GPU WORLD · GW-LC-GPU-H100-SXM
H100 SXM cold plate
Configuration reviewEvaluate cooling for a four- or eight-GPU HGX H100 tray. The server revision determines how the plates mount and how the branches connect to the chassis.
Fit and compatibility
- Confirm the H100 SXM module, HGX board revision, GPU population and enclosure. Send the tray drawing or existing cooling assembly part number so the mounting and hose paths can be checked.
- Check the GPU and memory contact surfaces, access around the retention hardware, and any keep-out zones around the NVLink switch area.
Scope of supply
- Quote a plate assembly with its thermal interfaces and fasteners, or a tray loop with matched supply and return connections and service couplings.
- Define which CPU, switch or other board loads are in the liquid loop. Keep the chassis fans and air-side components in the acceptance scope.
Operating conditions
- NVIDIA specifies configurable H100 SXM TDP up to 700 W. Four GPUs at that limit account for 2.8 kW of device power before CPUs and other loads. Determine how much of the complete tray load enters the liquid loop.
- Balance the parallel GPU branches using the supplier flow and pressure-drop curves. Select the pump duty with the manifold, connectors and filter in the circuit.
Acceptance checks
- Measure each populated branch at the agreed supply temperature. Inspect joints after a tray service cycle and document any connector spill against its limit.
- Log GPU and memory temperatures, configured power and throttling indicators through the specified workload soak. Confirm airflow and temperatures for the other tray components.
Technical references
Use these references to identify the platform and design requirements. GPU WORLD confirms the applicable assembly and operating conditions for each project.
CONFIGURATION & QUOTATIONSRequest a cooling quoteShare your server configuration and site conditions.Request a quote
Plan your cooling project
Our team will review server fit, heat rejection and the scope of supply.