- Brand: NVIDIA
- Model: DGX Spark 4TB Dual Pack
- UPC: 810152850398
- MPN: 940-54242-0008-000
- 1 PFLOPS FP4 AI Compute: Accelerates LLM inference, model prototyping, fine-tuning, and generative AI workloads
- Powered by NVIDIA’s advanced GPU architecture to accelerate AI training, inference, and data analytics
- 128GB LPDDR5x Unified Memory: CPU-GPU coherent memory helps run large AI models and heavy datasets smoothly
- 4TB Self-Encrypting NVMe SSD: High-speed secure storage for models, datasets, containers, and AI project files
- Dual DGX Spark Bundle: Two systems for scalable AI workflows, larger model experimentation, and professional AI labs
- ConnectX-7 + 10GbE Networking: Built for high-speed data transfer, multi-system AI workloads, and enterprise lab integration
- NVIDIA DGX Spark 4TB x 2 Units
- 1 Years Warranty
- Dispatch after 26-02-2025
Ask a Question About This Product
NVIDIA DGX Spark 4TB Dual Pack
Grace Blackwell AI Supercomputing Power for Your Desk
The NVIDIA DGX Spark 4TB Dual Pack is a powerful two-system desktop AI supercomputer solution designed for AI developers, researchers, data scientists, robotics teams, labs and advanced AI development environments. Powered by the NVIDIA GB10 Grace Blackwell Superchip, each DGX Spark delivers up to 1 petaFLOP FP4 AI performance, 128GB coherent unified system memory, 4TB self-encrypting NVMe M.2 storage, NVIDIA DGX OS and the NVIDIA AI software stack for local AI prototyping, fine-tuning, inference and data science workflows.
Key Highlights
A compact dual-system AI platform built for local model development, high-memory AI workflows and scalable desktop AI computing.
Grace Blackwell
NVIDIA GB10 Grace Blackwell Superchip for desktop AI supercomputing.
FP4 AI Power
Up to 1 petaFLOP FP4 AI performance per DGX Spark system.
Unified Memory
128GB LPDDR5x coherent unified system memory per unit.
NVMe Storage
4TB self-encrypting NVMe M.2 storage per DGX Spark system.
Dual Scaling
Connect two DGX Spark systems for AI models up to 405B parameters.
ConnectX-7
High-performance ConnectX-7 networking for Spark-to-Spark workloads.
DGX OS
NVIDIA DGX OS with AI-optimized software environment.
Compact Design
150mm x 150mm x 50.5mm small-form-factor desktop system.
Desktop AI Supercomputer for Developers and Researchers
DGX Spark brings NVIDIA Grace Blackwell AI acceleration into a compact desktop form factor. It is designed for users who want local AI compute for testing, validating, fine-tuning and deploying modern AI models without depending only on cloud resources.
Compact AI Workstation for Local Development
Designed for AI developers, data scientists, researchers and advanced labs, DGX Spark gives teams a local AI development platform for prototyping, validation, fine-tuning, inference and data science workloads.
Dual Pack Advantage: Scale Beyond One System
This Dual Pack includes two NVIDIA DGX Spark 4TB systems. Each unit delivers its own GB10 Grace Blackwell compute, 128GB unified memory and 4TB NVMe storage. When connected, two DGX Spark systems can work with AI models up to 405 billion parameters, making this bundle highly suitable for advanced AI labs and professional model development workflows.
Two-System AI Workflow
A dual-pack setup gives AI teams more flexibility for distributed experimentation, larger model testing, parallel workloads, multi-user lab setups and advanced local AI development environments.
ConnectX-7 High-Speed Networking
NVIDIA ConnectX-7 networking enables high-performance system-to-system connectivity, helping two DGX Spark systems work together for larger AI workloads and expanded development scenarios.
NVIDIA GB10 Grace Blackwell Superchip
At the heart of DGX Spark is the NVIDIA GB10 Grace Blackwell Superchip. It combines NVIDIA Blackwell GPU architecture, fifth-generation Tensor Cores, fourth-generation RT Cores and a 20-core Arm CPU to deliver a balanced AI compute platform for large language models, generative AI, data science and physical AI development.
