NVIDIA Adds 64GB DGX Spark at $4,999, Cluster Support for Two Units

What the 64GB DGX Spark offers

NVIDIA says the new 64GB configuration is available exclusively from its manufacturer partners and keeps the core of the platform intact. It retains the GB10 Grace Blackwell Superchip, DGX OS and the full NVIDIA AI software stack, the same as the 128GB model. According to the company, it supports models of up to 100 billion parameters and the agentic applications built on them, fully on device.

DGX Spark combines NVIDIA Grace Blackwell compute, unified memory, NVIDIA ConnectX-7 networking and a CUDA-accelerated AI software stack in one system. NVIDIA describes it as a complete local AI platform for agents, inference, fine-tuning, data science and edge development. The pitch is that developers can experiment with models and their own data without turning to a cloud instance for every task.

The system ships ready for agent development. NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron open models, and runtimes such as Ollama, vLLM and PyTorch with CUDA are supported out of the box. NVIDIA also says Blender is among the first major creator application providers to support the platform, with a prebuilt, downloadable installer coming soon.

Clustering two units with NVIDIA Sync

Every DGX Spark includes a built-in NVIDIA ConnectX-7 NIC. Two units can be linked directly with a QSFP cable, which pools their memory to 128GB and expands model support to up to 200 billion parameters. NVIDIA says this delivers twice the memory bandwidth and up to 1.7x the performance.

The 1.7x figure also appears in a specific test. In NVIDIA's Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7x performance compared with a single system, according to the company.

The NVIDIA Sync app handles the setup. Its Cluster Assistant feature detects connected units, validates device configuration and configures the ConnectX-7 network. Because every node runs the same NVIDIA software stack, nothing needs to be reconfigured when moving from one unit to two, NVIDIA says.

At the end of the month, NVIDIA Sync Model Launcher is due to arrive. It is meant to let developers download and launch Qwen3.8 27B on a single system or a cluster with a few clicks, make it accessible from their laptops, and set up OpenCode to use the model so they can start coding in a browser.

Suggested workflows

NVIDIA lists three example uses:

• Run an agent around the clock: keep a coding or research agent running to review code, analyze documents or carry out multistep tasks. A cluster adds capacity for larger models, longer context windows or several agents at once.
• Power AI apps on an everyday PC: run a language or image-generation model on DGX Spark while using an agent or creative application on a laptop or desktop, leaving the PC free for other work.
• Scale when the work grows: when a task outgrows one unit, two 64GB systems connected over the 200 GbE fabric pool their memory to 128GB without changing the software environment.

Price, availability and getting started

DGX Spark 64GB goes on sale from Acer, ASUS, Dell, Gigabyte, HP and MSI on Friday, Oct. 23, starting at $4,999. NVIDIA recommends downloading a supported inference framework (llama.cpp, Ollama, vLLM or LM Studio) and the recommended local model for the intended workflow. To scale to two units, users connect them through their ConnectX-7 ports and launch Sync Cluster Assistant, which configures the network and routes workloads automatically.

Playbooks for 64GB devices are coming soon on build.nvidia.com/spark: Serve LLMs With vLLM, Run OpenClaw With a Local LLM, and Connect Multiple DGX Sparks for Distributed Workloads.

For developers, the 64GB model offers a lower-memory entry point to the same software environment as the 128GB version, with a documented route to more memory by adding a second unit. How the price compares with other configurations is not stated in NVIDIA's announcement.