NVIDIA DGX SPARK
A Desktop Grace Blackwell AI Supercomputer
DGX Spark redefines what’s possible on the desktop, packing 128 GB of unified system memory and native support for the FP4 data format—empowering developers to train, fine-tune, and inference AI models of up to 200 billion parameters. Built from the ground up for the AI era, DGX Spark bridges the gap between local prototyping and scalable deployment, allowing seamless migration to DGX Cloud or any accelerated infrastructure with virtually no code changes. Its compact, plug-and-play design fits into any workspace, yet delivers a radically different architecture from traditional AI workstations—making it the ideal platform for developers who need powerful AI capabilities without compromising on space or cloud flexibility.
Book your NVIDIA DGX Spark ?
With limited availability, early reservation ensures you’ll be among the first to experience the power of desktop-scale large model development. Don’t miss the opportunity to bring datacenter-class performance to your workspace—register your interest now and be ready for what’s next.
Features
NVIDIA GB10 Superchip
It uses NVLink™-C2C technology to provide a coherent CPU and GPU memory model with five times the bandwidth of PCIe Gen 5.
NVIDIA ConnectX Networking
With the NVIDIA ConnectX-7 networking, the user can connect two NVIDIA DGX Spark systems together, enabling seamless collaboration on AI models with up to 405 billion parameters.
Blackwell GPU
The NVIDIA Blackwell GPU delivers 1 PetaFLOP of FP4 performance, featuring fifth-generation Tensor Core technology for cutting-edge AI acceleration.
Grace CPU
The Grace CPU features a high-performance 20-core Arm architecture, combining ten powerful Cortex-X925 cores with ten efficient Cortex-A725 cores.
NVMe Storage
The NVIDIA DGX Spark offers up to 4TB of ultra-fast NVMe storage, providing ample space and rapid data access essential for managing large AI datasets and accelerating development workflows.