DGX Spark

NVIDIA DGX Spark is a compact, desktop-class "AI supercomputer" designed for developers and researchers to build, fine-tune, and run inference on large AI models locally.

Key Features and Specifications

• Architecture: Powered by the NVIDIA GB10 Grace Blackwell Superchip, which combines a 20-core Arm CPU with a Blackwell-architecture GPU.
• Performance: Delivers up to 1 petaFLOP of AI performance (FP4 with sparsity).
• Memory: Features 128 GB of coherent unified system memory.
• Model Capacity: Can run inference on models with up to 200 billion parameters locally. Two units can be linked via NVIDIA ConnectX networking to handle models up to 405 billion parameters.
• Form Factor: Extremely compact (approximately 6 x 6 x 2 inches, weighing 2.6 lbs), designed to sit on a desk.

Purpose and Use Cases

• Prototyping & Development: Allows developers to move the "make it work on my desk" phase of AI development out of the data center, reducing reliance on cloud-based resources for prototyping and testing.
• Software Stack: Comes with the NVIDIA AI software stack preinstalled—including DGX OS, container runtimes, and optimized frameworks (PyTorch, TensorFlow, JAX)—ensuring a ready-to-use environment.
• Deployment Path: Because it uses the same NVIDIA software stack found in larger data center and cloud infrastructures, code developed on a DGX Spark can be easily migrated to production-scale systems when needed.
• Autonomous Agents: Specifically positioned as a platform for building "always-on" local autonomous agents and AI-augmented applications.