Surprisingly Useful AI Article Enhancements
Nvidia is expanding its DGX Spark lineup with a new 64GB unified memory configuration, aimed at developers, researchers and AI enthusiasts looking to run AI models and agents locally without relying on cloud infrastructure.
The 64GB model retains the same GB10 Grace Blackwell Superchip, DGX OS and Nvidia AI software stack found in the existing 128GB version of the compact personal AI system.
It will be available from October 23 through manufacturer partners Acer, Asus, Dell, Gigabyte, HP and MSI, with pricing starting at US$4,999.
A lower entry point
The new configuration is designed to offer a more accessible price point for the DGX Spark platform while preserving its core capabilities.
Nvidia states the 64GB model supports models with up to 100 billion parameters running entirely on device, along with the agentic applications built on top of them.
The system ships with Nvidia Agent Toolkit, CUDA-X AI libraries, Nemotron open models and support for inference frameworks including Ollama, vLLM and PyTorch with CUDA out of the box.
Blender is among the first creator applications to support the platform, with a prebuilt installer expected to follow.
Clustering with Nvidia Sync
Alongside the new configuration, Nvidia is introducing the Sync Cluster Assistant, a tool that allows two DGX Spark units to be connected directly via their built-in Nvidia ConnectX-7 network ports using a QSFP cable.
The tool detects connected units, validates device configuration and sets up the ConnectX-7 network automatically, allowing developers to pool the memory of two systems to a combined 128GB.
Nvidia claims two clustered 64GB systems can support models with up to 200 billion parameters while delivering twice the memory bandwidth. In Nvidia’s own Qwen 3.8 27B testing, the clustered configuration delivered up to 1.7x the performance compared with a single unit.
Because each node runs the same Nvidia software stack, no reconfiguration is required when scaling from one unit to two.
Model launcher on the way
Nvidia has also flagged the forthcoming Sync Model Launcher, due later in October, which aims to simplify running local AI workloads to a few button clicks.
The launcher will allow developers to download and run the Qwen3.8 27B model on a single DGX Spark or across a cluster, with Nvidia Sync handling the distribution of the model across connected devices and making it accessible from a user’s laptop.
The launcher will also configure OpenCode, enabling developers to begin coding through their browser.
Intended use cases
Nvidia has outlined several target workflows for the 64GB DGX Spark.
Related: Best Business Laptops for work & school
Related: Best Gaming Laptops
Related: Best Portable Laptop
These include running a coding or research agent continuously on device, ready to review code, analyse documents or carry out multistep tasks. A clustered setup provides capacity for larger models, longer context windows or multiple agents operating concurrently.
The system can also serve as a dedicated inference backend, handling language or image generation workloads while freeing up a developer’s laptop or desktop for other tasks.
When a workload outgrows a single unit – whether due to a larger model, extended context or concurrent agent requests – two 64GB systems connected via Nvidia Sync Cluster Assistant can pool their resources to 128GB without requiring changes to the software environment.
Getting started
Developers looking to begin working with DGX Spark can download a supported inference framework such as llama.cpp, Ollama, vLLM or LM Studio, then download a recommended model for their workflow.
To scale to a two-unit cluster, the process involves connecting the systems via their ConnectX-7 ports and launching the Sync Cluster Assistant, which configures the network and routes workloads automatically.
Nvidia has published agentic AI playbooks for DGX Spark on build.nvidia.com, covering projects including NemoClaw, OpenClaw, Hermes Agent and OpenShell. Playbooks for the 64GB configuration – including guides on serving LLMs with vLLM, running OpenClaw with a local LLM and connecting multiple DGX Sparks for distributed workloads – are expected to follow.
Separately, new Windows PCs powered by Nvidia RTX Spark are also due this month from Acer, Asus, Dell, HP, Lenovo, Microsoft and MSI.
Last Updated on October 3, 2026 by Nick Ross
