NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI

NVIDIA is adding a 64GB unified memory configuration to its DGX Spark personal AI computer, the company said. The new SKU will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI on Friday, Oct. 23, starting at $4,999, and is available exclusively from manufacturer partners.

The 64GB model retains the GB10 Grace Blackwell Superchip, DGX OS and the full NVIDIA AI software stack, the same as the 128GB version. It supports up to 100-billion-parameter models running entirely on device.

Two 64GB units can be connected directly with a QSFP cable, pooling memory to 128GB and expanding model support to up to 200 billion parameters while delivering twice the memory bandwidth. In NVIDIA's Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7x performance compared with a single system, according to the company.

The NVIDIA Sync app configures the multi-node cluster, with a cluster assistant feature that detects connected units, validates device configuration and configures the ConnectX-7 network. Every DGX Spark ships with a built-in NVIDIA ConnectX-7 NIC.

DGX Spark ships with the NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron open models, and runtimes including Ollama, vLLM and PyTorch with CUDA. Blender is among the first creator application providers to support the platform, with a prebuilt downloadable installer coming soon.

NVIDIA said a NVIDIA Sync Model Launcher is coming at the end of the month, allowing developers to download and launch Qwen3.8 27B on a single system or a cluster, and to set up OpenCode to use the model.

Supported inference frameworks listed include llama.cpp, Ollama, vLLM and LM Studio. Playbooks coming soon to 64GB devices include serving LLMs with vLLM, running OpenClaw with a local LLM, and connecting multiple DGX Sparks for distributed workloads.