Nvidia says a 64GB unified memory version of DGX Spark will arrive on October 23, starting at $4,999. Acer, ASUS, Dell, Gigabyte, HP and MSI will sell the configuration. It gives builders a smaller-memory option alongside the existing 128GB model for running AI models and agents locally.
In its October 2 announcement, Nvidia says the partner-only configuration retains the GB10 Grace Blackwell Superchip, DGX OS and the full Nvidia AI software stack. The company claims support for models with up to 100 billion parameters on one system.
What the hardware and software include
DGX Spark has a built-in ConnectX-7 network interface. Two 64GB systems can connect directly through a QSFP cable over a 200 GbE network. Nvidia says that gives the cluster 128GB of pooled memory and support for models with up to 200 billion parameters.
Nvidia Sync Cluster Assistant detects connected systems, validates their configuration and sets up the ConnectX-7 network. In Nvidia’s Qwen3.8 27B test, two 64GB systems delivered up to 1.7 times the performance of one system. That result applies to the company’s reported test, so buyers should check their own model and workload before budgeting for the same gain.
The software stack supports Nvidia Agent Toolkit, Nemotron models and CUDA-enabled runtimes including Ollama, vLLM and PyTorch. Nvidia also lists llama.cpp and LM Studio as supported inference options.
At the end of October, Nvidia plans to release Sync Model Launcher. It will download and launch Qwen3.8 27B on one Spark or a cluster, make the model accessible from a laptop and configure OpenCode to use it for coding in a browser.
