.png){kind=icon}
NVIDIA DGX™ Spark is a new type of computer designed specifically for building and running AI, delivering up to 1,000 AI TOPS of performance through high energy efficiency and compact form factor.
At the same time, with pre-installed NVIDIA AI software stack and 128GB of unified addressing system memory, developers can locally prototype, fine-tune, and inference large AI models with up to 200B parameters, and seamlessly deploy
them to data centers or the cloud.
1. Based on NVIDIA Grace Blackwell Architecture
At the core of DGX Spark is the new GB10 Grace Blackwell superchip, which is based on the Grace Blackwell architecture and optimized for desktop form factor. GB10 features the powerful Blackwell GPU, supporting 5th generation Tensor Core and FP4, delivering up to 1000 TOPS of AI computing performance.
GB10 also includes a high-performance Grace 20-core Arm CPU, which powerfully assists data preprocessing and orchestration, thereby accelerating model tuning and real-time inference. The GB10 superchip uses NVLink™-C2C interconnect technology, providing a unified memory model combining CPU+GPU with 5x the bandwidth of 5th generation PCIe.
2. Working with Next-Generation Large Parameter Generative AI Models
With 128 GB of unified addressing system memory and support for FP4 data format, DGX Spark can support up to 200B parameter AI models, enabling AI developers to prototype, fine-tune, and inference next-generation AI inference models (such as DeepSeek R1 distilled versions with up to 70B parameters) on desktop.
With built-in NVIDIA ConnectX™ networking technology, two DGX Spark systems can be connected to handle larger models such as Llama 3.1 405B.
3. Local Development, Large-Scale Deployment Anytime, Anywhere
DGX Spark provides developers with a powerful and cost-effective experimental playground for prototyping models and AI applications, thereby freeing up valuable computing resources in cluster environments that are better suited for training and deploying production models.
NVIDIA's full-stack AI platform supports DGX Spark users to seamlessly migrate their models from 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 their workflows.
Starting today, pre-order NVIDIA DGX Spark to secure priority delivery schedule.
As an NVIDIA Elite Partner, XindaMeng Technology provides comprehensive services from product consultation, solution selection to deployment implementation, helping you efficiently build enterprise-level AI infrastructure.
First batch delivery slots are now open for pre-order, welcome to inquire~
The copyright of images or videos (complete or partial) related to NVIDIA products belongs to NVIDIA Corporation.
We will contact you within 24 hours.