⚡ Powered by the NVIDIA GB10 Grace Blackwell Superchip, the NVIDIA DGX™ Spark workstation delivers extraordinary AI compute with up to 1 petaFLOP (FP4) of performance in a compact, energy‑efficient design. Preloaded with the NVIDIA AI software stack and equipped with 128 GB unified memory, developers can build, refine, and deploy large language models (DeepSeek, Meta, NVIDIA, Google, Qwen) up to 200B parameters locally—perfect for generative AI, multimodal reasoning, retrieval‑augmented generation, and agentic AI workflows. 🧩 Key Features 🚀 NVIDIA GB10 Superchip Delivers 1 petaFLOP FP4 compute with Grace Blackwell, tuned for LLM inference, fine‑tuning, and AI agent frameworks. 🧠 128 GB Unified Memory Run 200B‑parameter models seamlessly with LPDDR5x unified memory for fast context switching and zero‑copy access. 🌐 High‑Speed Networking ConnectX™ networking links two DGX Spark units to scale workloads up to 405B parameters. 🧱 NVIDIA AI Software Suite Includes NVIDIA NIM microservices, CUDA, Triton Inference Server, TensorRT‑LLM, libraries, and pre‑trained models for generative AI. 📊 Specifications 🏗 Architecture NVIDIA Grace Blackwell 🎮 GPU NVIDIA Blackwell Architecture 🖥 CPU 20‑core Arm (10 Cortex‑X925 + 10 Cortex‑A725) 🔢 CUDA Cores Blackwell Generation 🧮 Tensor Cores 5th Generation 💡 RT Cores 4th Generation ⚙️ Tensor Performance 1 PFLOP 🧠 Memory 128 GB LPDDR5x unified 🛣 Interface 256‑bit 🚅 Bandwidth Up to 273 GB/s 💽 Storage 4 TB NVMe M.2 (self‑encrypting) 🔌 USB 4× USB Type‑C 🌐 Ethernet 10 GbE RJ‑45 ⚡ NIC ConnectX‑7 @ 200 Gbps 📶 Wi‑Fi Wi‑Fi 7 🟦 Bluetooth BT 5.4 with LE 🔊 Audio HDMI multichannel output 🔋 Power 240W 🖥 Display 1× HDMI 2.1a 🎥 NVENC | NVDEC 1× | 1× 🧪 OS NVIDIA DGX™ OS 📐 Dimensions 150 × 150 × 50.5 mm ⚖️ Weight 1.2 kg 🛠 Hardware Overview 🎯 Applications 📚 Documents Datasheet User Manual Quick Start Guide 📦 Included Parts Quick Start Guide x1 Power Supply x1 Power Cord x1