泰鉑科技模塊化液冷賦能AI數(shù)據(jù)中心Taybo Modular Cooling Powers AI Data Centers
應(yīng)上海市某AI服務(wù)商邀請，泰鉑科技為其AI算力中心定制了積木式AI算力熱管理液冷模塊的應(yīng)用方案。
Upon request from a leading AI service provider in Shanghai, Taybo developed a bespoke modular liquid cooling solution for thermal management in its AI computing infrastructure.
項目概況 Project Overview
本項目為AI訓(xùn)練服務(wù)器熱管理項目。本項目甲方需求AI服務(wù)器的每個機柜熱負(fù)荷20kW，共設(shè)置40個服務(wù)器機柜，總熱負(fù)荷800kW。目前，泰鉑AI算力熱管理液冷模塊的服務(wù)器機柜單機柜最大散熱量為120kW。
This project is an AI training server thermal management solution. As required by the client, each AI server rack has a thermal load of 20kW, with a total of 40 server racks deployed, resulting in a combined thermal load of 800kW. Currently, TechPro's AI computing thermal management liquid cooling module supports a maximum heat dissipation capacity of 120kW per server rack.
氣候條件 Meteorological Conditions
項目位于上海市，該地區(qū)夏季（7-8月）濕球溫度常達(dá)28-30℃，歷史最高濕球溫度約31℃。夏季極端最高氣溫39.4℃，冬季極端最低氣溫-10.1℃。
The project site is situated in Shanghai, characterized by summer (July-August) wet-bulb temperatures ranging between 28-30°C and a historical maximum of 31°C. Recorded extreme temperatures include 39.4°C in summer and -10.1°C in winter.
熱管理解決方案 Thermal Management Solution
硬件組件工作溫度
Hardware Component Operating Temperature
AI服務(wù)器的整體工作允許溫度60℃，為保證設(shè)備高效運行，控制溫度在55℃以下。
AI servers are designed to operate up to 60℃, but a sustained temperature ≤55℃ is recommended to maintain computational efficiency and extend hardware lifespan.
采用泰鉑科技積木式AI算力熱管理液冷模塊，室外冷源1.5倍冗余設(shè)計。設(shè)計方案如下圖所示：
The solution utilizes Taybo's modular AI computing thermal management liquid cooling modules, with a 1.5x redundant outdoor cooling source design. The detailed system design is shown in the following diagram:
項目共設(shè)置3臺冷卻塔（每個冷卻塔額定散熱量400kW），搭配6組CDU，一組CDU搭配5-7個AI服務(wù)器液冷機柜，UPS單元單獨設(shè)置。一體化整合CTS + CDU + PDU等設(shè)備，減少現(xiàn)場安裝復(fù)雜度，提升系統(tǒng)可靠性。其中冷卻塔采用316不銹鋼一體化彎折工藝，耐腐蝕、抗嚴(yán)寒，適應(yīng)極端氣候環(huán)境，智能控制系統(tǒng)優(yōu)化水、電消耗，實現(xiàn)高效節(jié)能運行。本項目系統(tǒng)通過精確的控制冷卻液的流速、溫度和壓力，PUE低至1.03，顯著降低數(shù)據(jù)中心運營成本。通過隔離TCS和SCS回路，減少交叉污染和泄漏風(fēng)險，提高系統(tǒng)的整體可靠性。
The project deploys 3 cooling towers (each rated at 400kW heat dissipation capacity), paired with 6 CDUs (Cooling Distribution Units). Each CDU supports 5-7 AI server liquid-cooled racks, with UPS units configured separately. The system features integrated CTS (Cooling Tower System) + CDU + PDU (Power Distribution Unit), reducing on-site installation complexity and enhancing overall reliability. The cooling towers utilize 316 stainless steel with seamless bending technology, offering corrosion resistance and extreme cold tolerance for harsh climates. An intelligent control system optimizes water and power consumption, ensuring energy-efficient operation. The system achieves a PUE as low as 1.03 through precise control of coolant flow rate, temperature, and pressure, significantly reducing data center operational costs. By isolating the TCS (Tertiary Cooling System) and SCS (Secondary Cooling System), the design minimizes cross-contamination and leakage risks, improving overall system reliability.
溫度論證：AI服務(wù)器側(cè)供液最高溫度40℃與服務(wù)器運行溫差最小15℃，滿足散熱需求。上海市室外濕球溫度最高約為31℃，與一次側(cè)載冷劑溫差5℃，滿足換熱需求。高效板換對流換熱，換熱溫差僅為2-3℃。因此，本項目配置滿足AI算力需求，且在滿足冗余的情況下最大限度的節(jié)能。
Thermal Performance Verification: The maximum supply coolant temperature for AI servers is 40℃, with a minimum 15℃ temperature differential from server operating temperatures, ensuring effective heat removal. Shanghai’s peak wet-bulb temperature reaches 31℃, maintained at a 5℃ differential from the primary coolant, meeting heat exchange requirements. The high-efficiency plate heat exchanger achieves a 2-3℃ temperature differential during convective heat transfer, optimizing energy efficiency. So,The system design meets AI computing thermal demands while balancing redundancy and energy-saving optimization.
泰鉑科技積木式AI算力熱管理液冷模塊——以創(chuàng)新散熱技術(shù)，賦能AI計算未來！
Taybo Modular AI Computing Thermal Management Liquid Cooling Module-Empowering the Future of AI Computing with Innovative Cooling Technology！
