QAI-h1290FX

A GPU-ready edge AI storage server supporting NVIDIA® GPUs, U.2 NVMe SSDs, and 25GbE connectivity—designed for on-premises AI, virtualization, and compute-intensive workloads.

QNAP QAI-h1290FX

TechRadar Pro Picks Awards Winner for CES 2026

QAI Ideal applications

• Internal Chatbot & Knowledge Base

  Deploy private ChatGPT-like bots using AnythingLLM or OpenWebUI. Securely connect to internal documents for employee Q&A, policy search, and training support—no internet required.

  

• Private RAG Search Engine

  Run Retrieval-Augmented Generation (RAG) locally with full control over data. Enable natural-language document search across contracts, reports, and archives—ideal for legal, finance, and enterprise teams.

  

• AI Inference & Content Generation

  Use Stable Diffusion or ComfyUI for image generation, or deploy custom models for video tagging, document summarization, and medical analysis. Benefit from GPU acceleration and all-flash storage.

  

Enterprise-Grade Edge AI and High-Performance Computing

QAI-h1290FX is more than a storage system — it is a compute-ready, enterprise-grade edge computing platform. Built on a high-performance computing architecture, it supports configurable NVIDIA® RTX™ Pro Blackwell GPUs, making it well-suited for large language model (LLM) inference, image generation, RAG search, and a wide range of compute-intensive and virtualized workloads.

Whether for AI inference, research and development, data analytics, or enterprise applications requiring high core counts and sustained performance, a single desktop-class enterprise platform can deliver outstanding compute efficiency and data security entirely on-premises.

Maximum AI Compute Performance (Optional GPU Configuration)

• 3511 AI TOPS (FP4)
  333 TFLOPS (RT Core)

GPU-Ready Architecture — Supporting NVIDIA® RTX™ Pro Blackwell

QAI-h1290FX features a GPU-ready architecture designed to support NVIDIA® RTX™ Pro Blackwell GPUs. Built on the Blackwell architecture, it supports acceleration technologies such as CUDA, TensorRT, and the Transformer Engine, making it well-suited for modern AI and GPU-accelerated computing workloads.

From large language model (LLM) inference and computer vision to generative AI and other GPU-accelerated professional applications, workloads can be deployed and executed entirely on-premises—delivering strong performance while maintaining data privacy and full system control. The platform can also operate as a CPU-centric high-performance computing system, supporting virtualization and a wide range of enterprise computing scenarios.

• 

  Flagship

  NVIDIA® RTX™ Pro 6000 Blackwell

  • 96 GB GDDR7 ECC memory
  • 24,064 CUDA cores, 752 Tensor cores, 188 RT cores
  • 125 TFLOPS (FP32), up to 4000 AI TOPS
  • 1,792 GB/s memory bandwidth
  • 300W power consumption
  • PCIe 5.0 x16 interface

  • Designed for large-scale LLMs (up to 70B+ parameters), multi-model parallel workloads, and high-throughput AI pipelines

• 

  Performance

  NVIDIA® RTX™ Pro 4500 Blackwell

  • 32 GB GDDR7 ECC memory
  • 10,496 CUDA cores, 328 Tensor cores, 82 RT cores
  • 54.94 TFLOPS (FP32)
  • 896 GB/s memory bandwidth
  • 200W power consumption
  • PCIe 5.0 x16 interface

  • Ideal for mid-sized LLMs (up to ~30B parameters), retrieval-augmented generation (RAG) pipelines, and high-performance graphics applications

NVIDIA® RTX™ Pro Blackwell Series — Redefining AI and High-Performance Computing Workflows

The NVIDIA® RTX™ Pro Blackwell series GPUs are purpose-built for high-intensity AI, compute, and creative workloads, combining the next-generation Blackwell architecture with ultra-fast GDDR7 ECC memory. This delivers a level of compute performance and VRAM capacity on a single professional GPU that previously required multiple consumer-grade graphics cards.

With support for up to 96GB of VRAM and enhanced AI acceleration capabilities, the RTX™ Pro Blackwell series is ideal for advanced LLMs, generative models, data analytics, and complex 3D visualization and professional compute workflows.

• 5th Gen Tensor Cores

  Up to 3× faster AI processing with FP4 precision and DLSS 4 acceleration.

• 4th Gen RT Cores

  2× faster ray tracing for photorealistic rendering and real-time simulations.

• Up to 96GB GDDR7 Memory

  Massive capacity and bandwidth to handle complex AI and graphics tasks.

