Nvidia links eight DGX Sparks into one cluster: 1TB of unified memory lets small teams run GLM 5.3 locally
Synopsis
At GTC Berlin, Nvidia laid out a local AI hardware line running from Windows PCs to desk-side workstations, and said DGX Spark can be manually configured into clusters of up to eight units with 1TB of unified memory, enough to run frontier-class models such as GLM 5.3. Previously, local devices were bounded by single-machine memory and ran smaller models or leaned on the cloud. A companion NVIDIA Sync Model Launcher reduces model download, container setup and OpenCode connection to a few clicks.
Interpretation
Local AI hardware moves from single-machine use to small-team shared clusters, with DGX Spark supporting up to eight manually configured units and 1TB of unified memory in total. Previously, the memory capacity of a single DGX Spark limited the model size it could run, and larger models required cloud or shared infrastructure. Nvidia's stated comparison is that the eight-unit cluster's 1TB of unified memory is enough to run frontier-class models like GLM 5.3, suited to small teams sharing one cluster as a backend with each developer running concurrent sessions.
The cluster connects through built-in NVIDIA ConnectX-7 networking and NVIDIA Sync Cluster Assistant, so developers do not build their own interconnect. Multi-machine scaling previously required separate network configuration and cluster management tooling. Nvidia lists coding models the cluster can run, including Qwen 3.8 Flash Next, DeepSeek v4 Flash 0731 and GLM 5.3 Flash, and publishes a DGX Spark cluster playbook alongside.
NVIDIA Sync Model Launcher reduces local model acquisition and hookup to a few clicks, downloading a model and container matched to the configuration and connecting it to OpenCode. Previously, deploying a model on a local device meant manually choosing the model, configuring a container and wiring it into developer tools. The feature arrives later this month for a single DGX Spark or a cluster, and automatically loads DGX Spark Skills for troubleshooting and device maintenance.
Perspective
The report covers Nvidia's local AI hardware and software updates announced at GTC Berlin, spanning the RTX Spark, DGX Spark and DGX Station lines plus companion models and inference accelerations. The intended readers are developers and engineers who need to run models and agents locally or within a team. Nvidia's stated next step is to consult the DGX Spark cluster playbook and try the local AI inference recipes.
The claim that 1TB of unified memory suffices for GLM 5.3 comes from Nvidia's own description without independent reproduction. Clusters are manually configured and capped at eight units. DGX Station for Windows and early-access Cloud Functions are expected or forthcoming, not yet broadly available.