NVIDIA presents local AI hardware in Europe

During GTC Berlin, NVIDIA showcased its complete range of local AI hardware, with key innovations including the RTX Spark, DGX Spark, and DGX Station. These systems are designed for various applications; RTX Spark supports personal agents on Windows PCs, while DGX Spark and DGX Station cater to developers and enterprise teams running advanced workloads on Linux. Recent innovations include the availability of the Underdog Desktop Agent and new updates for the DGX Spark software.

The RTX Spark allows users to run personal agents, accelerating content creation and enhancing gaming experiences. The system offers a memory capacity of up to 128 GB and impressive AI computing power of 1 petaflop. The Underdog agent utilizes a model that runs entirely locally, meaning users do not have to share their data with external servers. Various manufacturers, including ASUS, Dell, and Microsoft, already offer compatible laptops.

The DGX Spark is aimed at developers and offers systems with 64 GB of memory, enabling them to run more powerful models by connecting multiple systems together. The NVIDIA Sync cluster assistant enables teams to run simultaneous sessions against the same model, which is ideal for collaborative projects. With the DGX Station, which offers up to 748 GB of coherent memory and more than 20 petaflops of AI performance, users can run advanced models and benefit from new integrations with various enterprise management tools.

Additionally, NVIDIA introduces Recursive Flint, a local coding model optimized for the RTX Spark, and provides day-zero support for various new AI models. By collaborating with open-source communities such as llama.cpp and vLLM, NVIDIA also accelerates the agentic workloads on local hardware, improving the speed and efficiency of AI applications. With these innovations, NVIDIA positions itself as a leading force in the local AI market.

Read the full article from Nvidia.