# NVIDIA DGX Spark The power of NVIDIA Grace Blackwell on your desk. ### Overview ## Spark Something Big Powered by the NVIDIA GB10 Grace Blackwell Superchip, NVIDIA DGX Spark™ delivers up to one petaFLOP1 of FP4 AI performance in a power-efficient, compact form factor. With a preinstalled NVIDIA AI software stack and 128 GB of memory, developers can prototype, fine-tune, and deploy the latest reasoning AI models. Additionally, NVIDIA NemoClaw, part of the NVIDIA Agent Toolkit is an open source agent development platform for building, evaluating, and optimizing safer, long-running autonomous agents directly from the desktop. 1 Theoretical FP4 TOPS using the sparsity feature. ### NVIDIA Announces NemoClaw for OpenClaw Community NemoClaw adds security and privacy to run secure, always-on AI assistants on NVIDIA RTX PCs, DGX Station, and DGX Spark. [Read Press Release](https://nvidianews.nvidia.com/news/nvidia-announces-nemoclaw) ### DGX Spark and DGX Station with NVIDIA NemoClaw for Autonomous Agents The ultimate platform for locally developing and deploying secure, supercomputing-intelligent, always-on AI agents. [Read Blog](https://blogs.nvidia.com/blog/gtc-2026-news/#dgx-spark-station) * * * ### Run Autonomous Agents More Safely NVIDIA NemoClaw is an open source stack that adds privacy and security controls to OpenClaw. With one command, anyone can run always-on, self-evolving agents anywhere. [Learn More About NemoClaw](https://www.nvidia.com/en-us/ai/nemoclaw/) * * * ## Playbooks for Everyone From novice to expert, leverage our curated playbooks to get off to a running start on various projects. Playbooks contain step-by-step recipes to demonstrate the art of the possible. [Browse DGX Spark Playbooks](http://build.nvidia.com/spark) ### Getting Started With Your DGX Spark Here are some resources that can help you get up and running fast. [Quick Start Guide](https://www.nvidia.com/content/dam/en-zz/Solutions/dgx-spark/DGX-Spark-Quick-Start-Guide.pdf) \| [User Manual](https://docs.nvidia.com/dgx/dgx-spark/index.html) \| [Unboxing Video](https://www.youtube.com/watch?v=AamP-LbGHXQ&feature=youtu.be) \| [FAQ](https://forums.developer.nvidia.com/t/dgx-spark-gb10-faq/347344) \| [Support](https://www.nvidia.com/en-us/support/dgx-spark/) * * * ## Features ## NVIDIA GPU, CPU, Networking, and AI Software Technologies ### NVIDIA GB10 Superchip Experience up to 1 petaFLOP of AI performance at FP4 precision with the NVIDIA Grace Blackwell architecture. ### 128 GB of Coherent Unified System Memory Run AI development and testing workloads with AI models up to 200 billion parameters at your desktop with a large, unified system memory. ### NVIDIA ConnectX Networking High-performance NVIDIA ConnectX™ networking enables the connection of two NVIDIA DGX Spark systems to work with AI models of up to 405 billion parameters. ### NVIDIA AI Software Stack Utilize a full-stack solution for generative AI workloads, encompassing NVIDIA tools, frameworks, libraries, and pre-trained models including NVIDIA NIM. * * * ## Workloads ## Accelerate All AI Workloads Delivering the power of Grace Blackwell in a desktop-friendly size, NVIDIA DGX Spark is ideal for AI developer, researcher, and data scientist workloads. #### Prototyping Develop, test, and validate AI models and applications. With the NVIDIA AI software stack, NVIDIA DGX Spark provides a platform for developers to create, test, and validate AI models, as well as build AI agents, AI-augmented applications, and solutions. For final tuning or deployment, conveniently evaluate the work for eventual migration to the NVIDIA DGX cloud or other NVIDIA-accelerated data centers or cloud infrastructures. #### Fine-Tuning Fine-tune AI models up to 70 billion parameters. Improve the performance of pre-trained models by fine-tuning on NVIDIA DGX Spark. With 128GB of unified system memory, fine-tune models up to 70 billion parameters to customize AI models and solutions for specific needs and use cases. #### Inference Test, validate, and inference with AI models up to 200 billion parameters. Fifth-generation Tensor Cores with support for FP4 deliver up to 1 petaFLOP of AI computing performance, combined with 128GB of system memory, accelerate inference of [state-of-the-art AI models](http://build.nvidia.com/) to test, validate and deploy from your NVIDIA DGX Spark. #### Data Science High-performance data science at your desk. NVIDIA DGX Spark’s combination of 128GB of unified memory and 1 petaFLOP of parallel throughput maximizes performance of large, computationally complex data analytics and machine learning workflows at your desk. #### Edge Applications Develop edge applications with NVIDIA AI frameworks, including NVIDIA Isaac™, Metropolis, and many others. NVIDIA DGX Spark offers an exceptional platform for developing robotics, smart city, and computer vision solutions. NVIDIA frameworks include [Isaac](https://developer.nvidia.com/isaac), [Metropolis](https://www.nvidia.com/en-gb/autonomous-machines/intelligent-video-analytics-platform/), and [Holoscan](https://developer.nvidia.com/holoscan-sdk), enabling developers to leverage the power of NVIDIA DGX Spark to develop edge applications quickly. ### NVIDIA AI Enterprise—DGX Spark A cloud-native suite of software tools, libraries, and frameworks accelerating production-grade AI development. Combines enterprise-grade security, optimized performance, and enterprise-level support to streamline prototype-to-production for next-gen agentic AI. [Learn More](https://www.nvidia.com/content/dam/en-zz/Solutions/dgx-spark/workstation-print-gtc26-nvaie-spark-solution-overview-5004550-r7.pdf) ### DGX Spark AI Software DGX Spark includes the NVIDIA AI software stack preinstalled with support for the NVIDIA AI software ecosystem to get AI projects up and running quickly. [Learn More](https://developer.nvidia.com/topics/ai/dgx-spark) ## NVIDIA DGX Spark Specifications | | | | --- | --- | | Architecture | NVIDIA Grace Blackwell | | GPU | Blackwell Architecture | | CPU | 20-core Arm, 10 Cortex-X925 + 10 Cortex-A725 Arm | | CUDA Cores | Blackwell Generation | | Tensor Cores | 5th Generation | | RT Cores | 4th Generation | | Tensor Performance1 | Up to 1 PFLOP FP4 | | System Memory | 128 GB LPDDR5x, coherent unified system memory | | Memory Interface | 256-bit | | Memory Bandwidth | 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 | | Audio-output | HDMI multichannel audio output | | Power Supply | 240 W | | GB10 TDP2 | 140 W | | 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 | 1 Theoretical FP4 TOPS using the sparsity feature. 2 TDP: Thermal Design Power of the GB10 chip including CPU and GPU [View Datasheet](https://nvdam.widen.net/s/tlzm8smqjx/workstation-datasheet-dgx-spark-gtc25-spring-nvidia-us-3716899-web)