Nvidia is adding a 64GB configuration to its DGX Spark lineup, giving local AI users a cheaper way to buy into the GB10 Superchip platform as memory prices continue to climb. Systems built on the 64GB GB10 from Acer, Asus, Dell, Gigabyte, HP, and MSI are scheduled to launch on October 23 at a starting price of $4,999. The new option arrives as prices for memory and storage remain unsettled, meaning those figures could shift.
Until now, unified memory machines such as Nvidia's GB10, AMD's Strix Halo, and Apple's M-series have commonly shipped with 128GB or more, based on the early assumption that large amounts of RAM would be needed to hold the most capable models. That calculation has changed as dense models including Qwen 3.8 27B can now run within 32GB of memory, though with limited context. Because a chip permanently paired with a large amount of costly LPDDR5X has become more of an entry barrier than a benefit, Nvidia is offering a lower-memory Spark for buyers who want the GB10 capabilities and software stack but do not need the extra capacity.
The price gap is significant. A 128GB GB10 system, when one can be found in stock, currently runs roughly $7,000 to $9,000, above even Nvidia's adjusted MSRP. The 64GB units keep the same ConnectX 7 RDMA NIC as their 128GB counterparts, so users who eventually exceed 64GB of RAM can still cluster multiple systems to expand memory capacity and inference performance. The 20-core Arm CPU complex and the 273 GB/s of shared memory bandwidth for the CPU and GPU carry over unchanged from the original Spark.
To make that networking easier to use, Nvidia is introducing the Nvidia Sync Cluster Assistant, a software tool that automates the process of linking Sparks together. It joins Sync, the remote connectivity and management app that lets users tap their Spark's AI horsepower from other Macs and PCs. The assistant supplements community-developed utilities that have appeared over the past year to simplify clustering, including ones used in experiments that joined two Dell GB10 systems earlier this year.
Nvidia is also adding a Model Launcher feature to Sync that automatically downloads and starts models such as Qwen 3.8 27B across one or more systems. The feature links those models to the OpenCode browser-based coding agent, allowing users to put their local token factories to work on agentic workflows right away. The 128GB DGX Spark and similar systems will stay on the market for buyers who need room for larger models, cluster scaling, or memory-heavy local work such as model fine-tuning.
For US buyers weighing a local AI setup, the addition expands the DGX Spark lineup at a moment when component pricing is volatile, giving enthusiasts another option to match their budgets and workflows.
More AI news from TechManNews.