• Power consumption and performance balance

  Powerful AI performance with a power consumption of only 300W

Powered by Server-Class AMD EPYC™ Processors for High-Performance Compute

QAI-h1290FX is built on a server-class AMD EPYC processor platform, delivering high core counts and massive multithreaded performance.
Designed for long-term, stable operation under highly parallel workloads, it is well suited for virtualization, multithreaded computing, data processing, and edge computing scenarios—while also supporting AI inference and a wide range of compute-intensive applications.

• CPU

  Server-class AMD EPYC™ processors
  High core-count and multithreaded architecture

  Learn more spec

• Memory

  128 GB RDIMM DDR4 ECC memory (Expandable up to 1 TB)

采用 QuTS hero 操作系统

专为企业所打造的 QuTS hero 操作系统采用高可靠 ZFS 文件系统，为关键数据存储提供强悍数据安全与系统稳定，更拥有专注于提升 SSD 性能与寿命的先进技术，满足企业对于高性能与可靠度的严苛要求。探索 QuTS hero 操作系统了解 QuTS hero 最新功能

• 自动修复损坏数据

  ZFS 支持静态数据自我修复能力 (Self-healing)，可侦测并自动修复损坏数据，确保数据完整性。

• 数据不可变性

  通过创建 WORM（Write Once, Read Many）文件夹来防范勒索软件，防止数据被覆盖、修改或删除，同时启用不可变备份。

• 写入保护避免数据毁损

  ZIL 断电保护机制在 NAS 遭遇无预警断电后启用写入保护，避免修改中的数据毁损。

• 资料缩减

  在线数据重复删除 (Inline data deduplication1) 节省存储空间占用，让可用硬盘容量保持优化的状态。

• 防止多个 SSD 同时故障

  自主研发 QSAL 专利算法可自动定期侦测 RAID 层级的 SSD 寿命，防止多个 SSD 同时毁损，避免 RAID 崩溃造成数据遗失。

Remote access to on-prem AI – anytime, anywhere

Create a seamless hybrid work environment with multiple remote access options offered by QNAP. Whether you're managing AI applications or accessing files, the QAI-h1290FX ensures you're always connected—without compromising security.

Direct or relay access options

• myQNAPcloud DDNS:
  Access your QuTS hero interface from anywhere via a custom domain, without remembering IP addresses.
• myQNAPcloud Link:
  Establishes a secure relay connection through QNAP servers—no need to open router ports or modify firewall settings.
• VPN Server Support:
  Set up a private VPN using QVPN Service, enabling secure encrypted tunnels for full network access.

Whether you're fine-tuning your LLM container setup, reviewing inference logs, or collaborating across locations, the QAI-h1290FX offers reliable access to your on-prem AI environment from any device, anytime.

A More Powerful Container Station: A New Experience in AI Application Deployment

To promote practical AI adoption, the QAI series integrates Container Station with a wide selection of AI application templates. These templates support one-click deployment of popular AI tools and frameworks, with regular updatesfrom QNAP to ensure access to the latest technologies.

Whether you're new to AI or looking to move workloads on-premises, QAI makes it easy to explore AI, reduce costs, enhance data security, and even develop custom AI tools to boost business innovation.

Containerized AI deployment made simple

Enhance your AI infrastructure with seamless container integration. Explore Container Station

AI-centric container environment

The QAI-h1290FX comes with Container Station, enabling the deployment of Docker and LXD containers. As most AI tools today are delivered as containerized applications, QAI-h1290FX provides a direct and efficient way to deploy models such as LLM, RAG Search, image generation (e.g. Stable Diffusion), or knowledge base engines like AnythingLLM—all without complex setup.

Persistent data storage for AI models

With Docker Volume support, the QAI-h1290FX allows containers to mount shared folders from the NAS, ensuring persistent storage even after container rebuilds. This is especially beneficial for AI workloads involving large model files, training data, or logs. No more worrying about losing key data during updates or version switching.

Make GPU acceleration simple and accessible for all AI workloads

The QAI-h1290FX removes the complexity of configuring GPU resources in Docker environments. With Container Station’s intuitive interface, when creating a new container, simply select the desired NVIDIA® GPU from the dropdown menu, and the container will instantly gain access to GPU compute capabilities. Whether you're deploying LLMs, running image generation, or executing deep learning inference, the QAI-h1290FX ensures the process is seamless and efficient.

Redefining creativity with AI-powered visual design

ComfyUI empowers artists, designers, and content creators with a powerful, modular interface for AI-driven image and video creation. Through its intuitive node-based design and support for advanced models like Stable Diffusion, users can effortlessly generate, transform, and animate visual content. Combined with GPU acceleration and flexible workflows, ComfyUI lowers the barrier to complex visual design—unlocking unprecedented creative freedom.

• Text-to-Image Generation

  Turn simple text prompts into high-quality visuals using Stable Diffusion and custom models. Create concept art, product designs, or stylized scenes—perfect for creative and marketing teams looking to accelerate ideation.

