Build Daily
Tinley Park · September 9, 2026GB10 Dgx Spark 2 Node Cluster From Scratch
This video demonstrates the process of setting up a two-node GB10 DGX Spark cluster from scratch using Nvidia Sync on a MacBook. It covers the initial boot, software installation, and running a VLLM inference test to verify that both nodes are active and functioning correctly.
The outline
Tap a timestamp to jump there in the video.
Introduction and Setup
The creator introduces the GB10 hardware and explains that the setup will be performed from a MacBook using Nvidia Sync.
Nvidia Sync and Initial Boot
The video shows how to access the initial first boot software and connect to the Spark via the Nvidia Sync interface.
Cluster Memory and Testing
After a long setup duration, the cluster is confirmed to be working with 218 gigabytes of total memory, and VLLM is introduced for inference testing.
Inference Performance
The creator runs a test showing both machines are active with GPUs hitting 93-94 percent utilization, confirming the cluster is operational.
Conclusion
The video concludes by highlighting the successful creation of a two-node cluster.
A two-node cluster built from compact hardware and managed from a laptop shows how solo builders can scale local inference without cloud overhead. The setup relies on standard networking and vendor tooling, making it a practical path for testing distributed AI workloads on owned infrastructure.