Blackwell AI Architecture
NVIDIA Blackwell architecture brings fifth-generation Tensor Cores with FP4 support, helping accelerate modern AI workloads including inference, model validation and AI application development.
20-Core Arm CPU
The integrated 20-core Arm CPU includes 10 Cortex-X925 and 10 Cortex-A725 cores, providing a capable processing foundation alongside the Blackwell GPU for AI development workloads.
128GB Unified Memory for Large AI Models
DGX Spark uses 128GB LPDDR5x coherent unified system memory with 273GB/s memory bandwidth. This large shared memory design helps developers run AI development and testing workloads with models up to 200 billion parameters on a single DGX Spark system.
128GB LPDDR5x
Large coherent unified memory for AI model development and testing.
273GB/s
High memory bandwidth for demanding AI and data workflows.
Model Support
Supports AI models up to 200B parameters on a single system.
Dual Mode
Two connected systems support AI models up to 405B parameters.
Built for AI Prototyping, Fine-Tuning and Inference
NVIDIA DGX Spark is designed to accelerate real AI development workflows. Developers can build AI agents, prototype AI applications, fine-tune models, test inference pipelines, run data science projects and prepare workloads for migration to NVIDIA DGX Cloud or other NVIDIA-accelerated data center infrastructure.
AI Prototyping and Validation
Create, test and validate AI models locally before moving projects to cloud or data center infrastructure. This helps developers iterate faster while keeping early-stage work close to the desk.
Fine-Tuning up to 70B Parameters
With 128GB unified memory, DGX Spark is suitable for fine-tuning AI models up to 70 billion parameters, helping teams customize pre-trained models for specific use cases.
Inference up to 200B Parameters
DGX Spark can test, validate and run inference with AI models up to 200 billion parameters on a single unit, making it useful for local LLM, reasoning AI and application development.
Data Science and Edge AI
Ideal for computational analytics, machine learning, robotics, smart city, computer vision and edge AI development using NVIDIA AI frameworks such as Isaac, Metropolis and Holoscan.
Official NVIDIA DGX Spark Video
Watch the official NVIDIA DGX Spark unboxing video to understand the setup experience and compact desktop AI supercomputer form factor.
Video Link: https://www.youtube.com/watch?v=AamP-LbGHXQ
AI Software Stack and DGX OS
DGX Spark includes NVIDIA DGX OS and a preinstalled NVIDIA AI software stack to help AI projects get started faster. The software environment supports NVIDIA tools, frameworks, libraries, pre-trained models and NVIDIA NIM for generative AI development.
DGX OS
Stable Linux-based operating foundation optimized for NVIDIA AI hardware.
NVIDIA NIM
Supports NVIDIA NIM microservices for generative AI development workflows.
AI Stack
Includes NVIDIA tools, frameworks, libraries and model ecosystem support.
Developer Ready
Designed to help AI developers start building and testing quickly.
High-Speed Connectivity and Compact Hardware
Each DGX Spark system includes 4x USB Type-C, HDMI 2.1a, 10GbE RJ-45 Ethernet, ConnectX-7 networking, Wi-Fi 7 and Bluetooth 5.4. Its compact 150mm x 150mm x 50.5mm chassis and 1.2kg weight make it easy to place on a desk, in a lab or in a compact AI development workspace.
Modern I/O for AI Workstations
With USB Type-C, HDMI 2.1a, 10GbE Ethernet and wireless connectivity, DGX Spark is built for professional desktop usage, AI labs and development environments.
Small Form Factor, Serious Compute
The 150mm square footprint allows DGX Spark to deliver Grace Blackwell AI acceleration in a compact desktop design that fits easily into professional workspaces.
Ideal Use Cases
NVIDIA DGX Spark 4TB Dual Pack is designed for professionals and teams who need local AI computing capacity with scalable dual-system flexibility.
