GB10 (DGX Spark)

01Introduction

The launch of NVIDIA DGX Spark on March 18, 2025, marks a significant milestone in AI computing, bringing supercomputer-grade performance to the desktop. Originally developed as Project DIGITS, DGX Spark is powered by the NVIDIA GB10 Grace Blackwell

Superchip, designed to handle large-scale AI workloads, including fine-tuning and inference for models up to 200 billion parameters. With its compact form factor, high energy efficiency, and seamless cloud deployment capability, DGX Spark offers a new paradigm for researchers, developers, and data scientists.

02Architectural and Performance Overview

At the core of DGX Spark lies the NVIDIA GB10 Grace Blackwell

Superchip, which integrates a high-performance Blackwell GPU and a 20-core Arm CPU. The GPU leverages fifth-generation Tensor Cores, RT Cores, and CUDA cores to deliver an astonishing 1,000 AI TOPS of compute performance. This level of power enables users to run state-of-the-art AI reasoning models such as NVIDIA Cosmos Reason and GR00T N1 while maintaining a compact and power-efficient design.

The GB10 Superchip also incorporates NVIDIA’s NVLink-C2C interconnect technology, providing a unified memory model that significantly enhances data transfer speeds between the CPU and GPU. This architecture allows DGX Spark to achieve a memory bandwidth of

273 GB/s, ensuring optimal performance for memory-intensive AI tasks.

03Key Specifications

NVIDIA DGX Spark specifications:

• SoC – NVIDIA GB10
• CPU – 20-core Armv9 processor with 10x Cortex-X925 cores and 10x Cortex-A725 cores
• Architecture – NVIDIA Grace Blackwell
• GPU – Blackwell Architecture
• CUDA Cores – Blackwell Generation
• 5th Gen Tensor cores
• 4th Gen RT (Ray Tracing) cores
• Tensor Performance – 1000 AI TOPS (FP4)
• VPU – 1x NVENC, 1x NVDEC
• System Memory – 128 GB 256-bit LPDDR5x with 273 GB/s memory bandwidth
• Storage – 1 TB or 4 TB NVMe M2 SSD with self-encryption
• Display – HDMI 2.1a port
• Audio – HDMI multichannel audio output
• Networking
• 10GbE RJ45 port
• ConnectX-7 Smart NIC to connect two Spark DGX together with a speed of up to 200Gbps
• WiFi 7 and Bluetooth 5.3
• USB – 4x USB4 Type-C ports
• Power Consumption – 170 Watts
• Dimensions – 150 x 150 x 50.5 mm
• Weight – 1.2 kg

04Brands Offering GB10 Machines

NVIDIA designed the GB10 Superchip as the core platform, but it partnered with major PC manufacturers to bring it to market under their own branding. All these systems share the

same underlying GB10 Superchip and core specs (128GB unified memory, Blackwell GPU, 20-core Arm CPU) — the differences come down to chassis design, cooling, SSD capacity options, and price.

05AI Workloads and Use Cases

DGX Spark is tailored for AI developers, researchers, and data scientists working on various machine learning and deep learning tasks. Its capabilities extend across multiple AI

workloads, including:

• Generative AI & Large Language Models (LLMs): Capable of handling models with up to 200 billion parameters, enabling real-time inference and fine-tuning.
• Robotics & Simulation: Optimized for NVIDIA GR00T N1 robot foundation model, ideal for robotics research and training.
• Data Science & Analytics: High-speed processing for large datasets and AI-driven analytics.
• Edge AI & Autonomous Systems: Compact and energy-efficient, making it suitable for real-time AI inference at the edge.
• Cloud-Ready AI Development: Users can seamlessly transition from local development to NVIDIA DGX Cloud and other accelerated cloud platforms.

06Taking Advantage for GeoAI, LLM, and AI/ML

Given its high-performance AI computing capabilities, the DGX Spark is an ideal system for various advanced AI applications, including GeoAI, large language models (LLMs), and AI/ML research:

• GeoAI: The ability to process geospatial data at high speeds makes DGX Spark perfect for spatial analysis, predictive modeling, and real-time environmental monitoring.

Researchers can leverage its GPU power for deep learning-based remote sensing, geospatial feature extraction, and urban planning applications.

• Large Language Models (LLMs): The DGX Spark provides the computational power needed to fine-tune and deploy LLMs locally, reducing reliance on cloud-based

models while maintaining performance for natural language processing, machine translation, and chatbot development.

• AI/ML Research: For machine learning and AI researchers, DGX Spark offers an affordable yet powerful alternative to cloud-based AI infrastructure. It enables experimentation with AI models, neural network training, and high-performance computing workflows without the need for expensive cloud subscriptions.

The ability to run these workloads locally, combined with its compact form factor and

seamless cloud integration, makes the DGX Spark an attractive option for AI practitioners looking for a cost-effective yet powerful AI workstation.

07Competitive Edge and Market Positioning

The DGX Spark is poised to disrupt the AI hardware market by offering a desktop-class supercomputer with performance that rivals data center AI accelerators. Compared to traditional workstation GPUs or cloud-based solutions, DGX Spark provides:

• Cost-Effective AI Development: Eliminates the dependency on expensive cloud compute resources by allowing local AI processing.
• Compact and Power-Efficient Design: At just 1.2 kg and 170W power consumption, it is an ideal solution for home, lab, or office environments.
• Seamless Model Deployment: Native support for NVIDIA’s AI software stack and DGX Cloud integration ensures ease of deployment.
• Industry Adoption: With support from major PC manufacturers like ASUS, Dell Technologies, HP, and Lenovo, DGX Spark is expected to see broad adoption across academia and industry.

08Challenges and Considerations

While DGX Spark presents a compelling solution, there are some challenges and limitations to consider:

• Limited Expandability: The compact form factor restricts upgrades, unlike modular AI servers.
• NVIDIA Ecosystem Lock-in: Optimal performance relies on NVIDIA’s software stack, which may limit flexibility for users preferring open-source alternatives.

09Conclusion

• NVIDIA DGX Spark represents a groundbreaking step in AI computing, making high-performance AI processing accessible to individuals and small teams. With its

powerful Grace Blackwell Superchip, optimized memory architecture, and seamless cloud integration, DGX Spark is set to empower the next generation of AI research and development. While pricing and expandability remain concerns, its potential impact on AI innovation is undeniable. As AI workloads continue to evolve, DGX

Spark positions itself as a game-changer in the world of personal AI supercomputing.

Source : https://www.nvidia.com/en-us/products/workstations/dgx-spark/