Product of the Week: NVIDIA's DGX Spark Personal AI Computer
January 19, 2026
Sponsored Story
Today’s AI models are growing exponentially. The process of building, testing, and running AI models is continuously impacting developers and enterprises everywhere, as they increase in size and complexity and are integrated into various compute platforms.
To build better AI, it may be time to move away from relying on the cloud and data centers, and toward fast and locally processed advanced AI that gives developers the freedom and flexibility to prototype, fine-tune, and inference on a personal AI computer.
Designed to ease this challenge is the NVIDIA DGX Spark, built on NVIDIA GB10 GraceBlackwell Superchip and based on the NVIDIA Grace Blackwell architecture. The GB10 includes the NVIDIA Blackwell GPU with fifth-generation Tensor Core technology and an NVIDIA Grace CPU with 20-core Arm architecture in a compact desktop form factor.
The NVIDIA DGX Spark in Action
The NVIDIA DGX Spark provides up to 1 petaFLOP1 of AI compute using FP4 for large AI workloads, and 128 GB of integrated system memory, supporting developers’ versatility with up to 200 billion parameters. For even larger models, the solution includes NVIDIA ConnectX networking, allowing two NVIDIA DGX Spark supercomputers to connect, enabling inference on models with up to 405 billion parameters.
A step-up from a standard bus, the GB10 Superchip enables fast shared data using the NVIDIA NVLink-C2C, providing a CPU+GPU memory model with increased bandwidth than the PCIe Gen 5. To ensure large models run smoothly, the NVIDIA DGX Spark supports a 256-bit memory interface, a memory bandwidth of 273 GB/s, and 4 TB of NVMe M2 storage with self-encryption.
Getting Started with NVIDIA DGX Spark
Designed to support students, researchers, and developers with creative and practical endeavors, the NVIDIA DGX Spark supports those looking to improve existing skills and those who want to explore AI independently.
When you’re ready to get started, the DGX personal AI computer features an NVIDIA DGX OS with support for Ubuntu Linux and is preconfigured with the latest NVIDIA AI software stack. The solution also includes developer program access to NVIDIA NIM and NVIDIA Blueprints, providing users with tools such as PyTorch, Jupyter, and Ollama.
Additional Resources:
Tiera Oliver, Assistant Managing Editor for Embedded Computing Design, is responsible for web content edits, product news, and constructing stories. She develops content and constructs ECD podcasts, such as Embedded Insiders. Before working at ECD, Tiera graduated from Northern Arizona University, where she received her B.S. in journalism and political science and worked as a news reporter for the university’s student-led newspaper, The Lumberjack.
