NVIDIA is expanding its DGX Spark personal AI supercomputer line with a 64GB unified-memory configuration that gives developers more headroom to run larger models and more complex agentic workflows entirely on device.
Starting October 23, the new SKU will be available from Acer, Asus, Dell, Gigabyte, HP, and MSI with the same GB10 Grace Blackwell Superchip, DGX OS and full NVIDIA AI software stack as the 128GB model but at a more accessible price point.
The capacity upgrade will support running up to 100-billion-parameter models locally, compared with smaller models on lower-memory configurations, while retaining the platform’s ability to handle agents, inference, fine-tuning, and data science workloads without cloud dependency.
Developers can run capable local agents on device — privately, without cloud dependency, and keep coding or research agents running around the clock for tasks such as code review, document analysis and multi-step agent workflows.
For workloads that outgrow a single box, two 64GB units can be clustered via the built-in NVIDIA ConnectX-7 networking and NVIDIA Sync Cluster Assistant to pool memory to 128GB and expand model support to up to 200 billion parameters.
In NVIDIA’s own tests using the Qwen 3.8 27B model, two clustered 64 GB systems delivered up to 1.7x performance compared with a single system, with room to keep scaling as workloads demand.
Unlike earlier setups that required manual network and software tuning, the Sync app configures this multi-node cluster seamlessly, detecting connected units, validating device configuration and setting up the ConnectX-7 network so developers can focus on their work rather than the infrastructure.
As every node runs the same NVIDIA software stack, no reconfiguration is needed when scaling from one unit to two, enabling teams to start with a single 64GB system and add a second as models, context windows or concurrent agent requests grow.
The DGX Spark 64GB configuration will be available from October 23 starting at US$4,999 through the OEM partners mentioned earlier.
