NVIDIA DGX Spark

The increasing size and complexity of generative AI models is making development efforts on local systems challenging. Prototyping, tuning, and inferencing large models locally requires large amounts of memory and significant compute performance. As enterprises, software providers, government agencies, startups, and researchers staff up AI development efforts, the need for AI compute resources continues to grow.

NVIDIA DGX Spark Personal AI Supercomputer delivers a petaflop of AI performance from a power-efficient, compact form factor. With the preinstalled NVIDIA AI software stack, 1000 TOPS of performance, and 128GB of memory, developers can prototype, fine-tune, and inference large AI models of up to 200 billion parameters locally and seamlessly deploy to the data centre or cloud.

NVIDIA DGX Spark is a comprehensive solution, providing the hardware along with expert support and integration services to ensure seamless deployment and optimal performance for AI projects. This makes it an accessible and efficient tool for those looking to advance their AI research and applications.

For more information, see the: Product Datasheet

Built on NVIDIA Grace Blackwell

At the heart of NVIDIA DGX Spark is the new NVIDIA GB10 Grace Blackwell Superchip based on the NVIDIA Grace Blackwell architecture optimised for a desktop form factor. GB10 features a powerful NVIDIA Blackwell GPU with fifth-generation Tensor Cores and FP4 support, delivering up to 1000 TOPS of AI compute. GB10 includes a high-performance Grace 20-core Arm CPU to supercharge data preprocessing and orchestration, speeding up model tuning and real-time inferencing. The GB10 Superchip uses the NVLink™-C2C to deliver a CPU+GPU coherent memory model with 5X the bandwidth of PCIe Gen 5.

NVIDIA GB10 Superchip

Experience up to 1000 AI TOPS of AI performance at FP4 precision with the NVIDIA Grace Blackwell architecture.

Coherent Unified System Memory

Run AI development and testing workloads with AI models up to 200 billion parameters at your desktop with large, unified system memory.

NVIDIA ConnectX Networking

High-performance NVIDIA Connect-X networking enables connecting two NVIDIA DGX Spark systems together to work with AI models up to 405 billion parameters.

NVIDIA AI Software Stack

Use a full-stack solution for generative AI workloads, including tools, frameworks, libraries, and pretrained models.

Work With the Latest Generation of Large-Parameter Generative AI Models

With 128 GB of unified system memory and support for the FP4 data format, DGX Spark can support AI models of up to 200B parameters, enabling AI developers to prototype, fine-tune and inference the latest generation of AI reasoning models—such as DeepSeek R1 distilled versions up to 70 billion parameters—on their desktop. With built-in NVIDIA ConnectX™ network technology, two DGX Spark systems can be connected to work on even larger models such as Llama 3.1 405B.

Develop Locally, Deploy Anywhere at Scale

DGX Spark provides developers with a powerful, experimentation ground for prototyping models and AI applications, freeing up valuable compute resources in their cluster environments better suited for training and deploying production models. Leveraging the NVIDIA AI platform software architecture makes it possible for DGX Spark users to seamlessly move their models from their desktop to DGX Cloud or any accelerated cloud or data center infrastructure with virtually no code changes, making it easier than ever to prototype, fine-tune, and iterate.

Key Features

Built on NVIDIA GB10 Grace Blackwell Superchip
NVIDIA Grace Blackwell GPU with fifth-generation Tensor Core technology
NVIDIA Grace Blackwell CPU with 20- core high-performance Arm architecture
Up to 1000 TOPS of AI performance using FP4
128 GB of coherent, unified system memory
Support for up to 200 billion parameter models
NVIDIA ConnectX™ networking to link two systems to work with models up to 405 billion parameters
Up to 4 TB of NVMe storage
Compact desktop form factor

Desktop AI Compute Demands

The increasing size and complexity of generative AI models is making development efforts on local systems challenging. Prototyping, tuning, and inferencing large models locally requires large amounts of memory and significant compute performance. As enterprises, software providers, government agencies, startups, and researchers staff up AI efforts, the need for AI compute resources continues to grow.

200B Parameter Models on Your Desk

NVIDIA DGX™ Spark is part of a new class of computers designed from the ground up to build and run AI. Powered by the NVIDIA GB10 Grace Blackwell Superchip and based on the NVIDIA Grace Blackwell architecture, NVIDIA DGX Spark delivers up to 1000 TOPS of AI performance to power large AI workloads. With 128 GB of unified system memory, developers can experiment, fine-tune, or inference models of up to 200B parameters. Plus, NVIDIA ConnectX™ networking can connect two NVIDIA DGX Spark supercomputers to enable inference on models up to 405B parameters. To give developers a familiar experience, NVIDIA DGX Spark mirrors the same software architecture that powers industrial-strength AI factories. Using the NVIDIA DGX OS with Ubuntu Linux and preconfigured with the latest NVIDIA AI software stack, along with developer program access to NVIDIA NIM™ and NVIDIA Blueprints, developers can hit the ground running using common tools such as Pytorch, Jupyter, and Ollama to prototype, fine-tune, and inference on NVIDIA DGX Spark and seamlessly deploy in the data center or cloud. By delivering massive performance and capabilities in a compact package, NVIDIA DGX Spark lets developers, researchers, data scientists, and students continue to push the boundaries of generative AI.

ConnectX-7 Smart NIC

Upgrade easily by Connecting two DGX Spark systems together via ConnectX-7 and LinkX cable to work with even larger models of up to 405 billion parameters utilising the full 256GB of unified memory for better than multi-GPU workstation performance.

Workflow, Models and Tools

DGX Spark comes with all the software AI developers need to get up running AI workloads right away. Users can install their favourite tools, and frameworks or take advantage of the latest NVIDIA workflows, models, microservices and tools such as NVIDIA NIM, AI Blueprints or AI Workbench.

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	1000 AI TOPS
System Memory	128 GB LPDDR5x, unified system memory
Memory Interface	256-bit
Memory Bandwidth	273 GB/s
Storage	1 or 4 TB NVME.M2 with self-encryption
USB	4x USB TypeC
Ethernet	1x RJ-45 connector 10 GbE
NIC	ConnectX-7 Smart NIC
Wi-Fi	WiFi 7
Bluetooth	BT 5.3 w/LE
Audio-output	HDMI multichannel audio output
Power Consumption	TBD
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