NVIDIA will release a 64GB configuration of its DGX Spark personal AI supercomputer this month, cutting the entry price to $4,999 while keeping the same GB10 Grace Blackwell Superchip, DGX OS and full NVIDIA AI software stack as the 128GB model. Six manufacturers — Acer, ASUS, Dell, Gigabyte, HP and MSI — will sell the device starting Friday, Oct. 23.

The new SKU targets developers, researchers and AI enthusiasts who need to run agents and inference locally, without cloud dependency. It supports models of up to 100 billion parameters fully on device, along with the agentic applications built on them.

Same Architecture, Lower Entry Point

DGX Spark combines NVIDIA Grace Blackwell compute, unified memory, ConnectX-7 networking and a CUDA-accelerated AI software stack in a single system. NVIDIA positions the 64GB configuration as an accessible starting point for local AI work — agents, inference, fine-tuning, data science and edge development — at a price $2,000-class below typical dual-unit configurations, with the same architecture throughout the lineup. The system ships ready for agent development from day one: NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron open models and popular runtimes including Ollama, vLLM and PyTorch with CUDA work out of the box. Developers can go from power-on to running models in minutes, according to NVIDIA.

Blender is among the first major creator application providers to support the platform, with a prebuilt, downloadable installer coming soon.

Clustering: 128GB Pooled, Up to 1.7x Performance

Every DGX Spark includes a built-in ConnectX-7 NIC. Two units connect directly with a QSFP cable, pooling memory to 128GB and expanding model support to up to 200 billion parameters, with twice the memory bandwidth. In NVIDIA's Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7x the performance of a single system.

The NVIDIA Sync Cluster Assistant detects connected units, validates device configuration and configures the ConnectX-7 network automatically. Every node runs the same software stack, so nothing needs reconfiguration when scaling from one unit to two.

At the end of the month, NVIDIA will also ship the Sync Model Launcher. It lets developers download and launch Qwen3.8 27B on a single system or a cluster with a few clicks, with NVIDIA Sync configuring the model across connected devices and exposing it on users' laptops. The launcher also sets up OpenCode so developers can start coding in the browser.

Workflow Scenarios

NVIDIA outlines three concrete use cases for the 64GB configuration:

• Around-the-clock agents. Keep a coding or research agent running on DGX Spark for code review, document analysis or multistep tasks. A cluster adds capacity for larger models, longer context windows or multiple concurrent agents.
• Offloaded inference. Run a language- or image-generation model on DGX Spark while a laptop or desktop runs the agent or creative application. DGX Spark handles inference; the PC stays free for other work.
• Scale on demand. When a task outgrows one unit — a larger model, longer context window or concurrent requests — two systems connected over the 200 GbE fabric pool memory to 128GB without reconfiguring the software environment.

Availability and Getting Started

DGX Spark 64GB goes on sale Oct. 23 from Acer (Veriton GN100), ASUS (Ascent GX10), Dell, Gigabyte (AI TOP Atom), HP and MSI (EdgeXpert MS-C931), starting at $4,999.

To deploy: download a supported inference framework — llama.cpp, Ollama, vLLM or LM Studio — then pull the recommended model for the workflow. To scale, connect two units via their ConnectX-7 ports and launch Sync Cluster Assistant, which configures the network and routes workloads automatically.

Agentic AI playbooks are available on build.nvidia.com for NemoClaw, OpenClaw, Hermes Agent and OpenShell. Three more playbooks are coming soon to 64GB devices: serving LLMs with vLLM, running OpenClaw with a local LLM, and connecting multiple DGX Sparks for distributed workloads.

Related Updates

Windows PCs powered by NVIDIA RTX Spark arrive this month from Acer, ASUS, Dell, HP, Lenovo, Microsoft and MSI. Separately, Alibaba's Qwen-Image-2.1, a lightweight open-weight model combining image generation and editing, now runs locally on NVIDIA RTX GPUs, DGX Spark and DGX Station.