Unified Memory
NVIDIA DGX™ Spark delivers up to 1 petaFLOP of AI compute and 128GB of LPDDR5 unified memory for extreme low-latency data access, speeding up inferencing and fine-tuning.
Wide Data Format
Inheriting the Blackwell GPU’s transformer engine, DGX Spark supports FP4 data format. Handle models upwards of 200B parameters for extreme AI development and fine-tuning.
Deploy Anywhere
DGX Spark fits in your bag, travelling wherever your development needs bring you. Develop on the NVIDIA AI platform and move work to and from the data center with ease.
The Personal AI Workstation
NVIDIA DGX Spark is everything NVIDIA envisioned for the data center, packed into a small 6”x 6” box. Made for developers, researchers, data scientists, and enthusiasts who need an AI solution that can travel with them while being abstracted from the data center.
Prototype, fine-tune, run local LLM inference and Agentic AI, and more with the DGX Spark, offering a large 128GB of unified CPU and GPU memory, optimized for enterprise workloads. Port your work over to the data center or cloud with confidence that it will scale.
DGX Spark is available as a Founder Edition from NVIDIA as well as from board partners like MSI and Asus.
NVIDIA AI Enterprise Now for NVIDIA DGX Spark
NVIDIA’s cloud-native suite of software tools, libraries, and frameworks assists users in accelerating production-grade AI development with NVIDIA DGX Spark from the jump. Enterprise-grade security, optimized performance, and enterprise-level support are included to streamline prototype/production for next-gen agentic AI.
How NVIDIA DGX Spark Accelerates Your Workload
AI Prototyping
Quickly test and validate new AI concepts on the powerful DGX Spark, designed for experimentation. Harness speed and flexibility to move from idea to working prototype without delays.
AI Model Fine-tuning
Adapt pre-trained models to your data confidently and fine-tune with NVIDIA DGX Spark so your AI performs better for your specific use case, all with zero interruption to your main computing infrastructure.
Local LLM Inferencing
Run large language models directly on your portable workstation for faster responses and full data control. DGX Spark enables high-performance inferencing without relying on external cloud services.
NVIDIA DGX Spark Specifications
| NVIDIA DGX Spark | |||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NVIDIA Grace CPU | 20 ARM Cores (10 Cortex®-X925 + 10 Cortex-A725) | ||||||||||||||||||||||||||||
| CPU Memory | 128GB Unified LPDDR5X at 273 GB/s | ||||||||||||||||||||||||||||
| NVIDIA GPU | NVIDIA Blackwell | ||||||||||||||||||||||||||||
| Networking | 1x 10Gb Ethernet | ||||||||||||||||||||||||||||
| Networking NIC | NVIDIA ConnectX-7 NIC (at 200Gb Ethernet | ||||||||||||||||||||||||||||
| Net. Bandwidth | Up to 200Gb/s | ||||||||||||||||||||||||||||
| Storage | Up to 4TB NVMe SSD | ||||||||||||||||||||||||||||
| Connectivity | WiFi 7 & Bluetooth 5.3 | ||||||||||||||||||||||||||||
| Output | 1x HDMI 2.1a | ||||||||||||||||||||||||||||