NVIDIA announced the release of a new 64‑GB configuration of its DGX Spark personal AI supercomputer, which will be available from major OEM partners—Acer, ASUS, Dell, Gigabyte, HP, and MSI—on Friday, Oct. 23, at a starting price of $4,999. The system ships with the NVIDIA Grace Blackwell superchip, 64 GB of unified memory, NVIDIA ConnectX‑7 networking, and the full NVIDIA AI software stack, including DGX OS, CUDA‑accelerated libraries, and the NVIDIA Agent Toolkit. It is designed to run large open models locally, supporting up to 100‑billion‑parameter models entirely on device. When workloads exceed a single unit’s capacity, two DGX Spark systems can be clustered using NVIDIA Sync Cluster Assistant, which automatically configures the ConnectX‑7 network and pools memory to 128 GB, delivering up to 1.7× performance compared to a single system. The platform also includes a model launcher that allows developers to download and run Qwen‑3.8 27B with a few clicks, and it supports popular runtimes such as Ollama, vLLM, and PyTorch with CUDA. Blender is among the first major creator applications to support the platform, with a prebuilt installer forthcoming. The 64‑GB SKU maintains the same software stack as the 128‑GB model, providing an accessible price point while enabling local AI development and scaling for larger pipelines.
NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI
NVIDIA is launching a new 64‑GB version of its DGX Spark personal AI supercomputer, available from major OEMs starting Oct. 23. The system ships with DGX OS, NVIDIA AI software, and a 64‑GB Grace Blackwell superchip, enabling local deployment of up to 100‑billion‑parameter models. Two units can be clustered via NVIDIA Sync Cluster Assistant for up to 128 GB and 1.7× performance gains. The platform supports popular runtimes and includes a model launcher for Qwen‑3.8 27B. The 64‑GB SKU is priced at $4,999.
Why it matters
The 64‑GB DGX Spark lowers entry barriers for developers needing large‑model inference without cloud costs, while its clustering capability offers scalable performance for growing workloads.