NVIDIA launches DGX Spark 64GB

NVIDIA launches DGX Spark 64GB with 64GB of unified memory. The device is aimed at users who work with fewer resources. It costs US$ 4,999 in the US.

The DGX Spark 64GB uses 64 GB of unified memory, allowing AI models with up to 100 billion parameters to run in environments with fewer resources. The memory is shared between the CPU and GPU, which reduces latency and improves efficiency. Connectivity via NVIDIA ConnectX-7 makes communication between devices easier, while the Sync Cluster Assistant allows scaling up to 128 GB using two units.

NVIDIA launches DGX Spark 64GB with unified memory for local AI — Image: blogs.nvidia.com

This configuration is. Horizontal scaling offers flexibility for growth without interrupting operations. The combination of memory and connectivity is essential for tasks that demand local AI processing. NVIDIA expands its local AI hardware lineup with the DGX Spark 64GB, complementing coverage of tools for secure access to high-performance storage and generative recommendation systems.

Adapter, Electronics, Hardware — Image: digitaltrends.com

The new configuration is part of a strategy to meet demand for local AI hardware, focused on inference, fine-tuning and the development of autonomous agents. The company already offers the DGX Spark 128GB, which costs US$ 4,699, but the 64GB model is more expensive, at US$ 4,999, despite having half the memory. This runs counter to the trend of lower prices for versions with fewer resources, but it reflects demand for performance in memory-constrained environments.

Computer, Electronics, Tablet Computer — Image: digitaltrends.com

The configuration is aimed at users who can work with less memory, offering an alternative for developers who don't need full scale. The system can scale to 128GB with two devices via NVIDIA Sync Cluster Assistant, keeping flexibility for growth without interrupting operations. The launch takes place in the US, with availability exclusively through OEM partners starting at US$ 4,999.