LLM Development
Prototype, test, validate and inference large language models locally.
Fine-Tuning
Customize pre-trained AI models for business, research and project-specific use.
Data Science
Run machine learning, analytics and computational data workflows at your desk.
AI Labs
Ideal for universities, AI research teams, development labs and training centers.
Robotics AI
Develop robotics and physical AI applications with NVIDIA AI frameworks.
Computer Vision
Build and validate smart vision, edge AI and video analytics solutions.
AI Agents
Develop AI agents, AI-assisted applications and autonomous workflows locally.
Local AI
Useful where local data privacy and controlled development environments matter.
Specification Overview
| Product Name | NVIDIA DGX Spark 4TB Dual Pack |
|---|---|
| Bundle Type | Dual Pack / 2 x NVIDIA DGX Spark 4TB Systems |
| Architecture | NVIDIA Grace Blackwell |
| Superchip | NVIDIA GB10 Grace Blackwell Superchip |
| GPU | NVIDIA Blackwell Architecture |
| CPU | 20-core Arm processor, 10 Cortex-X925 + 10 Cortex-A725 |
| CUDA Cores | Blackwell Generation |
| Tensor Cores | 5th Generation Tensor Cores |
| RT Cores | 4th Generation RT Cores |
| AI Tensor Performance | Up to 1 PFLOP FP4 per system |
| System Memory | 128GB LPDDR5x coherent unified system memory per system |
| Memory Interface | 256-bit |
| Memory Bandwidth | 273GB/s |
| Storage | 4TB NVMe M.2 self-encrypting storage per system |
| Total Bundle Storage | 2 x 4TB systems, storage operates per individual system |
| Single-System Model Support | AI models up to 200 billion parameters |
| Dual-System Model Support | AI models up to 405 billion parameters when two DGX Spark systems are connected |
| Networking | 1 x RJ-45 10GbE, NVIDIA ConnectX-7 Smart NIC |
| ConnectX Networking | ConnectX-7 NIC at 200Gbps |
| Wireless | Wi-Fi 7 |
| Bluetooth | Bluetooth 5.4 |
| USB | 4 x USB Type-C |
| Display Output | 1 x HDMI 2.1a |
| Audio Output | HDMI multichannel audio output |
| Video Processing | 1 x NVENC, 1 x NVDEC |
| Operating System | NVIDIA DGX OS |
| Power Supply | 240W external power supply per system |
| GB10 TDP | 140W |
| System Dimensions | 150mm L x 150mm W x 50.5mm H per system |
| System Weight | 1.2kg per system |
| Best For | AI development, LLM inference, fine-tuning, data science, robotics, edge AI and research labs |
FAQ
What is NVIDIA DGX Spark 4TB Dual Pack?
It is a two-system bundle of NVIDIA DGX Spark 4TB desktop AI supercomputers, designed for AI developers, researchers, data scientists and AI labs that need local AI computing and dual-system scalability.
How much memory does each DGX Spark include?
Each DGX Spark system includes 128GB LPDDR5x coherent unified system memory.
How much storage is included?
Each DGX Spark 4TB system includes 4TB NVMe M.2 self-encrypting storage. In this dual pack, there are two individual 4TB systems.
Can two DGX Spark systems work together?
Yes. NVIDIA specifies that high-performance ConnectX networking enables two DGX Spark systems to work with AI models up to 405 billion parameters.
What AI workloads is DGX Spark suitable for?
DGX Spark is suitable for AI prototyping, fine-tuning, inference, data science, AI agents, robotics, computer vision, edge AI and research workloads.
Does DGX Spark include NVIDIA software?
Yes. DGX Spark includes NVIDIA DGX OS and the NVIDIA AI software stack to help developers start AI projects quickly.
Build, Fine-Tune and Run AI Locally with Dual DGX Spark Power
Choose the NVIDIA DGX Spark 4TB Dual Pack for a compact, scalable and AI-ready desktop supercomputing platform built for local AI development, LLM testing, fine-tuning, data science and advanced research workflows.
