What Is Compute Tray?_
In high-density GPU systems like the NVL72, a compute tray is a hot-swappable assembly that contains multiple GPU modules, their associated host processors, and system memory. Each tray plugs into a backplane or midplane that provides power, data, and cooling connections, simplifying maintenance and scaling by allowing individual units to be replaced without disturbing the entire rack.
Technical Details
Compute trays typically integrate NVLink copper spine connections for GPU-to-GPU communication within the rack, avoiding fiber for that internal fabric. They align with liquid cooling manifolds or cold plates via quick-disconnect coolant fittings. The tray's backplane interface carries high-speed PCIe or NVLink signals, plus management and power lines, following the OEM's specified pinout and keying. Structural rails and alignment pins ensure repeatable seating and proper thermal contact with the rack's cooling infrastructure.
How Leviathan Systems Works with Compute Tray
During rack assembly, Leviathan Systems field crews install compute trays into the rack chassis, securing them with captive fasteners and connecting coolant hoses to the tray's manifolds. We verify each tray's seating and perform a power-on check before commissioning the full NVL72 system.
Related service
GPU Rack Assembly & Rack-and-Stack →