It is a next-generation desktop AI computer powered by the GB10 Superchip, built on NVIDIA’s Grace Blackwell architecture and capable of delivering 1 PFLOPS of AI performance.

For installation and technical information: https://build.nvidia.com/spark

• NVIDIA GB10 Grace Blackwell Superchip
• 5th-generation Tensor Core–enabled NVIDIA Blackwell GPU
• 20-core Arm CPU (10 Cortex-X925 + 10 Cortex-A725)
• 128 GB of unified system memory
• 1 PFLOP of tensor performance
• CPU+GPU memory coherence via NVLink-C2C (5× the bandwidth of PCIe Gen5)
• 4 TB NVME.M2 SSD
• 10 GbE Ethernet + ConnectX-7 Smart NIC
• Wi-Fi 7, Bluetooth 5.4, HDMI 2.1a output

Technological Innovations

Experience up to 1 petaFLOP of AI performance at FP4 precision with the NVIDIA Grace Blackwell architecture.

Run AI development and testing workloads on your desktop with models of up to 200 billion parameters, enabled by large and unified system memory.

The high-performance NVIDIA ConnectX network enables you to connect two NVIDIA DGX Spark systems, allowing you to work with AI models of up to 405 billion parameters.

Use a full-stack solution for generative AI workloads, including tools, frameworks, libraries, and pre-trained models.