By PCMag BrandX for NVIDIA
Publication Date: 2026-10-02 16:35:00
The age of agentic AI, in which systems can be fully autonomous and act on their own, has arrived. And as the availability of open models that fit into the VRAM of a single GPU grows, so does the number of developers, students, and AI tinkerers looking to build and run their own tools.
But the hardware ecosystem has been slow to accommodate the rapid shift in resources that local AI developers need. Many of today’s AI PCs come with 32GB of RAM and integrated GPUs that fall short, especially when it comes to running inference on large language models (LLMs). While gaming rigs generally have more compute power, they also hit their limits when it comes to handling demanding local models.
When the NVIDIA DGX Spark AI supercomputer was unveiled at CES 2025, it boasted a high-performance Grace 20-core Arm CPU and a Blackwell GPU in a mini PC for desktops. It also came with 128GB of memory to help you build your most ambitious creations yet.
Introducing the New NVIDIA DGX Spark 64GB
Starting October 23, the new NVIDIA DGX Spark 64GB configuration ships from NVIDIA OEM partners – Acer, ASUS, Dell, Gigabyte, HP and MSI – with the same architecture as the DGX Spark 128GB, but with 64GB of coherent unified memory. The new spec meets local resource requirements for running a new generation of 30-35B class models and autonomous agents, without requiring an enterprise-grade price tag, making it more accessible to developers, students, and enthusiasts.
…
