Testing AI datacentres

The 720kW IT Rack

Fitting 15 x 2kW GPU on a 2U server will deliver 720kW IT load in a 48U cabinet.

At $25K per GPU it equates to $9,000K per IT cabinet.

This is going to demand a robust test & commissioning programme.

The solution to protect this investment will be to use a Chip Load Emulator CLE to deliver a dynamic heat flux direct to chip load profile .

Integration testing of the AI datacentre

Testing only to the CDU secondary cooling loop is not sufficient to protect the AI GPU investment.

We now need 360 CLE secondary loop on chip testing.

With the need for integrated commissioning testing for AI datacentres extending beyond the CDU cooling system infrastuture, a CLE proves inline manifolds, server tray valves pressure and flow rates to the chip cold plate.

The ability to dynamically profile the heat flux to specific chips performances, will ensure future proof assurance of the AI datacentre in a post Robin Ultra timeline.