针对 GPU 功耗升至数千瓦、单机峰值电流近千安的趋势,ITECH 用 IT-M3900D 直流电源 + IT8100E 大电流电子负载,覆盖 AI 服务器 DC-DC 转换器的低压输入模拟与动态负载仿真。
本文为英文技术文章,以下正文为 ITECH 官方原文(英文)。概述:随着 AI 算力爆发,GPU 供电从 12V 走向 48V/54V 乃至 800V HVDC,DC-DC 转换器须在极低电压下输出上千安电流;ITECH 方案以 IT-M3900D 模拟母线供电、以 IT8100E 以 75A/μs 高速瞬态拉载,完成动态响应、电压调整率、纹波与效率等全面验证。
GPU Power Architectures and Key Test Focuses
1. Traditional Architecture: 12 V Bus → VRM
2. Mainstream Architecture: 48 V/54 V → Power Distribution Board (PDB) → 12 V → VRM
(e.g., ORv3, NVIDIA DGX platforms)
3. Next-Generation Architecture: 800 V HVDC → 12 V → Vertical Power Delivery (VPD)
(e.g., NVIDIA MGX and next-generation AI computing centers)
Traditional low-voltage bus architectures are increasingly constrained by excessive conduction losses and thermal challenges associated with high-current transmission, making them unsuitable for next-generation high-power AI computing systems. As a result, 48 V/54 V intermediate bus architectures have become the industry standard, delivering an optimal balance between efficiency and power density through multi-stage power conversion. These architectures are now widely adopted in high-density AI servers and computing clusters.
Looking ahead, 800 V HVDC architectures are emerging as the preferred solution for hyperscale AI data centers. By simplifying the power conversion chain and significantly reducing transmission losses, they enable megawatt-class power delivery per rack.
Regardless of the system architecture, the GPU DC-DC converter remains the critical component responsible for precise voltage conversion and stable high-current delivery, serving as the foundation for reliable AI accelerator operation.
For high-power GPU applications, DC-DC converters and power modules are characterized by low output voltage, ultra-high output current, and extremely high conversion efficiency. These devices must support multiple bus voltage levels while accurately stepping down the voltage to the GPU operating range and delivering kiloampere-level output currents with high efficiency across the entire operating range.
To address the demanding load characteristics of modern GPUs, the industry has established comprehensive and rigorous qualification requirements for DC-DC converters. Critical evaluation items include:
These tests ensure the overall performance and long-term reliability of the power converter.
To meet these testing requirements, ITECH provides a comprehensive solution combining the IT-M3900D Series Programmable DC Power Supply with the IT8100E High-Current Electronic Load. This solution covers the two key aspects of DC-DC converter validation:

Figure 1: ITECH AI Server DC-DC Converter Test Principle Diagram
Advantages and Product Features of the ITECH Solution
? IT-M3900D Series High-Power Programmable DC Power Supply
The IT-M3900D series features high power density and wide output range, enabling simulation of 48 V and 12 V bus environments within a compact footprint. Its CC/CV priority function allows users to switch control loop response modes according to different DC-DC module startup behaviors, preventing voltage overshoot and improving test safety and compatibility.
? IT8100E Series Ultra-Low Voltage High-Current DC Electronic Load
The IT8100E series high-speed electronic load provides core support for AI server DC-DC dynamic testing. With extremely low operating voltage and ultra-high current slew rate, it accurately reproduces current transients caused by GPU workload transitions between active and idle states. The three-range current design enables high-precision measurement from mA-level standby current to kA-level full load, covering full operating conditions of DC-DC converter validation.
System-Level Testing Capability
The ITECH test solution addresses two key requirements in the evolution of AI server power architectures:
In addition to DC-DC converter testing, ITECH also provides complete AI server test solutions for PSUs, BBUs, and SST systems. Through coordinated AC/DC source and load integration, ITECH delivers a comprehensive and industry-focused validation approach for verifying transient response, efficiency, and stability of next-generation AI computing power systems.
For more information, please visit: https://www.itechate.com