Updated status: NVIDIA DGX Spark is shipping now!
The NVIDIA DGX Spark delivers a full AI supercomputer experience in a remarkably compact, desktop-friendly form. Powered by the GB10 Grace Blackwell Superchip, it offers up to 1 petaFLOP of AI performance with 128 GB of unified memory. It’s designed for developers, researchers, and data scientists to prototype, fine-tune, and run large models locally—without waiting on a datacenter.
Architecture, Performance and Platform Capabilities
Built on NVIDIA’s Grace Blackwell platform, DGX Spark integrates a 20-core Arm CPU with a Blackwell GPU, along with 5th-gen Tensor cores and 4th-gen RT cores. Its unified memory architecture (128 GB LPDDR5x) allows ultra-efficient data access across CPU and GPU. Connectivity includes Wi-Fi 7 and ConnectX-7 Smart NIC, making it both flexible and powerful for AI workloads.
AI Workflows, Practical Applications and Developer Advantages
DGX Spark is perfect for prototyping, fine-tuning, and inference of generative and reasoning models up to 200 billion parameters. With its full NVIDIA AI stack preinstalled, you can transition seamlessly from desktop experiments to production environments on DGX Cloud or larger DGX clusters. Scientists, developers, and edge-AI teams benefit from its unique mix of performance, portability, and development convenience.
