Nvidia is positioning its new DGX Spark and DGX Station systems as deskside supercomputers designed for developers working with open-source and frontier Artificial Intelligence models. Announced at the CES trade show, the machines are aimed at organizations and teams that want high performance model development capabilities in a local environment instead of relying solely on remote data centers.
The company says the DGX Spark and DGX Station let developers run the latest open and frontier Artificial Intelligence models directly on a deskside system, with support ranging from 100-billion-parameter models on DGX Spark to 1-trillion-parameter models on DGX Station. By offering this level of capacity in a form factor that can sit near a developer’s workspace, Nvidia is trying to shorten iteration cycles for training, fine-tuning, and experimentation while maintaining access to very large model sizes.
Both systems are powered by the Nvidia Grace Blackwell architecture, which is described as combining large unified memory with petaflop-level Artificial Intelligence performance. Nvidia presents this combination as giving developers new capabilities to develop applications locally while retaining a straightforward path to scale workloads to the cloud. The focus is on enabling a workflow where teams can prototype, refine, and validate models on DGX Spark or DGX Station, then extend or deploy those models across larger infrastructure as demand grows.
