Since its launch, NVIDIA has continuously enhanced the performance of the Grace Blackwell–based DGX Spark through ongoing software optimizations and close collaboration with software partners and the open‑source community. These efforts have delivered significant improvements in inference, training, and creative workflows.

At CES 2026, a new DGX Spark software release—combined with updated models and open‑source libraries—brings notable performance gains for DGX Spark and GB10‑based OEM systems.

Scale Large Models Locally with Unified Memory and NVFP4
Designed for localized large‑model applications, DGX Spark features 128 GB of unified memory in a compact desktop form factor. Connecting two DGX Spark systems provides a combined 256 GB of memory, enabling developers to run even larger models locally. With ConnectX‑7 networking delivering 200 Gbps of bandwidth, the systems support high‑speed, low‑latency multi‑node workloads. Support for the NVIDIA NVFP4 data format significantly reduces memory footprint for next‑generation models while boosting throughput.

A Powerful Desktop Platform for Creators
DGX Spark is not only an exceptional platform for AI developers—creators can also fully leverage its desktop‑grade capabilities. By offloading AI workloads to DGX Spark, creators keep their laptops or PCs responsive during content‑generation tasks. Leading diffusion models such as Alibaba’s Qwen‑Image are already using NVFP4 to reduce memory usage while achieving higher performance.

DGX Spark Now Part of the NVIDIA Certified Systems Program
The NVIDIA Certified Systems program validates system performance across a broad range of accelerated graphics, computing, and AI workloads. Certified systems provide a trusted foundation for AI development, desktop inference, data science, design, and content‑creation workloads, while also complementing data‑center and cloud resources.

Get Started Quickly
Simple, reliable, and clearly‑documented troubleshooting guides—verified on the latest DGX OS—help users reduce setup time and spend more time building.

Access DGX Spark Anywhere with NVIDIA Brev
With NVIDIA Brev, users can securely access DGX Spark from anywhere. Brev makes it easy for developers to create AI cloud instances; its Launchables feature enables one‑click configuration of complete AI environments. At CES, updates to Brev demonstrated the ability to register local compute resources such as DGX Spark. Once registered in Brev, users can access DGX Spark remotely anytime and securely share that access with their team.

The above is a summary. To read the full article, please scan the QR code below.

Scan the poster QR code to watch the replay of CEO Jensen Huang’s CES 2026 NVIDIA Live keynote from Las Vegas.
GTC 2026 will be held from March 16 to 19, 2026 in San Jose, California, USA, with both online and in-person participation available. Scan the QR code below to register for the online conference or purchase tickets for the in-person event now.