Odm Ai Server Manufacturer

With the rapid development of artificial intelligence (AI) technology, the demand for AI servers continues to rise. These servers require not only powerful computing capabilities but also face increasingly severe cooling challenges. This article will delve into the classification of AI servers, their cooling requirements, cooling methods, and the advantages and applications of liquid cooling circuits as a mainstream cooling solution.

AI Server Forms

Subtitle Description

Tower Server

A vertical and horizontal chassis structure, suitable for regular office environments. The chassis is large, offering expansion space and good cooling functionality. Typically, they have a single processor and lower reliability without redundancy. Cooling solutions mainly rely on internal fans and heat sinks. Suitable for corporate websites, multimedia apps, etc

Rack Server

Width of 19 inches, height measured in 'U' units (1U = 1.75 inches). Designed to minimize space usage and reduce data center hosting costs. Cooling solutions usually adopt front panel intake and rear panel exhaust, utilizing airflow design within the rack for effective cooling. Suitable for cloud computing, software-defined storage, etc.

Blade Server

A high-density server platform that provides greater computing power by optimizing space. The cooling solution is more complex, typically including internal heat sinks in the blades and fan walls at the back of the cabinet to ensure efficient cooling. Suitable for supercomputing centers, cloud computing platforms, etc.

Cabinet Server

The core form of future data centers, integrating computing, networking, and storage into one. Usually equipped with independent cooling units, using airflow and liquid cooling modules to effectively manage the cabinet’s temperature. Easy to maintain, suitable for virtualization, big data analytics, and rapid integrated deployment scenarios.

Cooling Requirements of AI Servers

The cooling requirements of AI servers mainly arise from the following aspects:

High Power Consumption Characteristics

AI computing usually requires a large amount of computational resources, resulting in a significant increase in the power consumption of processors and GPUs. According to IDC's forecast, by 2027, energy consumption in AI data centers will grow at a compound annual growth rate (CAGR) of 44.7%, necessitating corresponding cooling systems to quickly and effectively dissipate excess heat. The power consumption of a single AI server may reach several kilowatts, posing a tremendous challenge to the overall energy consumption and cooling capabilities of data centers.

High-Density Deployment

The high-density deployment of AI servers significantly increases the power density within a single cabinet, further intensifying cooling challenges. Effective cooling solutions are crucial for ensuring the stable operation of servers, especially in supercomputing centers and cloud computing data centers, where liquid cooling systems can effectively reduce hotspot issues caused by device density.

Green and Environmental Requirements

With global emphasis on sustainable development, cooling solutions for AI servers must meet low energy consumption and low carbon emission standards. Reducing Power Usage Effectiveness (PUE) through effective cooling technology has become an industry consensus, especially in companies that comply with Environmental, Social, and Governance (ESG) standards, where liquid cooling technology is seen as an important means to achieve green development goals.

Long-term High-load Operation

Many AI applications require 24/7 uninterrupted operation, which places higher demands on the stability and durability of cooling systems. Servers are prone to overheating under prolonged high-load conditions, leading to performance degradation or even hardware damage. The high heat capacity characteristics of liquid cooling systems enable them to effectively maintain safe operating temperatures for devices under extreme loads.

Cooling Solutions for AI Servers

Currently, the main cooling methods for AI servers include the following:

Air Cooling

Air cooling is a traditional cooling method that uses fans to circulate air to dissipate heat. This method is low-cost and technically mature, but in high power, high-density AI server environments, its cooling effectiveness often fails to meet demand. Under high-load conditions, air cooling systems may lead to increased fan noise and are prone to heat accumulation in cases of poor airflow.

Liquid Cooling

Liquid cooling is an efficient cooling method that utilizes the high heat capacity of liquids to quickly dissipate heat. The efficiency of liquid cooling systems far exceeds that of air cooling, making it suitable for high power density and long-term high-load operations of AI servers. Liquid cooling technology can be implemented by setting up cooling circuits inside and outside the server, ensuring that heat can be effectively removed in a timely manner.

Phase Change Cooling

Phase change cooling technology utilizes materials that absorb and release heat during phase changes, making it suitable for specific application scenarios. Although phase change materials theoretically have good thermal management properties, their complexity and cost limit their application in AI servers, which is still in the exploratory stage and requires more experimental validation for feasibility under high-load conditions.

