DigiKey TechForum: 800V Hot-Swap-Based AI Data Centers, “Power Protection” and “Telemetry” - 2
TECHWORLD ·
✦ AI Summary
Hot-swap controllers sit at the very front end of the system power path, making them the basis for telemetry that measures voltage, current, and power, detects and records events such as overcurrent, undervoltage, and thermal shutdown, and provides temperature information for the MOSFET and adjacent PCB area.
Accurate hot-swap telemetry analyzes load current in real time to improve energy demand forecasting accuracy and is used for predictive maintenance, event logging, and load profiling.
Data centers are adopting high voltage in response to rapidly rising per-rack power demand, with a distributed power architecture that directly supplies ±400V or 800V from a power sidecar to the IT rack and a 1MW rack target mentioned.
Telemetry’s role in high-voltage power delivery is introduced. The hot-swap controller sits at the very front end of the system power path.
For that reason, hot-swap controllers are well suited to monitoring and data collection. Normal operating conditions mean staying within the specified range.
The essential measurement items that require accurate readings here are voltage, current, and power. They must also detect and record events such as overcurrent, undervoltage, and thermal shutdown.
In addition, important information about the temperature of the power switch (MOSFET) and the adjacent PCB area is also included among the data to be provided. This electrical and thermal data forms the basis for accurate hot-swap telemetry.
Accurate hot-swap telemetry is used to analyze load current in real time. Analyzing load current in real time improves the accuracy of energy demand forecasting.
The functions supported by hot-swap telemetry include predictive maintenance, event logging, and load profiling. These supporting functions improve data center uptime and optimize energy planning.
In a distributed power architecture, configuration changes are made by separating the PDU, BBU, and CU from the main IT rack and moving them to adjacent power sidecars. This structure moves existing power devices out of the IT rack and into power sidecars.
In the new voltage distribution scheme, ±400V or 800V is supplied directly from the power sidecar to the IT rack. In other words, high voltage is sent directly from the power sidecar to the rack.
The main reason data centers are pushing to adopt high voltage is the sharp increase in power demand per rack. As rack power demand rises, busbar current increases as well.
As busbar current increases, larger and heavier busbars are needed to maintain thermal and electrical performance. As a result, larger and heavier busbars increase the burden of design and installation.
Voltage increases are presented as a way to address this burden. Raising the voltage makes it possible to reduce the current needed to deliver the same amount of power.
As voltage rises, busbars can be smaller, lighter, and easier to handle. This is important for maintaining efficiency, space utilization, and scalability in AI racks, where power demand continues to grow.
The current in-rack power used by some systems is 48V.
Some current configurations use a method that raises PSU output to maintain 48V inside the rack, receives ±400V or 800V at the rack input, then steps it down to 48V and connects it to the existing busbar.
However, this approach is transitional in nature, and the general per-rack limit for such systems is about 250kW.
The goal of increasing rack power to 1MW is then mentioned, with the aim of maximizing the computational performance of a single rack.
To accommodate more compute equipment, securing space is presented as a prerequisite, and installing the PSU and BBU outside the IT rack is recommended as the preferred arrangement. For source information on the related content, DigiKey’s TechForum/Article is cited.
Source: TECHWORLD · Lee Gwang-jae
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406703
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.