NVIDIA DGX Spark Personal AI Supercomputer – GB10 Grace Blackwell, 128GB Unified Memory, 4TB NVMe SSD
Key Highlights
- 1-PetaFLOP FP4 AI Performance
- 128GB Coherent Unified Memory
- 4TB Self-Encrypting NVMe Storage
- Supports AI Models up to 200 Billion Parameters
Develop, fine-tune and run advanced artificial-intelligence models locally with NVIDIA DGX Spark—a compact personal AI supercomputer built on the same NVIDIA Grace Blackwell architecture used across modern AI infrastructure.
At its core is the NVIDIA GB10 Grace Blackwell Superchip, combining a 20-core Arm CPU with an integrated NVIDIA Blackwell GPU. Fifth-generation Tensor Cores support FP4 processing and deliver up to 1 petaFLOP of theoretical AI performance for generative AI, machine learning, computer vision, data science and model inference.
The processor features ten high-performance Arm Cortex-X925 cores and ten Cortex-A725 cores. This hybrid CPU architecture handles application processing, data preparation and system operations while the integrated Blackwell GPU accelerates AI workloads.
Unified Memory for Large AI Models
DGX Spark includes 128GB of LPDDR5x coherent unified system memory shared between the CPU and GPU. Unlike conventional systems with separate system RAM and graphics memory, its unified architecture allows both processors to work within the same large memory pool.
The 256-bit memory interface delivers bandwidth of up to 273GB/s. A single DGX Spark can support compatible AI models containing as many as 200 billion parameters, depending on model architecture, precision and quantization.
4TB Self-Encrypting Storage
The integrated 4TB NVMe M.2 solid-state drive provides high-speed storage for AI models, datasets, containers, development environments and project files.
Self-encryption provides hardware-level protection for stored information, making DGX Spark suitable for professional environments working with private datasets, proprietary models and sensitive development projects.
Complete NVIDIA AI Development Platform
NVIDIA DGX OS is an Ubuntu-based Linux operating system optimized for accelerated AI workloads. The system includes the NVIDIA AI software stack, allowing developers to begin working with established tools and frameworks without building the complete environment from scratch.
Supported development technologies include:
- NVIDIA CUDA
- PyTorch
- Jupyter
- Ollama
- NVIDIA TensorRT
- NVIDIA NIM microservices
- NVIDIA Blueprints
- Containerized AI development workflows
Projects developed locally can be transferred to compatible NVIDIA-accelerated cloud or data-centre environments with fewer architecture changes.
Dual-System Scalability
The integrated NVIDIA ConnectX-7 SmartNIC allows two compatible DGX Spark systems to be connected through a supported QSFP interconnect cable.
A two-system configuration provides 256GB of combined unified memory and can support compatible AI models containing up to 405 billion parameters. This makes it possible to expand beyond a single system without immediately moving development to large data-centre infrastructure.
The QSFP interconnect cable is sold separately.
Advanced Connectivity
DGX Spark includes 10Gb Ethernet, Wi-Fi 7 and Bluetooth 5.4 for high-speed wired and wireless networking.
Four USB Type-C connections support peripherals and external storage, with up to three DisplayPort outputs available through USB-C Alternate Mode. An HDMI 2.1a output supports compatible monitors and multichannel audio.
Compact Desktop Design
Measuring only 150 × 150 × 50.5mm and weighing approximately 1.2kg, DGX Spark delivers powerful AI capabilities in a small-form-factor desktop enclosure.
It is suited to AI developers, researchers, universities, data scientists, technology companies and businesses that require local AI experimentation while keeping models and data within their own environment.
Ideal Applications
- Generative AI development
- Large language model inference
- Model prototyping and fine-tuning
- Retrieval-augmented generation
- AI agent development
- Computer vision
- Data science and analytics
- Robotics development
- Educational AI laboratories
- Private and on-premises AI workloads
Box Includes
1 × NVIDIA DGX Spark Personal AI Supercomputer
1 × 240W External Power Adapter
1 × Regional Power Cord
1 × Quick Start Documentation
| Brand | NVIDIA |
| Product Name | NVIDIA DGX Spark |
| Edition | Founders Edition |
| Hardware Model | P4242 |
| Architecture | NVIDIA Grace Blackwell |
| Processor | NVIDIA GB10 Grace Blackwell Superchip |
| CPU Architecture | Arm |
| CPU Core Count | 20 Cores |
| CPU Configuration | 10 × Cortex-X925 + 10 × Cortex-A725 |
| GPU Architecture | NVIDIA Blackwell |
| GPU Type | Integrated |
| CUDA Cores | NVIDIA Blackwell Generation |
| Tensor Cores | Fifth Generation |
| RT Cores | Fourth Generation |
| AI Performance | Up to 1 petaFLOP FP4 |
| System Memory | 128GB LPDDR5x |
| Memory Architecture | Coherent Unified System Memory |
| Memory Speed | 4266MHz |
| Memory Interface | 256-Bit |
| Memory Bandwidth | 273GB/s |
| Storage Capacity | 4TB |
| Storage Type | NVMe M.2 SSD |
| Storage Security | Self-Encrypting Drive |
| Operating System | NVIDIA DGX OS |
| Operating-System Foundation | Ubuntu Linux |
| Software Platform | NVIDIA AI Software Stack |
| Single-System Model Support | Up to 200 Billion Parameters |
| Dual-System Model Support | Up to 405 Billion Parameters |
| Network Processor | NVIDIA ConnectX-7 SmartNIC |
| ConnectX-7 Connection Speed | Up to 200Gbps Per Port |
| QSFP Network Ports | 2 |
| Wired Network | 1 × RJ-45 10Gb Ethernet |
| Wireless Network | Wi-Fi 7 |
| Bluetooth | Bluetooth 5.4 |
| USB Connections | 4 × USB Type-C |
| DisplayPort Outputs | Up to 3 via USB-C Alternate Mode |
| HDMI Output | 1 × HDMI 2.1a |
| Audio Output | HDMI Multichannel Audio |
| Video Encoder | 1 × NVENC |
| Video Decoder | 1 × NVDEC |
| External Power Supply | 240W |
| GB10 TDP | 140W |
| Form Factor | Small-Form-Factor Desktop |
| Length | 150mm |
| Width | 150mm |
| Height | 50.5mm |
| Weight | 1.2kg |
| Recommended Operating Temperature | 5°C to 30°C |
| Operating Humidity | 10% to 90%, Non-Condensing |
