NVIDIA has made a critical move for developers and researchers who want to run AI models directly on their own hardware without cloud dependence. The company officially announced the new member of its series, positioned as a personal AI supercomputer, the NVIDIA DGX Spark 64 GB model. Based on the Grace Blackwell architecture, the compact system is available to users through leading manufacturer partners such as Acer, ASUS, Dell, Gigabyte, HP and MSI, with a starting price of $4,999.

Grace Blackwell Power in a Compact Case: What Does the DGX Spark 64 GB Offer?

The new DGX Spark configuration retains the GB10 Grace Blackwell Superchip processor, DGX OS operating system, and NVIDIA’s comprehensive software ecosystem found in the top 128 GB model of the series. Thanks to its unified memory architecture, the device can run open-source AI models and autonomous spyware with up to 100 billion parameters entirely locally, without risking data confidentiality.

The system is ready for developers’ workflows right out of the box. It directly supports NVIDIA Agent Toolkit, CUDA-X libraries, and Nemotron models, as well as popular development environments such as Ollama, vLLM, LM Studio, and PyTorch. Additionally, it was shared that a platform-specific pre-compiled installer for the well-known 3D modeling and animation software Blender will be released very soon.

Two Units, One System: Easy Clustering with NVIDIA Sync Cluster Assistant

One of the most striking aspects of the DGX Spark 64 GB is its scalability. Thanks to the integrated NVIDIA ConnectX-7 network adapter and QSFP connector on the device, users can connect two systems directly to each other. At this point, NVIDIA Sync Cluster Assistant takes over, automatically handling network configuration and load balancing.

• Memory Capacity: When two units are connected, the combined memory pool reaches 128 GB.
• Model Support: A wide range of license models with up to 200 billion parameters can run on a single cluster.
• Performance Increase: In Qwen 3.8 27B tests, two clustered systems offer up to 1.7 times performance increase and twice the memory bandwidth compared to a single device.

Running AI Models is Reduced to a Single Button

NVIDIA focuses on simplifying the software experience as well as the hardware. NVIDIA Sync Model Launcher, which will be available for use at the end of the month, reduces the launching of models on a local device or cluster to just a few clicks. Users can launch the Qwen3.8 27B model on the DGX Spark with a single touch from their laptops and code directly through the browser thanks to OpenCode integration.

DGX Spark 64 GB, which relieves personal computers from high processing loads, eliminates complex cloud rental costs, and prevents company data leaks, will be available on the global market from October 23rd.

NVIDIA announced the DGX Spark model with 64 GB of memory, which facilitates local AI development. Details of the $4,999 system are in our news.