NVIDIA DGX Spark Brings Arm-Powered AI Supercomputing to Developer Desks
NVIDIA unveils the DGX Spark, a personal AI supercomputer powered by the Arm-based GB10 Grace Blackwell Superchip. The compact Linux system delivers up to one petaflop of AI performance to help developers prototype and run large models locally.
NVIDIA introduces the DGX Spark at CES 2025, a personal AI supercomputer that brings high-performance computing directly to developer desks. This Linux-based system relies on the NVIDIA GB10 Grace Blackwell Superchip, which combines an NVIDIA Grace CPU with a Blackwell GPU to deliver up to one petaflop of AI computing performance at FP4 precision.
The Grace CPU inside the DGX Spark utilizes leading-edge Arm architecture, featuring 10 Arm Cortex-X925 cores and 10 Cortex-A725 cores. These powerful CPU components connect to the latest-generation CUDA and fifth-generation Tensor Cores via NVLink-C2C interconnect, paired with 128GB of unified memory to handle demanding AI workloads.
This compact device enables AI researchers, data scientists, and students to prototype, fine-tune, and run inference on large AI models without relying solely on cloud infrastructure. By leveraging the ubiquitous Arm compute platform, the DGX Spark ensures that new AI models run efficiently at the edge and seamlessly integrate with existing consumer technology ecosystems.