NVIDIA DGX Spark 64GB Enhances Local AI Capabilities

NVIDIA introduces the DGX Spark with 64GB unified memory, available from October 23 through leading manufacturers such as Acer, ASUS, and Dell. This system is designed for developers and researchers looking to build local AI applications without relying on the cloud. With cutting-edge specifications and a wide range of supporting software, the DGX Spark is ready to use from day one.

The DGX Spark combines powerful technologies such as NVIDIA Grace Blackwell compute, unified memory, and NVIDIA ConnectX-7 network functionality. This provides a robust local AI platform where developers can test their models with their own data. The new configuration supports up to 100 billion parameters and positions itself as an accessible alternative to the 128GB version, allowing more users to take advantage of powerful AI.

In the new 64GB configuration, two systems can be easily clustered using the NVIDIA Sync Cluster Assistant. This system automatically detects the connection between the units and configures the network, allowing developers to focus on their projects rather than the infrastructure. This enables users to effectively scale their usable memory capacity as their needs grow and provides improved performance up to 1.7 times that of a single unit.

The DGX Spark is designed with various applications in mind. Developers can use the system to run AI agents continuously, support powerful AI applications on regular computers, or easily scale their infrastructure as their processes become more complex. The 64GB configuration is available starting at a price of $4,999 and is aimed at various workflows, from model inference to extensive collaboration between multiple systems.

Read the full article from Nvidia.