NVIDIA has recently showcased its AI technology at the CES trade show, unveiling two deskside AI supercomputers: the DGX Spark and DGX Station. These are built for developers and researchers, with the aim of leveraging comprehensive AI models practically and conveniently from their desktops.
With NVIDIA's Grace Blackwell architecture, both the DGX Spark and DGX Station offer unified memory and AI performance at petaflop levels. This advancement aims to empower users to develop locally before scaling to the cloud as their needs evolve.
Running highly optimised open AI models once required large-scale infrastructure in data centres. Advancements in hardware and software now allow these models to run on desktop setups. Pre-configured with NVIDIA AI software and CUDA-X libraries, these systems hope to provide a simple, plug-and-play optimisation process for developers, researchers, and data scientists.
The DGX Spark is designed to be the foundation for developers, enabling the running of AI models directly from their desks. DGX Station supports the execution of larger and more intricate AI models for enterprises and research institutions. This includes running expansive models like NVIDIA's Nemotron 3 and others, from a desktop environment.
The DGX Station, enhanced with the GB300 Grace Blackwell Ultra superchip, can manage models of up to 1 trillion parameters, giving AI labs a significant tool for local scale model deployment. NVIDIA's collaborations are contributing to improved performance, yielding a 35% uplift when processing AI models, facilitated by partnerships like its collaboration with llama.cpp.
Beyond research, NVIDIA's deskside systems aim to cater to the needs of modern creators. By supporting the entire AI development lifecycle, from prototype to production, these deskside supercomputers accommodate a wide range of AI applications across various industries.
The DGX Spark enables creators to run video generation models, such as those from Black Forest Labs and Alibaba, at faster acceleration rates using NVFP4 technology. With these advancements, creators can offload workloads from conventional laptops, freeing up resources for uninterrupted creative workflows.
Alongside software leaders and the open-source community, NVIDIA's DGX systems are aiming to enable faster iteration cycles and greater data control on AI projects, providing a more interactive and user-friendly AI experience on the desktop.
As these systems become more accessible, DGX Spark is being used in projects such as enhancing urban mobility with TRINITY and developing AI-powered interactive agents.
