Rack Emulator Design for Multigenerational IT Racks

Author: Pardeep Shahi, Senior Mechanical Engineer, NVIDIA; Steve Medeiros, Boyd Collection: 2025 OCP Global Summit - Data Center Facility

Summary

This presentation by NVIDIA and Boyd introduces a purpose-built rack emulator designed to replace traditional load banks and TTV (Thermal Test Vehicle) emulators for data center commissioning. The core problem is that existing commissioning tools are inadequate for modern liquid-cooled data centers: TTVs are designed for in-rack chip-level validation and cannot simulate full rack thermal and hydraulic loads, while traditional load banks are bulky industrial units (designed for generator testing) that cannot replicate realistic coolant flow resistance, pressure drops, or fit within a standard server rack footprint. Neither technology offers the combination of rack-compatible sizing, adjustable pressure drop simulation, flow/temperature/pressure data measurement, and safety certification needed for holistic facility commissioning.

The NVIDIA-Boyd rack emulator addresses all of these gaps. It is sized at a standard rack form factor (24" W x 60" D x 94.5" H), rolls on casters into the exact server rack position, generates up to 250 kW of heat load (scalable to 1 MW with a different form factor) in eight 31.25 kW increments, simulates pressure drops from 12 to 35 PSI via a PID-controlled valve with custom algorithm, measures inlet/outlet temperature, pressure, and flow rate with an electromagnetic flow meter accurate to +/-0.2%, and features an ASME-certified heater with emergency shutoff. It communicates via Ethernet Modbus TCP/IP and displays real-time data on a 15.6-inch color touchscreen HMI. All wetted materials are stainless steel with 3-inch tri-clamp sanitary connections. A safety interlock prevents heaters from activating beyond what the available flow rate can safely cool, eliminating the risk of glycol burning.

The key value proposition is multigenerational reuse: because pressure drop and heat load are software-configurable via the HMI, operators do not need to purchase new emulators each time NVIDIA releases a new rack generation. The emulator cost is roughly 1/100th of an actual IT rack, making it economically justified to validate infrastructure before deploying multi-million-dollar GPU servers. Prototype testing confirmed the unit can achieve any flow/pressure combination within its operating envelope, as demonstrated by PQ curves at valve positions from 20% to 100% open covering 0-600 LPM and 0-35 PSI.