Hybrid Cooling

Hybrid cooling solutions combine the advantages of air and liquid cooling, aiming to optimize cooling efficiency. By switching cooling methods based on different load conditions, overall system energy efficiency can be improved. This solution's flexibility allows it to be applied in various environments, especially during transitional periods or special projects.

Heat Pipe Technology

Heat pipes are a high-efficiency cooling technology that utilizes the principle of phase change and contains a working fluid inside. During heating, the fluid evaporates and transfers heat to the cooling area, and upon cooling, it reverts to liquid state, achieving rapid heat transfer. Heat pipes are increasingly being adopted in the cooling design of high-density AI servers, particularly where high efficiency and small size are required.

Cooling Solutions for AI Servers – Liquid Cooling Loop

Subtitle Description

Basic Principles of Liquid Cooling Loops

Liquid cooling loops utilize coolant flowing within the server to absorb heat through components like cooling plates and transfer it to radiators, ultimately releasing it into the external environment. The thermal conductivity efficiency of liquid cooling systems far exceeds that of traditional air cooling solutions, making them an ideal choice for cooling AI servers. The fundamental principle of this technology lies in leveraging the high specific heat capacity and good thermal conductivity of liquids to quickly dissipate heat generated by computing nodes.

1.Coolant

The coolant is a key component in the liquid cooling circuit, responsible for removing heat from within the server. Common coolants include water, glycol, and mineral oil.

2.Cooling Pipes

Cooling pipes are the channels through which the coolant flows, connecting the heat sinks inside the server to the external radiators. Cooling pipes need to have good thermal conductivity and corrosion resistance.

3.Cold Plate

Cold Plates are the heat dissipation devices in the liquid cooling circuit, transferring heat from the coolant to the external environment. Cold plates typically use finned or plate structures to increase the surface area and enhance cooling efficiency.

4.Pump

The pump is the power device in the liquid cooling circuit, responsible for driving the coolant to circulate within the loop. The performance of the pump directly affects the cooling efficiency and stability of the liquid cooling circuit.

5.Control System

The control system monitors and controls the operation of the liquid cooling circuit, including parameters like coolant flow, temperature, and pressure. This system ensures that the liquid cooling circuit operates at optimal conditions, enhancing cooling efficiency and stability.

Advantages of Liquid Cooling Loops

1. High Cooling Efficiency

Liquid cooling systems can quickly and effectively dissipate heat generated by servers, particularly excelling during high-load operations. This efficient cooling capability allows servers to operate stably at higher performance levels, thereby enhancing the overall efficiency of AI training and inference.

2. Reduced Energy Consumption

Servers using liquid cooling technology can significantly lower energy consumption, thereby reducing carbon emissions. Compared to air cooling, liquid cooling solutions demonstrate superior energy efficiency, especially important in data centers aiming to lower overall PUE.

3. Adaptability to High-Density Deployments

Liquid cooling systems are designed to be flexible, supporting high-density server deployments to meet modern data center requirements for high-performance computing. In high-density environments, the advantages of liquid cooling become even more apparent, effectively preventing the formation of hot spots.

4. Reduced Noise

Liquid cooling systems typically operate at lower noise levels than air cooling systems, making data centers quieter during operation and providing a better working environment, especially in places requiring minimal noise interference, such as high-performance computing labs.

Development of Liquid Cooling Technology

Liquid cooling technology is continually evolving, mainly including the following forms:

1.Cold Plate Liquid Cooling

Cold plate liquid cooling is the most widely used liquid cooling technology today. It involves installing cooling plates inside the server, designing the coolant's flow path into a loop to achieve efficient heat exchange. This technology performs exceptionally well in retrofitting traditional data centers and reusing infrastructure, making it the preferred solution for modernizing many data centers.

2.Full Liquid Cooling Solutions

Full liquid cooling systems immerse the entire server in coolant, achieving comprehensive thermal management. This method's advantages lie in its thorough heat management; however, its complexity and cost are relatively high, making it suitable for large data centers and specific application scenarios. Full liquid cooling solutions can provide more uniform cooling effects but require higher maintenance and technical support.

3.Single-Phase Immersion Liquid Cooling

Single-phase immersion liquid cooling involves completely immersing computing nodes in liquid, utilizing the liquid's high heat capacity for heat dissipation. This approach has a higher technical threshold and is suited for applications with extremely high cooling demands, such as large supercomputers. Such systems can achieve very low temperature gradients, ensuring efficient operation.

Liquid Cooling Loop For AI Servers

Subtitle Description