NVIDIA DGX Spark GB10 AI Desktop — 128GB Unified Memory, 4TB NVMe

The NVIDIA DGX Spark GB10 AI desktop brings a compact personal AI workstation to developers, researchers, data scientists and advanced creators who want to explore demanding AI workloads locally.

Why the DGX Spark stands out

At its core is the NVIDIA GB10 Grace Blackwell Superchip, combining a 20-core high-performance Arm CPU with a Blackwell GPU and fifth-generation Tensor Core technology. NVIDIA quotes up to 1 petaFLOP of AI performance using FP4, giving this small desktop a clear focus on model development, inference and experimentation.

• 128GB of coherent, unified system memory

• 4TB of NVMe storage

• NVIDIA DGX OS for an AI-focused software environment

• Support for models of up to 200 billion parameters, according to the listed specifications

• ConnectX networking for linking two systems on compatible workloads

• Compact desktop form factor for a dedicated AI workspace

Who it is for

This NVIDIA DGX Spark is aimed at AI developers, machine-learning engineers, researchers, educators and technical teams building or testing generative AI applications. The unified memory capacity is particularly relevant for users who need to work with larger models without moving every task to a remote cloud environment.

Points to consider

The DGX Spark is a specialized AI computer rather than a conventional everyday desktop. Buyers should check software compatibility with its Arm-based platform, confirm the requirements of their preferred frameworks and verify the final configuration, price and availability before purchase.

Our verdict

The NVIDIA DGX Spark GB10 AI desktop is a distinctive choice for serious local AI development. Its combination of Blackwell acceleration, 128GB unified memory, 4TB NVMe storage and a compact design makes it especially compelling for professionals who need a focused AI workstation without building a larger server system.