NVIDIA GB10 Grace Blackwell Architecture
The HP ZGX Nano G1n is powered by the NVIDIA GB10 Grace Blackwell Superchip, combining a Blackwell GPU with a 20-core Arm CPU. It delivers up to 1,000 TOPS of FP4 AI compute for local AI development, including prototyping, fine-tuning, and inference workloads.
128GB Coherent Unified Memory
128GB of onboard LPDDR5X unified memory is shared across the system for AI development and testing workloads. With memory bandwidth up to 273GB/s, the ZGX Nano is designed to work with AI models up to 200 billion parameters, depending on model architecture, precision, and workload.
Local AI Storage and Software
The 2TB PCIe Gen4 NVMe SSD provides fast local storage for models, datasets, and development files. NVIDIA DGX OS, based on Linux, and the NVIDIA AI software stack provide the environment for building, fine-tuning, and running AI workloads locally.
High-Speed Connectivity for AI Workflows
The system includes 10Gb Ethernet, two QSFP connections supporting 200Gbps signaling, three 20Gbps USB-C data ports, and HDMI 2.1a. These connections support high-speed networking, compatible AI and storage equipment, peripherals, and a local display.
Compact Desktop Workstation Design
The HP ZGX Nano fits dedicated AI compute into a 5.9 × 5.9 × 2.1-inch desktop footprint and weighs approximately 2.76 lb. It is air cooled and powered by a 240W external USB-C adapter, making it suitable for local development environments where space is limited.
Built for Local AI Development
Connect the ZGX Nano to a compatible Windows, Mac, or Linux computer to offload AI workloads without relying solely on cloud resources. HP ZGX Toolkit supports device discovery, local model serving, and export workflows for a more direct path from development to deployment.