What Is Switch Tray?_
In a GPU data-center rack, the switch tray sits between the compute trays and contains the NVSwitch chips that enable high-bandwidth, low-latency GPU-to-GPU communication over copper NVLink cables. It typically occupies a specific slot in the rack’s backplane and is hot-swappable for serviceability. The switch tray does not handle external network traffic; that role belongs to separate NICs and top-of-rack switches connected via fiber or copper to the compute nodes.
Technical Details
The switch tray uses copper NVLink cables to connect each GPU in the compute trays, forming a fully connected all-to-all topology within the rack. Each NVSwitch ASIC in the tray provides multiple NVLink ports, and the tray’s backplane routes these links to the compute trays via blind-mate connectors. Power and cooling for the switch tray are provided by the rack’s shared infrastructure, with airflow directed through the tray’s heatsinks per the OEM’s thermal design. The tray’s firmware manages link training and error handling, and it must be updated in coordination with the GPU drivers to maintain compatibility.
How Leviathan Systems Works with Switch Tray
During rack assembly, Leviathan Systems field crews install the switch tray into its designated slot before cabling the compute trays, ensuring the backplane alignment is correct. In commissioning, we verify that all NVLink links show as active in the management interface and that the switch tray’s firmware matches the compute tray’s GPU driver version.
Related service
GPU Rack Assembly & Rack-and-Stack →