System Overview#
Powered by the NVIDIA Grace Blackwell architecture, DGX Spark enables developers, researchers, and data scientists to prototype, deploy, and fine-tune large AI models on their desktop.
Flexible Deployment Options#
The DGX Spark is designed for maximum flexibility in how you use it:
Desktop Mode: Use with keyboard, mouse, and monitor for local development work
Network Appliance Mode: Operate headless for remote access and server-style deployments
Both modes are fully supported and equally capable, allowing you to choose the setup that best fits your workflow and environment.
Key Capabilities#
Your DGX Spark enables you to:
Run Inference: Deploy models for real-time AI applications
Develop AI Models: Train and fine-tune models with up to 200 billion parameters
Process Data: Handle large datasets with high-performance computing
Experiment Freely: Test new ideas without cloud computing costs
Scale Workloads: Connect multiple systems for larger projects
System Architecture#
The DGX Spark is built on NVIDIA’s Grace Blackwell architecture, providing:
Unified Memory: 128 GB of high-bandwidth memory for large models
High-Performance Computing: 20-core ARM64-based processor with integrated GPU
Advanced Connectivity: Wi-Fi 7, 10 GbE, CX7 NIC, and multiple I/O options
Compact Form Factor: 150mm x 150mm x 50.5mm desktop design
For detailed hardware specifications, see Hardware Overview.
Software#
Your system comes pre-configured with:
NVIDIA DGX OS: Optimized operating system for AI workloads
Development Tools: CUDA, cuDNN, and NVIDIA’s development ecosystem
Container Support: Docker and NVIDIA Container Runtime for easy deployment
NGC Integration: Access to NVIDIA’s container registry
For detailed software information, see Software.
Getting Started#
To begin using your DGX Spark:
Initial Setup: Follow the Initial Setup - First Boot to configure your system
Explore Examples: Try sample workloads to understand capabilities
Configure Development Environment: Set up your preferred tools and frameworks
Start Building: Begin your AI development projects
Note
For the most up-to-date tutorials and examples, visit https://build.nvidia.com/spark. This site is regularly updated with new content and serves as the primary resource for practical guides and use cases.