• Style Customization with Multiple Models

  Choose from a wide range of supported checkpoints and LoRA models to achieve distinct visual styles. Whether it's photorealism, anime, or painterly art, every output can match your creative direction.

• Image-to-Video Animation

  Bring still images to life by creating smooth animations and cinematic sequences. ComfyUI supports frame interpolation and style transfer to turn static visuals into engaging motion graphics.

• Visual Workflow Editing for Custom Pipelines

  Build and refine complex visual pipelines using ComfyUI’s intuitive node editor. Connect models, effects, and processing steps in a visual interface that empowers experimentation and creative control & no coding required.

Real-world AI performance – measured on QAI-h1290FX

AI deployment performance is validated through real-world benchmark data. Under a high-end GPU test configuration, QAI-h1290FX was fully evaluated with the NVIDIA® RTX™ PRO 6000 Blackwell Max-Q Workstation GPU, verifying its performance in on-premises AI inference and enterprise deployment scenarios.

Ollama LLM Inference Benchmark (Rapid Deployment)

Leveraging the GPU acceleration capabilities of the Blackwell architecture, QAI-h1290FX can run a wide range of large language models locally via Ollama.
Ollama enables rapid deployment and simplified management, making it well suited for proof-of-concept (PoC) projects, single-user environments, and small to mid-scale use cases such as RAG-based search, AI assistants, and offline inference.

Mode	Token/sec	VRAM Usage
gpt-oss:120b (MXFP4)	90 Token/sec	~63GB
deepseek-r1:70b (q4_K_M)	24 Token/sec	~41GB
qwen3:32b (q4_K_M)	46 Token/sec	~21GB
gemma3:27b (q4_K_M)	54 Token/sec	~19GB
deepseek-r1:8b (q4_K_M)	140 Token/sec	~7GB
qwen3:8b (q4_K_M)	172 Token/sec	~7GB

vLLM Concurrent Inference Benchmark (Enterprise-Grade Throughput)

To address multi-user and high-concurrency AI service requirements, QAI-h1290FX also supports deployment with the vLLM inference engine.
Compared to single-request–oriented inference approaches, vLLM significantly improves GPU utilization and overall throughput through Paged Attention and efficient scheduling mechanisms. This makes it particularly suitable for enterprise AI services, multi-user RAG systems, and API-based AI applications.

Under the same GPU configuration, vLLM demonstrates more consistent latency characteristics and higher tokens-per-second throughput in concurrent request scenarios, making it ideal for production environments and long-running enterprise AI deployments.

Tested Large Language Model : deepseek-ai/DeepSeek-R1-Distill-Qwen-7B (Hugging Face)

Thread	Total Token/sec	avg Token/Thread/Sec
1	79 Token/sec	79 Token/sec
2	166 Token/sec	83 Token/sec
5	410 Token/sec	82 Token/sec
10	688 Token/sec	68.8 Token/sec
20	810 Token/sec	40.5 Token/sec
50	850 Token/sec	17 Token/sec

Tested Large Language Model : openai/gpt-oss-20b (Hugging Face)

Thread	Total Token/sec	avg Token/Thread/Sec
1	218 Token/sec	218 Token/sec
2	340 Token/sec	170 Token/sec
5	1045 Token/sec	209 Token/sec
10	880 Token/sec	88 Token/sec
20	600 Token/sec	30 Token/sec

Benchmark results are based on the average performance from multiple prompts across various domains, such as mathematics, physics, computer science, and philosophy.

Unlock AI potential with practical use cases

From document automation to creative workflows and system-wide automation, the QAI-h1290FX empowers every department to apply AI in meaningful, measurable ways—securely hosted on your own infrastructure.

No cloud lock-in, no complex setup—just real results driven by local LLMs, secure containers, and integrated QNAP features.

Smart HR Assistant – Internal Policy Chatbot

Build an internal Q&A assistant using AnythingLLM + Ollama. Upload HR documents such as employee handbooks, leave policies, and benefits to the QAI-h1290FX. Employees can ask natural questions like:

“How do I apply for family care leave?”

The system performs local RAG search and replies instantly, reducing HR workload and improving response time.

Creative Team AI Studio – Image Generation Hub

Design teams use Stable Diffusion and ComfyUI deployed on the QAI-h1290FX to generate promotional images, mockups, or stylized artwork via prompt input.
Thanks to GPU acceleration and persistent NAS storage, designers get fast and reproducible results—no more starting from scratch.

