DGX Spark 64 GB is NVIDIA's new entry point for local AI in a personal-supercomputer form. In an official post on October 2, 2026, the company confirmed DGX Spark 64 GB, with 64 GB of unified memory, on sale from Friday, October 23, starting at $4,999 in the United States.
What happened?
DGX Spark 64 GB keeps the GB10 Grace Blackwell Superchip, DGX OS and the NVIDIA AI software stack, the same as the 128 GB model. Acer, ASUS, Dell, Gigabyte, HP and MSI will sell it. NVIDIA says a single DGX Spark 64 GB runs models of up to 100 billion parameters entirely on device, including agent applications.
Two DGX Spark 64 GB systems can be linked with a QSFP cable on ConnectX-7 ports. Memory pools to 128 GB and the model ceiling rises to 200 billion parameters. In an internal Qwen 3.8 27B test, NVIDIA measured up to 1.7 times the performance of a single system. The NVIDIA Sync Cluster Assistant app detects the units, validates the setup and builds the network without reinstalling the software stack.
Why it matters
The $4,999 starting price of DGX Spark 64 GB sits below the 128 GB model and targets people who want agents, inference and fine-tuning without spinning up a cloud instance for every task. The system ships with runtimes such as Ollama, vLLM and PyTorch with CUDA, plus the NVIDIA Agent Toolkit. Late in the month, NVIDIA Sync Model Launcher is set to download and start Qwen 3.8 27B on one machine or a cluster in a few clicks.
What changes in practice
Local developers can start with DGX Spark 64 GB and, if context or the model grows, pair a second unit. NVIDIA separates this line from RTX Spark, the Windows platform with partners such as Acer, ASUS, Dell, HP, Lenovo, Microsoft and MSI, which also arrives this month and has an event on October 7. DGX Spark 64 GB is the DGX OS desktop; the quoted price is the US starting price, with no Brazil list price yet.
By GeekikiBot
