NVIDIA's 64GB DGX Spark brings local AI down to $4,999
A 64GB DGX Spark goes on sale on 23 October from $4,999, and two of them can be clustered to pool 128GB of memory for models of up to 200 billion parameters.
NVIDIA has added a cheaper way into its desktop AI line. The company says a 64GB configuration of DGX Spark will go on sale on Friday 23 October through Acer, ASUS, Dell, Gigabyte, HP and MSI, starting at $4,999.
Half the memory, the same silicon
The new model is not a cut-down platform. It keeps the GB10 Grace Blackwell Superchip, DGX OS and the full NVIDIA AI software stack, so it is the same machine as the 128GB DGX Spark with half the unified memory. NVIDIA positions it for developers, researchers and enthusiasts who would rather run agents and inference on hardware they own than rent a cloud instance for every task.
Two units, 128GB and 200 billion parameters
Every DGX Spark ships with a ConnectX-7 network interface, and two 64GB units can be linked directly with a QSFP cable to pool 128GB of memory. NVIDIA says that raises model support to about 200 billion parameters, while a single 64GB system handles models of up to 100 billion parameters. In the company's own Qwen 3.8 27B test, two clustered 64GB systems returned up to 1.7x the performance of one.
The clustering is meant to be handled by software rather than a manual setup. NVIDIA Sync's cluster assistant detects connected units, validates their configuration and configures the ConnectX-7 network, and a Model Launcher due at the end of October will download and run Qwen 3.8 27B on a single system or a cluster.
Blender and the local software stack
NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron open models, Ollama, vLLM and PyTorch with CUDA are all supported out of the box. Blender is named as one of the first major creator applications to support the platform, with a prebuilt, downloadable installer due soon.
The company also says new Windows PCs powered by NVIDIA RTX Spark are coming this month from Acer, ASUS, Dell, HP, Lenovo, Microsoft and MSI.
Our opinion
The 64GB Spark is the more interesting of the two machines. Keeping the silicon and the software stack while halving the memory removes the price of entry that kept small teams away, and the two-unit cluster path quietly turns a desk-side box into a 200-billion-parameter system. NVIDIA is selling the upgrade path before it sells the upgrade, which is a harder trick than it sounds.