AI Co-Pilot for Developers – Docs, Code & Summaries

Engineering teams run LLMs like Qwen or Llama in Ollama on the QAI-h1290FX to assist with spec writing, code review, and technical translations.
Upload API docs or whitepapers, then chat with the model for clarifications, summaries, or even markdown formatting—completely offline and secure.

n8n + NAS Automation – Trigger AI from Anywhere

With n8n installed on the QAI-h1290FX, you can automate tasks that integrate AI with business operations. For example:

• When receiving a support email, trigger LLM to summarize and provide suggested responses, saving them in drafts.
• Connect to QNAP MCP to regularly check device status, and provide suggestions to users after AI analysis.
• Analyze NAS backed-up emails and conversations with LLM to check if there is any inappropriate content

AI Docker applications on QAI-h1290FX

Run powerful AI solutions via Container Station & GPU integration.

• AnythingLLM

  Enables teams to deploy customizable, private Large Language Models in a secure environment. Integrate various data sources and interact with your LLM for tailored Q&A, document parsing, and more—ideal for businesses prioritizing privacy and control.

• OpenWebUI

  A versatile and user-friendly web interface for interacting with local or remote LLMs. OpenWebUI supports chat, prompt engineering, and conversation management in a sleek UI, simplifying AI adoption for users and developers alike.

• Ollama

  Ollama makes running and experimenting with open-source language models effortless. With GPU support, developers can efficiently test, fine-tune, and serve powerful LLMs on-premises, keeping data secure and workflows agile.

• ComfyUI

  An advanced, modular UI for generative AI workflows. ComfyUI allows users to create, visualize, and customize complex AI pipelines—including image synthesis and text-to-image tasks—using an intuitive drag-and-drop interface powered by GPU acceleration.

• n8n

  n8n is a workflow automation tool that connects to hundreds of services, including AI models. Integrate business operations, trigger AI tasks, and automate repetitive processes—all within a secure, containerized environment, with support for powerful extensions and custom logic.

• Whisper

  Whisper delivers versatile, multilingual speech processing in a single, open-source model. It accurately transcribes, translates, and identifies languages from diverse audio sources, enabling seamless voice-to-text workflows across industries. With support for on-premises deployment, teams can maintain data security while integrating advanced speech capabilities into their applications.

Transform your NAS into an AI-powered knowledge hub

Empower enterprise users to search, understand, and retrieve information faster than ever—powered by Qsirch and next-generation RAG Search. The QAI-h1290FX brings intelligence to your documents while keeping everything secure and on-premises.

Qsirch – Smart Full-Text Search

Qsirch offers blazing-fast full-text search across all your NAS files—PDFs, Office docs, emails, and more. With advanced filters, indexing, and preview functions, users can locate relevant documents instantly across terabytes of data. Learn more

RAG Search – AI-Powered Contextual Answers

RAG (Retrieval-Augmented Generation) Search takes things further by combining Qsirch with a local Large Language Model (LLM). When users ask a question, the system first retrieves relevant files via Qsirch, then generates an accurate, natural language answer using the on-device AI. Learn more

On QAI-h1290FX, RAG Search is fully on-premise:

• No cloud APIs
• No data leaks
• Complete privacy and compliance

5 年标准保固

QAI-h1290FX 享有多至 5 年的免费产品保固，为您的企业组织提供更安心的保障。

QNAP 保固服务

NAS 恢复指南

Edge AI Storage Server Comparison：
QAI-h1290FX vs. Other AI NAS vs. AI Workstation

Features	QNAP QAI-h1290FX	Other NAS brand (AI NAS)	AI Workstation
Core AI Operating Model	On-prem LLM + Private RAG (models, data, and inference run locally)	Cloud-dependent AI (cloud search, labeling, or API-based services)	Local compute only (limited integration of data and AI workflows)
AI Compute	Supports enterprise NVIDIA RTX PRO 6000 workstation GPU (96GB VRAM)	No GPU, or iGPU / NPU only	Varies by model, typically a single GPU
Model Capability	Runs 70B+ LLMs, SDXL, and large open-source models	Limited to smaller models or cloud AI	Large models are costly and VRAM-limited
AI Experience	Built-in container management with ready-to-deploy AI templates (LLM / RAG / automation)	Complex setup: Docker/YAML tuning + manual GPU configuration	Fully manual setup: OS, drivers, and container stack required
Core NAS Capabilities	Enterprise NAS with ZFS, RAID, snapshots, WORM, and backup	Limited or consumer-oriented NAS features	No full NAS management or data protection
Deployment Fit	Quiet tower design—ideal for offices and studios	Varies by model	Large, loud, heat-intensive—often needs a dedicated room/rack
Total Value (TCO)	All-in-one AI + NAS with the lowest deployment and management overhead	Separate investments for AI and storage	Compute-focused system with limited data value built in