NVIDIA adds a 64GB DGX Spark configuration

NVIDIA says the new DGX Spark configuration has 64GB of unified memory and starts at $4,999.[1] It will be available exclusively through manufacturer partners, with systems available starting Friday, October 23, from Acer, ASUS, Dell, Gigabyte, HP and MSI.[1] NVIDIA says the 64GB version retains the GB10 Grace Blackwell Superchip, DGX OS and full NVIDIA AI software stack found in the 128GB model.[1] For teams weighing a local system, the announced price and partner route are concrete purchasing details; the evidence does not provide a price for the 128GB model.

NVIDIA positions DGX Spark for agents, inference, fine-tuning, data science and edge development.[1] The company says the system combines Grace Blackwell compute, unified memory, ConnectX-7 networking and a CUDA-accelerated AI software stack.[1] Supported software includes NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron open models, Ollama, vLLM and PyTorch with CUDA out of the box.[1] NVIDIA says the system operates privately without cloud dependency and supports models with up to 100 billion parameters.[1] These are NVIDIA’s stated capabilities, not independent performance results.

NVIDIA says two 64GB DGX Spark systems can be clustered using Sync Cluster Assistant without additional setup.[1] When connected, the systems pool memory to 128GB and support models with up to 200 billion parameters; NVIDIA also says the pair delivers twice the memory bandwidth of one system.[1] The systems can connect directly with a QSFP cable.[1] NVIDIA says Sync Cluster Assistant detects connected units, validates their configuration and configures the ConnectX-7 network.[1] The clustered systems run the same software stack, so moving from one to two does not require reconfiguration.[1]

In NVIDIA’s Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7 times the performance of a single system.[1] That figure is tied to the named test and does not establish results for other models or workloads. NVIDIA says Sync Model Launcher is coming at the end of October and will launch models on one DGX Spark or a cluster.[2] The company says the launcher will set up OpenCode for browser-based coding.[1] NVIDIA also identifies Blender as an early major creator application provider for DGX Spark and says a downloadable installer is coming soon.[1]