NVIDIA Introduces 64GB DGX Spark and Tool for Two-Node Systems

NVIDIA DGX Spark 64GB is a new system configuration for running AI workloads locally, announced by the company on October 2. NVIDIA also introduced Sync Cluster Assistant, a tool for automatically configuring a pair of interconnected devices. Sales of 64GB versions from Acer, ASUS, Dell, Gigabyte, HP, and MSI are expected to begin on October 23, according to the company, with prices starting at $4,999.

The new model expands the existing DGX Spark lineup, which NVIDIA continues to list with a configuration featuring 128 GB of shared memory. That configuration is officially intended for models with up to 200 billion parameters.

NVIDIA DGX Spark 64GB Uses GB10 and DGX OS

The 64GB configuration is expected to use the GB10 Grace Blackwell Superchip, the DGX OS operating system, and the NVIDIA AI software suite. These are core elements of the DGX Spark platform.

Shared memory is particularly important in these devices for working with larger language models and other AI models directly at a developer’s workstation. However, the required capacity does not depend solely on the model’s parameter count. Quantization, context size, the specific software, and the deployment method also matter.

NVIDIA therefore does not provide a universal usability threshold for all models with the new variant. The 64GB system’s real-world capabilities can only be assessed through independent testing in specific applications.

Sync Cluster Assistant Is Designed to Connect Two Systems

The announcement includes NVIDIA Sync Cluster Assistant. According to NVIDIA, it automates the configuration of two DGX Spark systems connected through a ConnectX-7 network interface. A pair of 64GB devices is intended to pool 128 GB of memory in total.

The tool is intended to automate the configuration of a small local cluster. NVIDIA also reports up to 1.7x acceleration for a pair of 64GB devices in an internal test using the Qwen 3.8 27B model. However, this result comes from the manufacturer and has not been independently verified.

NVIDIA has announced the release of Sync Model Launcher for the end of October. The company did not provide a detailed availability date or broader technical details in the announcement.

Another Configuration for Local AI Hardware

The new 64GB version adds another configuration alongside the 128GB DGX Spark. For developers who do not need the full capacity of the original configuration, it could be an option for local experiments, inference, and the development of AI applications and agents.

According to NVIDIA, two devices can pool 128 GB of memory. The practical result, however, will depend on the supported software and the nature of the task being processed.

What to Watch When It Launches

• whether partner manufacturers actually begin sales in multiple regions on October 23 and at the announced starting price,
• independent performance measurements for both the 64GB configuration and a pair of devices, especially for inference and long-context workloads,
• the availability and practical operation of NVIDIA Sync Model Launcher, announced for the end of October.

The sales date and price starting at $4,999 are still plans from NVIDIA and its partner manufacturers, not a confirmed indication of store availability.

Verified and updated: 10/02/2026 15:20