NVIDIA is expanding its DGX Spark lineup with a new 64GB unified-memory configuration designed to make local AI development more accessible while giving users an easier way to scale to larger workloads.

The new NVIDIA DGX Spark 64GB will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI starting October 23, with prices starting at $4,999. It uses the same GB10 Grace Blackwell Superchip, DGX OS and NVIDIA AI software stack as the existing 128GB configuration.

The bigger addition, however, is NVIDIA Sync Cluster Assistant. It allows two 64GB DGX Spark systems to be connected and configured as a multi-node cluster, creating a 128GB unified-memory pool for larger AI models and workloads.

NVIDIA DGX Spark gets a 64GB configuration

DGX Spark combines NVIDIA’s Grace Blackwell architecture, unified memory, ConnectX-7 networking and a CUDA-accelerated AI software stack in a compact desktop system designed for local AI development.

The new 64GB configuration retains the GB10 Grace Blackwell Superchip and the same software stack as the 128GB model. NVIDIA says it can run models with up to 100 billion parameters entirely on the device.

The system supports workloads including AI agents, inference, fine-tuning, data science and edge development. NVIDIA also supports popular local AI tools and runtimes including Ollama, vLLM, llama.cpp, LM Studio and PyTorch with CUDA.

That gives developers another entry point into the DGX Spark platform without requiring the 128GB configuration.

Two 64GB DGX Spark systems can create a 128GB cluster

NVIDIA is also making it easier to scale DGX Spark beyond a single system.

Each DGX Spark includes a built-in ConnectX-7 network interface. Two systems can be connected directly using a QSFP cable, after which NVIDIA Sync Cluster Assistant detects the machines, validates their configurations and sets up the ConnectX-7 network.

Two 64GB systems can pool their memory to provide 128GB of unified memory. NVIDIA says this expands support to models with up to 200 billion parameters while providing twice the memory bandwidth.

The company also reports up to 1.7x the performance of a single system in its test using Qwen 3.8 27B. That figure comes from NVIDIA’s own benchmark and should not be treated as a general performance guarantee across workloads.

The software environment remains consistent between the systems, so developers do not have to reconfigure the entire stack when moving from one DGX Spark to two.

NVIDIA Sync is designed to simplify local AI scaling

The new Sync Cluster Assistant is intended to remove much of the manual infrastructure setup normally associated with multi-node AI systems.

Once two DGX Spark systems are connected, Sync handles the network configuration and workload routing. NVIDIA says this allows developers to focus on their models and applications rather than configuring the underlying cluster.

NVIDIA is also preparing Sync Model Launcher, which will let developers download and launch Qwen3.8 27B on either a single DGX Spark or a cluster. Sync will configure the model to run across connected systems and make it accessible from a user’s laptop.

The launcher will also set up OpenCode to use the model for browser-based coding workflows. NVIDIA says the feature is coming later in October.

What can you do with DGX Spark?

NVIDIA is positioning the 64GB DGX Spark as a local platform for several types of AI workloads.

Developers can keep coding or research agents running locally, use DGX Spark to handle AI inference while their everyday PC handles other applications, or connect two systems when a model or workload exceeds the capacity of one machine.

The local approach can also reduce the need to send every AI workload to a cloud service, particularly when developers want to work with their own data on local hardware.

NVIDIA says DGX Spark supports its Agent Toolkit, CUDA-X AI libraries and Nemotron open models out of the box. Blender is also preparing a downloadable installer for the platform.

NVIDIA DGX Spark 64GB price and availability

The new DGX Spark 64GB configuration will be available starting Friday, October 23, through Acer, ASUS, Dell, Gigabyte, HP and MSI.

Pricing starts at $4,999.

NVIDIA is keeping the 128GB DGX Spark configuration alongside the new model. The 64GB version gives developers a lower-memory starting point, while two systems can later be connected to create a 128GB unified-memory cluster.

For developers working with increasingly capable local AI models, that makes the new DGX Spark configuration less about simply cutting memory in half and more about giving the platform a more flexible upgrade path.