NVIDIA DGX Spark is an AI supercomputer that changes the rules of the game in desktop computing. With the latest NVIDIA® GB10 Grace Blackwell chip, it achieves petaflop performance. Built for developers, AI researchers and data analysts, DGX Spark enables local development of AI solutions with unprecedented performance, scalability and advanced features-all in a compact chassis that fits on your desktop.

NVIDIA® Grace Blackwell Architecture

• NVIDIA Blackwell GPU and 20-core Arm processor Optimizes data preprocessing and organization, accelerating model tuning and enabling real-time inference with greater efficiency.
• These technologies are the most powerful in the world.
• NVLink®-C2C technology Provides seamless access to CPU and GPU memory with up to five times the bandwidth of PCIe 5.0 for ultra-fast data transfers.

Umatched AI computing power

• 1000 AI TOPS (FP4) tensor performance Provides lightning-fast performance when handling complex large-scale AI workloads.
• 128 GB LPDDR5x unified system memory Provides the high-capacity memory needed for smooth model development, fast experiments and efficient inference.

Running large language models locally, from 200 to 405 billion parameters

• Support AI models up to 200 billion parameters Run large language models locally while maintaining data security, low latency and cost control. Full AI workflow-from prototyping, to tuning, to inference.
• Connecting systems via NVIDIA ConnectX: bigger models and better performance High-performance NVIDIA ConnectX networking solutions allow you to connect two DGX Spark systems and work with AI models up to 405 billion parameters.