AI Server

Why Derive AI Servers

In the application of AI, 5G, Big Data, Internet of Things, Edge Computing, etc., the internet data has been increasing by a multiple of times, which is a serious test for general servers, which use CPUs as the main source of computational power. Currently, the number of cores of a single CPU is close to the limit for the CPU, but the data amount is still increasing, so it is necessary to increase the server's data processing power, and thus the emergence of AI servers.

Impact of AI servers?

Generative AI has ignited competition and business opportunities for technology giants. Taiwan holds nearly 60% of the world's advanced manufacturing capacity and nearly 80% of server shipments, and the main power module manufacturers supplying AI servers are all based in Taiwan. Generative Artificial Intelligence (AI) requires a huge amount of computing power, which has led to an increase in the power consumption of server chips. From all aspects, AI has led to the regeneration of the computer industry, and has also brought golden opportunities to Taiwanese enterprises, which is a new challenge and opportunity for power supply manufacturers.

What is an AI Server?

AI servers are advanced computing systems designed to handle complex AI training and inference, combining powerful hardware resources and specialized software tools to support a wide range of AI tasks that require efficient processing of large amounts of data and complex computations. They may employ multi-core processors, graphics processing units (GPUs), or specific AI gas pedals such as TPUs to speed up training and inference operations. These hardware components are capable of processing large amounts of data and performing parallel operations, providing faster and more efficient AI processing.

What are the AI server application scenarios?

Based on the advantages of AI servers, the application, in addition to chat robots, extending to the data searching, office work, finance, education, entertainment and other applications, from the enterprises to the clients have become more and more realistic applications appearing in the market. AI servers in the health care, search engine, gaming, e-commerce, finance, security, and other industries have a wide range of applications.

Smart Healthcare

Managing patients is expensive, and the healthcare system must shift from treatment to prevention and long-term care management. Through machine vision, knowledge mapping, deep learning and other artificial intelligence technologies, we can simulate the medical professional’s thinking, reasoning diagnosis, helping doctors to locate the condition, assisting in making diagnosis, and improve the efficiency of medical care.

Application Scope:

1. As a patient: Which department should I go to? Who is the most famous doctor? When will you arrive at the clinic? H ow to take the medicine prescription by the doctor, what to pay attention to, side effects?

2. As a doctor: a clinic queue up to 150 every day, there are patients requesting to be added to the list, but there is not enough time. Each patient can not speak two sentences and in some special cases, it is very challenging to take time to think about the treatment plan. By drilling AI a large amount of medical information exchange and analysis of the use of medical diagnostic assistance can enhance the quality of medical care.

3. Hospital management: how to design the medical check process to be efficient, saving patients' time, laboratory tests, how to design the examination process so that no mistake will be made, how to manage & warn of epidemic infectious diseases, how to avoid misplacing and losing tens of thousands of medical materials.

Face Recognition, Speech Recognition, Fingerprint Recognition

Through deep learning, machine learning and other technologies, it can realize image data training such as pictures and videos. High-intensity dynamic pedestrian flow, GPU acceleration is enabled to conduct ultra-fast search algorithms for massive face databases.

Application Scope:

1. Security control and preservation

2. Access control and access management

3. Employee time and attendance clocking

4. Public safety and technology law enforcement, wanted persons matching & assisting in searching for missing persons & lost elderly

5. Public computer login unlocking, with the popularization of the network government digitization push, the risk of leakage of government secrets or personal data is increasing daily, in the awareness of the rise of the protection for information security, in order to protect sensitive information, many agencies or units have begun to promote relevant information year by year. To protect sensitive information, many agencies and organizations have begun to promote relevant measures year by year.

Security Monitoring

The use of knowledge map technology, deep learning and other technologies can be applied to human body analysis, image analysis, vehicle analysis, behavior analysis and other security scenarios. Operators in large-scale and wide factory need to wear masks, helmets, goggles, sun hats, etc. all day long due to long hours of work in outdoor environments, coupled with epidemiological management and personal safety considerations, resulting in localized masking of the human face, which makes face recognition more difficult.

