPU handles and what a GPU handles. To be specific:
- Run large language models at interactive speeds: A 7B model on an 11 TOPS NPU will still be slow. Whether a 13B-or-larger model reaches acceptable latency depends on quantization, context, memory, runtime support, and the specific NPU or hybrid execution path.
- Replace a GPU for image generation (Stable Diffusion): Stable Diffusion does not universally require a discrete GPU; supported NPUs can accelerate selected diffusion pipelines, while ComfyUI compatibility depends on its available backend and the hardware. Do not assume a NAS NPU supports Stable Diffusion or ComfyUI; confirm the model, runtime, driver, memory, and application compatibility for the specific NAS.
- General parallel compute (video encoding, 3D rendering): NPUs are specialised for AI matrix operations, not general-purpose parallel work.
- Run multiple AI workloads simultaneously: NPU resources are shared; running face recognition while also doing LLM inference will compete for the same limited TOPS.
NPU vs GPU: When Each Makes Sense
The choice between an NPU-equipped NAS and a GPU-capable NAS is a workload question:
- Use NPU if: your primary AI workloads are photo recognition, document OCR, transcription, surveillance object detection, or small model (1B-3B) inference. An NPU-equipped NAS can handle supported AI workloads efficiently without a discrete GPU, but acceleration depends on the vendor software, runtime, driver, operators, and model format.
- Use GPU if: you need to run 7B+ language models at interactive speeds, generate images with Stable Diffusion, or serve inference to multiple concurrent users. A discrete GPU is often the practical choice for high-throughput LLM and image-generation workloads, but it is not universally required; supported NPUs and other accelerators can run selected models.
- Use CPU only if: your AI workloads are low-frequency (a few queries per day), quality matters more than speed, and hardware cost is a constraint. CPU inference on a 7B Q4 model may be workable for some batch tasks, but throughput varies substantially with CPU, memory bandwidth, model, context, and software configuration.
Australian Context: NPU NAS Buying Considerations
For NPU acceleration in Australian retail, consider verified NPU-equipped models such as QNAP's ARM NPU range or TVS-AIh1688ATX; the TVS-hx74 series does not contain an NPU. Australian availability varies across multiple retailers; check current stock and delivery estimates for the exact model. Confirm warehouse stock versus on-order before purchasing, because lead times vary by model, retailer, and supplier inventory.
The price premium for NPU capability is significant. A QNAP TVS-h474 (12th-generation Intel Core i3 or Pentium; no integrated NPU) runs $2,400+ versus the current Australian retail price of a QNAP TS-464 (Celeron, no NPU), which should be checked immediately before publication. Evaluate whether your AI workload justifies this premium. For most home users, the TS-464 running Ollama via CPU is adequate for the AI tasks they will actually use daily.
Mini-PCs with AMD Ryzen AI 300 (50 TOPS NPU) offer a higher-TOPS NPU at lower cost than the QNAP TVS-H series, but without the NAS storage architecture. The mini-PC vs NAS for local AI comparison covers this trade-off in detail.
Running a QNAP TVS-H with active AI inference adds measurable load to the system and electricity draw. Use the NAS power cost calculator with an estimate for continuous NPU workload when budgeting for ongoing running costs.
Related reading: our NAS buyer's guide.
For specific hardware tiers, AU prices, and what to look for when buying a mini-PC for local AI use, see the mini-PC buying guide for local AI in Australia.
Related reading: our NAS explainer.
See also: our complete QNAP NAS Australia guide.
Does my Synology NAS have an NPU?
No. As of April 2026, Synology does not sell consumer or prosumer NAS hardware with an integrated NPU. Synology Photos performs recognition locally; depending on the workflow, processing can involve the NAS or the mobile device's NPU. Synology has not announced an NPU-equipped NAS model for the consumer or prosumer market.
What is the difference between an NPU and a VPU?
A VPU (Vision Processing Unit) is a subset of NPU design optimised specifically for computer vision tasks: image classification, object detection, video analysis. Intel's older Movidius chips were VPUs. Modern Intel NPUs (AI Boost in Core Ultra) handle a broader range of AI workloads including language model inference, not just vision. In practice, the terms are often used interchangeably in marketing material. For NAS use, supported applications, runtimes, operators, memory, and drivers matter alongside the accelerator label and TOPS rating.
Can Ollama use an NPU for acceleration?
Ollama supports CUDA (NVIDIA GPU) and ROCm (AMD GPU) acceleration natively. Ollama's official hardware documentation does not currently list Intel NPU acceleration. Separately, llama.cpp has an in-progress OpenVINO backend that can be compiled to target Intel NPUs. CPU inference remains the stable path on NPU-equipped hardware without a discrete GPU. This is expected to improve as the llama.cpp project matures its OpenVINO/IPEX-LLM backend support.
Is 11 TOPS enough for useful AI tasks on a NAS?
Yes, for targeted tasks. An 11-TOPS Meteor Lake NPU can accelerate only workloads supported by its drivers, runtime, operators, and application; TOPS alone does not establish performance for these specific NAS tasks. It is not sufficient for interactive 7B+ language model inference at useful speeds. Think of it as capable for ambient AI features, not a replacement for a GPU for conversational AI.
Will future Synology or UGREEN NAS have NPUs?
It is likely, though not confirmed. As Intel's Core Ultra platform becomes more widely adopted and AMD Ryzen AI series reaches embedded NAS form factors, NPU-equipped consumer NAS will become more common. UGREEN has shown roadmap interest in AI-capable hardware. Synology has been conservative with hardware changes historically. The next generation of prosumer NAS (2026-2027) is the likely timeframe for NPU-equipped options beyond QNAP.
How does an NPU compare to the AI features on Apple Silicon (M-series chips)?
Apple's M-series chips (M1, M2, M3, M4) have some of the most capable integrated NPUs available. M3 delivers ~18 TOPS from its NPU alone, with the full unified memory architecture enabling GPU inference on large models at speeds no NAS CPU can match. This is why Mac Mini M-series hardware has become popular for local AI setups. Apple Silicon and NAS NPUs differ substantially in software, memory architecture, GPU capability, and NPU performance; meaningful comparisons require a specific chip, model, and workload., though the architecture serves the NAS use case (always-on, low power, storage-integrated) that Apple Silicon Mini cannot replace.
Wondering which NAS hardware is actually ready for AI workloads in Australia? The AI NAS hardware requirements guide covers RAM, CPU, NPU, and storage needs for every AI use case.