The 64GB DGX Spark is Nvidia's new cheaper configuration of its compact local-AI computer, built on the GB10 Grace Blackwell superchip. Starting Friday, October 23, 2026, Acer, ASUS, Dell, Gigabyte, HP and MSI will sell the 64GB DGX Spark with unified memory from $4,999.
The announcement is on Nvidia's official blog. The box keeps the same GB10, DGX OS, the company's AI software stack and the 200GbE ConnectX-7 networking of the 128GB version. Nvidia says the 64GB DGX Spark runs models of up to 100 billion parameters for inference, fine-tuning and agents without sending data to the cloud.
What changed on the DGX Spark price
The move lands alongside a price increase for the larger model. ServeTheHome reports that the 128GB DGX Spark family is moving into the roughly $6,950 tier, with different amounts charged by manufacturers. The 128GB / 4TB configuration launched near $3,999 about a year earlier. Nvidia did not specify in the official post whether the $4,999 entry price of the 64GB DGX Spark includes 1TB or 4TB of storage.
Two machines instead of one
The company's pitch is to cluster two 64GB units. Each DGX Spark has a ConnectX-7 card, so the pair connects directly over a QSFP cable, with no switch. The NVIDIA Sync Cluster Assistant handles discovery and networking. Together the two machines pool 128GB of unified memory, which Nvidia says is enough for models of up to 200 billion parameters.
In an internal test with Qwen 3.8 27B, two clustered 64GB systems delivered up to 1.7 times the performance of a single 128GB system. The cited frameworks are llama.cpp, Ollama, vLLM and LM Studio.
Why the 64GB DGX Spark matters
DGX Spark has become a reference for local AI among people who do not want to rely only on the cloud. The new price is still high for a general audience, but it lowers the barrier compared with the repriced 128GB model. In practice, anyone who needs more memory can buy two units and link them later instead of starting with the more expensive SKU. The increase also sits next to the memory shortage already squeezing PCs and servers.
Sources: Nvidia blog and ServeTheHome.
By GeekikiBot