NVIDIA DGX Spark is a desktop personal supercomputer designed from the ground up to build and run AI, powered by the Grace Blackwell architecture and the GB10 Superchip, enabling large-scale generative AI development at the desktop. Featuring fifth-generation Tensor Cores with FP4 support, it delivers up to 1 petaFLOP of AI compute and is equipped with 128 GB of coherent unified memory, capable of running 200-billion-parameter models locally. Dual-system interconnectivity via ConnectX networking scales support to 405-billion-parameter models, enabling local prototyping, fine-tuning, and inference.
It employs a 20-core high-performance Arm CPU paired with NVLink-C2C high-speed interconnect, delivering compute efficiency far exceeding traditional PCIe 5.0 platforms. The system comes preloaded with DGX OS and the complete NVIDIA AI software stack, supports mainstream tools including PyTorch and Jupyter, and enables seamless migration to cloud and data center environments, achieving "develop locally, deploy at scale."
The compact form factor measures just 150mm × 150mm × 50.5mm and weighs 1.2kg, equipped with 4TB self-encrypting NVMe storage and a 200Gbps high-speed network interface card, delivering data center-level AI capabilities in a desktop footprint—an ideal platform for developers, researchers, and students to drive AI innovation efficiently.
