What Is Sled?_
In GPU data-center platforms, a sled is a standardized unit that packages the GPU, CPU, memory, local storage, and liquid cooling cold plates into a single hot-swappable assembly. Sleds are inserted into a backplane or midplane within a rack enclosure, enabling dense, serviceable GPU compute nodes. They typically connect to the rack’s power, data, and liquid cooling infrastructure through blind-mate interfaces.
Technical Details
Sleds are built to a specific form factor defined by the platform vendor, with dimensions and connector locations that align with the rack enclosure’s backplane. They use blind-mate fluid couplings for liquid cooling loops, which automatically engage when the sled is fully inserted. The GPU-to-GPU NVLink spine is carried over copper traces on the backplane, not through the sled itself. Sleds are designed for tool-less insertion and removal, with alignment rails and latching mechanisms to ensure proper seating.
How Leviathan Systems Works with Sled
Leviathan Systems field crews handle sled installation as a core rack-population task, sliding each sled into its bay and verifying that the blind-mate power, network, and coolant connectors engage correctly. We also perform sled-level commissioning tests, including power-on and GPU link checks, before the rack is integrated into the broader cluster.
Related service
GPU Rack Assembly & Rack-and-Stack →