Powered by the NVIDIA GB10 Grace Blackwell Superchip, NVIDIA DGX Spark delivers up to 1 petaFLOP (FP4) of AI performance in a power‑efficient, compact workstation. With the NVIDIA AI software stack preinstalled and 128 GB unified memory, developers can prototype, fine‑tune, and inference reasoning LLMs (DeepSeek, Meta, NVIDIA, Google, Qwen) with up to 200B parameters locally—ideal for generative AI, RAG, multimodal, and agentic AI workflows. Feature NVIDIA GB10 Superchip Up to 1 petaFLOP (FP4) with Grace Blackwell—optimized for LLM inference, fine‑tuning, prompt engineering, and agent frameworks. 128 GB coherent unified system memory Run 200B‑parameter models locally with unified LPDDR5x memory for zero‑copy data access and fast context handling. NVIDIA ConnectX networking ConnectX enables pairing two DGX Spark systems to handle 405B‑parameter workloads—scalable multi‑node AI experiments. NVIDIA AI Software Stack Full‑stack generative AI: NVIDIA NIM microservices, CUDA, Triton Inference Server, TensorRT‑LLM, toolchains, libraries, and pre‑trained models. Specification Architecture NVIDIA Grace Blackwell GPU NVIDIA Blackwell Architecture CPU 20 core Arm, 10 Cortex-X925 + 10 Cortex-A725 Arm CUDA Cores NVIDIA Blackwell Generation Tensor Cores 5th Generation RT Cores 4th Generation Tensor Performance 1 PFLOP System memory 128 GB LPDDR5x, coherent unified system memory Memory Interface 256-bit Memory Bandwidth Up to 273 GB/s Storage 4 TB NVME.M2 with self-encryption USB 4x USB TypeC Ethernet 1x RJ-45 connector 10 GbE NIC ConnectX-7 NIC @ 200 Gbps Wi‑Fi WiFi 7 Bluetooth BT 5.4 w/LE Audio-output HDMI multichannel audio output Power Supply 240W Display Connectors 1x HDMI 2.1a NVENC | NVDEC 1x | 1x OS NVIDIA DGX OS System Dimensions 150 mm L x 150 mm W x 50.5 mm H System Weight 1.2 kg Hardware overview Application Documents Datasheet User Manual Quick Start Guide Part list Quick Start Guide x1 Power Supply x1 Power Cord x1