NVIDIA DGX Spark 64GB Brings Local AI Agents and 100B-Parameter Models to Developers
NVIDIA is expanding local AI development with a new 64GB configuration of NVIDIA DGX Spark. Designed for developers, researchers and AI enthusiasts, the compact personal AI supercomputer can run capable AI agents privately on-device without relying on the cloud.
Available this month through manufacturer partners including Acer, ASUS, Dell, Gigabyte, HP and MSI, the new configuration includes DGX OS and the NVIDIA AI software stack. Developers can also connect two DGX Spark systems to increase available memory and performance as their workloads grow.
NVIDIA DGX Spark 64GB: A Personal AI Supercomputer
DGX Spark combines NVIDIA Grace Blackwell computing, unified memory, NVIDIA ConnectX-7 Networking and the NVIDIA CUDA-accelerated AI software stack in one system.
The platform is designed for local AI agents, inference, fine-tuning, data science and edge development. By working with models and private data locally, developers can experiment without relying on cloud instances for every task.
The 64GB configuration uses the same GB10 Grace Blackwell superchip, DGX OS and full NVIDIA AI software stack as the 128GB model. It supports on-device AI applications and models with up to 100 billion parameters, helping keep local AI development accessible at a lower starting price.
Two 64GB systems can be clustered when a project requires additional memory or compute. In NVIDIA testing with Qwen 3.8 27B, two clustered 64GB systems delivered up to 1.7 times the performance of a single system.
DGX Spark is ready for agent development from day one. Supported software includes NVIDIA Agent Toolkit, CUDA-X AI libraries, the Nemotron open model, Ollama, vLLM and PyTorch with CUDA. Developers can begin running supported models within minutes of powering on the system.
Blender is among the first major creator application providers to support the platform. Pre-built downloadable installers are coming soon.
Scale Local AI with NVIDIA Sync Cluster Assistant
Developers can begin with one DGX Spark and expand to a two-node cluster as their workloads increase. Each system includes an NVIDIA ConnectX-7 network interface, and two units can be connected directly with a QSFP cable to pool memory.
A two-system configuration provides 128GB of pooled memory, support for models with up to 200 billion parameters, twice the memory bandwidth and up to 1.7 times the performance described in NVIDIA’s testing.
NVIDIA Sync simplifies the setup process. Its Cluster Assistant discovers connected systems, validates device configurations and configures the ConnectX-7 network. Because both nodes use the same NVIDIA software stack, developers do not need to reconfigure their environment when expanding from one system to two.
The NVIDIA Sync Model Launcher, scheduled for release later this month, will make it possible to run local AI models with a few clicks. Developers will be able to download and launch Qwen3.8 27B on a single DGX Spark system or a cluster.
NVIDIA Sync configures models to run across connected devices and makes them accessible from users’ laptops. The launcher also sets up OpenCode, allowing developers to begin coding in the browser.
What Can Developers Do with DGX Spark?
The DGX Spark 64GB configuration supports hands-on local AI development from day one. Developers and hobbyists can run models that fit in memory on a single system or connect multiple DGX Spark systems through NVIDIA Sync Cluster Assistant for larger workloads.
- Run AI agents around the clock: Use DGX Spark to run coding or research agents that review code, analyze documentation and complete multi-step tasks. Clusters add capacity for larger models, longer context windows or multiple agents running simultaneously.
- Power AI applications from a personal computer: Run language or image-generation models on DGX Spark while using agents or creative applications on a laptop or desktop. DGX Spark handles model inference and frees the primary computer for other work.
- Scale as workloads grow: Connect two DGX Spark 64GB systems through NVIDIA Sync Cluster Assistant to pool 128GB of memory for larger models, longer context windows or concurrent agent requests. The same workflow can be extended across two systems without reconfiguring the software environment.
How to Get Started with DGX Spark
The DGX Spark 64GB configuration is available through Acer, ASUS, Dell, Gigabyte, HP and MSI beginning Friday, Oct. 23, starting at $4,999.
- Download a supported inference framework, such as llama.cpp, Ollama, vLLM or LM Studio.
- Download the local model recommended for your workflow.
- To expand to two systems, connect them through the NVIDIA ConnectX-7 port and launch NVIDIA Sync Cluster Assistant. The software configures the network and routes workloads automatically.
For the DGX Spark Agent AI Playbook, visit the following NVIDIA resources:
More NVIDIA Local AI Updates
Explore the latest DGX Spark playbooks at build.nvidia.com/spark. The following playbooks will be available soon for 64GB devices:
- Delivering LLMs with vLLM
- Running OpenClaw with a local LLM
- Connecting multiple DGX Spark systems for distributed workloads
NVIDIA RTX Spark Windows PCs are also being released this month by Acer, ASUS, Dell, HP, Lenovo, Microsoft and MSI. Sign up for the RTX Spark Newsletter for future updates.
Alibaba’s Qwen-Image-2.1 unifies image generation and editing in a lightweight open-weight model. It runs locally on NVIDIA RTX GPUs, DGX Spark and DGX Station, giving creators more ways to generate and edit images on their own hardware.
Follow NVIDIA RTX Spark on X, Instagram, TikTok and Facebook. Subscribe to the NVIDIA Local AI Newsletter for updates. Follow NVIDIA Workstation on LinkedIn and X.
See the NVIDIA terms of service for information about software and product information.
Source: blogs.nvidia.com