NVIDIA launches 64GB DGX Spark AI supercomputer for local development

NVIDIA has introduced a new 64GB unified memory configuration for its DGX Spark personal AI supercomputer, aimed at providing a more affordable entry point for local AI development. The new model, based on the GB10 Grace Blackwell architecture, is scheduled for release on October 23 through partners including Acer, ASUS, Dell, Gigabyte, HP, and MSI.

Starting at $4,999, the 64GB version is designed to address rising memory costs and the specific needs of developers who require local inference and fine-tuning capabilities without the necessity of the 128GB capacity. Despite the reduced memory, the system retains core features such as the 20-core Arm CPU, Blackwell GPU, and ConnectX-7 networking.

Users can expand their capabilities by clustering up to four units using NVIDIA Sync, which can pool memory to support models with up to 200 billion parameters. The platform supports a wide range of software, including NVIDIA CUDA, DGX OS, and popular frameworks like PyTorch and Ollama, enabling the local execution of models with up to 100 billion parameters.