- Accelerate Local AI Workloads with 1000 TOPs AI Performance
- It's the DGX Spark, but Faster! Level1Techs Review
- Scale to 405B parameters by clustering two units with a QSFP/CX7 cable.
- Enable 4-unit clustering with a 400GbE switch for larger deployments
- NVIDIA GB10 Superchip Processor
- 20 Cores, 20 Threads
- NVIDIA Grace Blackwell GB10 Graphics
- 5th Gen Tensor Cores, 4th Gen RT Cores
- Supports Inferencing, RAG, Machine Learning Workloads
- NVIDIA NIM, NVIDIA Blueprints Technology
- Supports PyTorch, Jupyter, Ollama & More
- 128GB (On-Board) LPDDR5x Unified Memory
- Support for up to 200 Billion Parameter Models (Single Mode)
- Support for up to 405 Billion Parameter Models (ConnectX-7 Mode)
- 4TB M.2 2280 PCIe 5.0 x4 NVMe SSD
- 3x USB 10Gbps Type-C Ports
- 1x USB Type-C Power Port
- 1x HDMI 2.1a (With Multichannel Audio Output)
- 10 Gigabit Ethernet RJ-45 LAN Port
- 2x NVIDIA ConnectX-7 Teaming Ports
- WiFi 7 Dual-Band Adapter, Bluetooth 5.4
- 240W USB Type-C AC Adapter
- NVIDIA DGX OS Pre-Installed
- Supports Ubuntu Linux
- 1 Year Warranty


Read the NVIDIA Stacking/Clustering Guide