Application fields:

1. Intelligent factories and warehouses for security 2.

2. Intelligent security control systems for home use

3. Preventing unknown or blacklisted personnel from entering premises.

4. Public transportation passengers and flight crews boarding and checking in, border entry control

5. Using specific resources or professional machinery and equipment.

Intelligent Retail and Personalized Customer Experience

Utilizing technology, it is to provide customers with a more convenient, fast, and safe consumer experience, helping brands create online and physical store sales environments, optimizing the consumer shopping experience and process, and providing operators with more accurate decision-making analysis based on online consumer shopping behavior analysis and physical store historical sales data for more accurate future sales forecasts. In addition, in the retail industry, there are also common application scenarios such as unmanned sales and face payment.

Application Scene:

1. Retail: Smart advertising for electronic signage

2. Service industry: Identifying VIP customers and providing member benefits

3. Non-contact payment

4. AI analysis of customer behavior

Intelligent Financial Services Scenario

In the entire financial and insurance operations, it is necessary to strengthen fraud prevention, prevent identity theft, use a fast and cost-effective way to securely open an account, allowing customers to simplify the remote processing process, while protecting the integrity of the account, saving customer waiting time on-site, speeding up the efficiency of service delivery to speed up and strengthen the customer identity verification process.

Application Scope:

1. Determine credit and borrowing value intelligence analysis

2. Automated ordering of investment transactions

3. AI robot financial advisor

4. Customized and fast insurance recommendations

5. Credit card fraud detection

6. AI stock picking investment ordering, using machine learning algorithms to screen reliable investment strategies

The difference between the power supply used in generative AI server and general server?

In recent years, AI applications have entered a whole new generation of applications and development, and with the practical application and development of ChatGPT, the world's expectation on AI is full of unlimited imagination. And such an AI application must be built on a very powerful computing power. From the development of "Mining machine" (Digital currency mining) a few years ago to the widespread application of "AI" today, the most important and significant chip module, it is GPU module. Therefore, with the recent rapid development of GPU, all the computing power has been strengthened, which in turn creates the rapid response and learning of generative application systems. Compared to general servers, the narrow definition of AI servers refers to servers equipped with AI chips (such as the aforementioned GPU/TPU chips), while the broader definition is that servers equipped with at least one GPU are considered AI servers.

Under such a new application, are there any new requirements and specifications for the application and demand of "power supply"? The following points can be explained:

Power requirements:
Power demand: The wattage of power supply is multiplying.

In traditional server operation, a single unit is equipped with about two to four CPU (Central Processing Unit) modules and lower-order GPU modules. Generative AI servers, however, use software algorithms for large model training and inference, and not only does the computing core function from the CPU to the GPU (Graphics Processing Unit), but also to the GPU (Graphics Processing Unit). The number of chips has also doubled, and require as many as two CPUs and eight GPUs. Compared to general-purpose servers in the past, the power requirement has increased five to six times.

Voltage requirements:
System demand voltage varies.

Generative AI servers require as much as 54V, while traditional servers have 12V as the mainstream. The main reason for this is The main reason for this is for efficiency improvement. As power demand continues to increase, the related issue of energy consumption is becoming more and more apparent. The issues of energy saving and power saving (cost saving) have to be solved. The issue of energy saving and power saving (cost saving) is a major challenge that must be faced, so the only way to increase efficiency and reduce losses is to increase the "applied voltage". The only way to reduce losses is to increase the "applied voltage" so that efficiency can be improved and losses can be reduced. The conversion efficiency of traditional server voltage (12V) is about 96%, and the voltage of generative AI server (54V) is 97.5%. The conversion efficiency of traditional server voltage (12V) is about 96%, while that of generative AI server voltage (54V) is 97.5%.

Stability requirements:
Voltage drift requires extremely response stability.

In the process of powerful GPU computing, it will constantly generate transient high current (Peak Current), and in such a process, the PSU must still be able to maintain a stable voltage to accurately provide the power supply to maintain the system needs. During this process, the PSU must be able to maintain a stable voltage and provide accurate supply to meet the system requirements. Therefore, the response time must be fast and stable. Therefore, the related response speed must be fast and stable.

As a result of the above application development, the trend of power supply development will be oriented towards the "3 H": High performance, High power, and High density. ZIPPY has been providing the market with products with high power density, extremely small size, and the highest performance in the past development process. ZIPPY is looking forward to contributing its own technological energy in the field of AI and joining hands with global system developers to push the application of AI to the next era.

Recommended Models