Preen power supplies for AI server testing
In today's digital age, AI servers are driving some of the world's most advanced technologies. Unlike conventional servers, they are designed to handle the extreme computational demands of AI - equipped with powerful GPU accelerators, high-speed connectivity and extensive memory to process massive amounts of data in record time.
AI servers are being applied across industries - from accelerating drug development in healthcare, to fraud detection in finance, to predictive maintenance in manufacturing plants. So it's not just about hardware, it's driving innovation across the industry.
The present and future of AI servers
The older generation of AI servers were designed to run on 12 V DC power, which is no longer sufficient today. Modern systems are therefore increasingly switching to 48 V - 54 V DC power, which stays within the 60 V safety limit while improving efficiency and reducing current load. Most AI servers today operate at 48 VDC, but the move to 54 VDC brings significant reductions in line losses and increases the overall efficiency of advanced server infrastructures.
In large-scale data centers, rack-mounted power distribution systems with around 400 V DC are already being used as an intermediate step from which the voltage is reduced to 48-54 V DC for individual servers.
To meet the rapidly growing power demands of AI servers, NVIDIA is developing high-voltage DC (HVDC) architectures with up to 800 V that power the racks directly. This concept minimizes the number of gears, reduces power transmission losses, and simplifies cabling - a critical step for the future of sustainable data centers. These approaches will be key for new platforms such as the NVIDIA Rubin Ultra AI Server, which is expected to be released in 2027 and deliver up to 600 kW of power per rack.
Testing AI servers with Preen power supplies
The rapid development of AI computing systems and the diversity of their configurations increases the need for reliable power testing. Preen DC power supplies provide stable and accurate power for AI server testing and guarantee high accuracy, reliability and compliance with validation test requirements.
Key requirements for AI server testing:
- High power: stable power supply for GPU clusters in the range of 5 kW - 300 kW, up to 1.8 MW in parallel.
- Programmability: support for preset test parameters (burn-in, aging) for easy automation
- Safety and compliance: built-in protection against overload, overvoltage and overheating of sensitive AI components
- Power fault simulation: generation of voltage dips, ramps and interruptions according to IEC 61000-4-29 to verify device robustness
- Flexible integration: support for SCPI, Modbus and LabVIEW interfaces for easy interfacing with automated systems
Example from practice - testing AI servers
The customer used the ADG-L-80-510-15 from the ADG-L low voltage series, which provides 15 kW at 80 VDC in a compact 3U design. It is used for long-term reliability testing of AI servers at unbalanced voltages (54.5 V DC ± 10%).
Thanks to built-in programmability, the test parameters can be preset with a single click - the power supply then performs the test autonomously. The integrated remote interface and Preen Program software allow remote test control and automatic report generation. For customers with their own test system, SCPI and Modbus commands are available for seamless integration.
