"WAIC 2025 showcases cutting-edge technologies, and the TTY national production large-scale model all-in-one machine makes its debut."
The technological wave of large models is getting increasingly powerful, driving industrial transformation while also accelerating the need to restructure IT hardware infrastructure such as servers and computing power centers.
Especially at the beginning of this year, DeepSeek emerged unexpectedly, accelerating the explosive growth of the markets for large model integrated machines and AI servers.
"We conducted numerous tests and performance optimizations, eventually resulting in a series of extensive product lines." Recently, at the highly anticipated 2025 World Artificial Intelligence Conference (WAIC 2025), Shenzhen Tongtaiyi Information Technology Co., Ltd. (hereinafter referred to as "TTY") showcased its four product portfolios and an AI large model all-in-one machine.
TTY was established in 2016. It is a professional manufacturer specializing in server product design, technology research and development, production and manufacturing, and customer service. Its products are widely used in key technologies such as artificial intelligence, cloud computing, big data, and information security. It has been awarded the "Key Small Giant" title by the National Ministry of Industry and Information Technology. To gain a deeper understanding of TTY latest exhibited products at this conference, Data Ape conducted an exclusive interview with Ma Ze， the general manager of TTY product center. We hope that this interview can provide some valuable references and inspirations for the industry development.
From WAIC 2025, Observing the Roadmap for the Domestication of AI Hardware
The WAIC 2025 conference was grandly opened in Shanghai this year. This annual conference in the field of artificial intelligence brought together the world's top enterprises, experts and innovative achievements. As a seasoned professional in the artificial intelligence industry, Ma Ze has some of his own opinions on the industry trends reflected by the WAIC 2025 conference. He believes that this year's conference clearly demonstrated several key development directions.
Firstly, from the perspective of hardware development trends, super nodes and high-density computing have become hot topics in the industry. Ma Ze pointed out: "Companies like Huawei 384 super node cluster and H3C's similar super node products demonstrate the efforts and achievements of all-rack-class servers in striving for extremely high computing density. With the continuous development of artificial intelligence technology and the increasingly complex application scenarios, the market demand for computing power has shown an explosive growth, which has directly driven the rapid evolution of hardware equipment towards super nodes and high-density directions to meet the growing computing needs."
Secondly, in terms of energy efficiency improvement, liquid cooling technology has gradually emerged as a focus of industry attention. The PUE (Power Usage Effectiveness) of traditional data centers is usually 1.4 or higher, indicating that there is still significant room for improvement in energy utilization efficiency. Ma Ze emphasized: "Under the backdrop of green energy conservation and low-carbon development, liquid cooling technology, with its outstanding heat dissipation performance and energy efficiency, has received increasing attention and research from more and more enterprises. Currently, the application scope of this technology is continuously expanding, and it is expected to become the mainstream heat dissipation solution in high-performance computing scenarios such as data centers in the future, providing strong support for the sustainable development of the industry. "
Finally, the trend of domestic substitution is unstoppable. Ma Ze mentioned, "In the past, we mainly relied on foreign brands like NVIDIA in the GPU card field. However, in recent years, the intensification of Sino-US trade frictions and the implementation of related restrictive measures have severely affected the import of GPU cards. Against this backdrop, the state has introduced multiple policies to support the development of the domestic GPU industry. More and more domestic GPU manufacturers have emerged like bamboo shoots after a spring rain, and some have already achieved practical application cases. Although there is still a certain gap between domestic GPUs and NVIDIA in terms of performance and ecosystem completeness at present, we believe that with the continuous progress of technology and the gradual improvement of the industrial chain, domestic GPUs will definitely occupy a more important position in the future AI market."
From "High Threshold" to "Out-of-the-Box Usability": Unveiling the Large Model All-in-One Machine
Data Yuan observed that at the conference venue, many enterprises showcased their own large-scale model all-in-one machines. Among them were Lenovo， Enflame, and TTY, among others.
It is worth noting that the explosion of this large model all-in-one machine is mainly influenced by DeepSeek, as well as the development trend of miniaturization of large models and computing power. Through the characteristics of "software and hardware collaboration, ready to use out of the box", it meets the needs of enterprises to use large models at low cost in specific business scenarios, and solves the three major pain points in enterprise AI deployment, namely data security, computing power cost, and scenario adaptation.
The large model all-in-one machine is a comprehensive solution integrating hardware, software and large model methods, providing users with personalized intelligent computing devices. It usually includes a combination of computing power such as CPUs/GPUs, storage, operating systems, AI platform software, various model algorithms, and inference services, among other hardware and software components.
What are the features of TTY large-scale integrated machine?
It is reported that the all-native workstation all-in-one machine launched by TTY, based on Kunpeng module and Ascend accelerator card, has achieved complete independent innovation from motherboard design, operating system to AI framework. The product supports on-machine deployment of large models and does not rely on external networks, which can meet the strict requirements of sensitive industries for "data not leaving the domain". It can provide secure and reliable AI computing power equipment for key fields such as government, finance, and energy. In response to the demand for domestic scenarios, TTY has established a full-stack solution matrix covering "lightweight" to "full-blooded" large models. It can meet the diverse customization needs of various industries.
At the exhibition site, a team commented: "Under the same computing power, the domestic solution provided by TTY can fully integrate with the domestic operating system ecosystem, completely resolving the long-standing supply chain risk issues it has faced."
During the interview, Ma Ze introduced: "In the past, the use of AI technology often relied on expensive servers, which made many individual users and small and medium-sized enterprises hesitate. However, the DeepSeek all-in-one machine exhibited by the company can be deployed on a single machine. The company has also meticulously developed workstations with various specifications ranging from 7B, 8B, 14B, 32B, 70B to 671B. The product sizes are diverse and can be flexibly selected according to different model sizes and application scenarios. This measure not only lowers the usage threshold but also expands its application scenarios, enabling more users to enjoy the convenience and efficiency brought by AI."
Secondly, in the enterprise-level applications, Ma Ze pointed out that there are some differences in the needs of individual users and enterprise users. For example, when a company department wants to deploy a DeepSeek all-in-one machine, they need to consider the compatibility with the company's existing product line and the stability of the equipment performance, among other issues. Since enterprise-level applications are usually deployed in data center environments, the traditional server architecture remains the mainstream choice. To meet the diverse needs of enterprise users, TTY has launched a complete set of all-in-one machine solutions, covering various configurations such as 2-card, 4-card, 8-card, and cluster solutions, providing enterprise users with comprehensive and customized options to ensure that enterprises can efficiently and stably apply AI technology in complex business scenarios, achieving digital transformation and intelligent upgrading.
In response to the extreme computing demands of large model training and inference, TTY also showcased a variety of different forms of AI servers at this event. These flagship products, with their outstanding performance, reliability, and high cost-effectiveness, have become the benchmarks in the global AI infrastructure field.
It is worth mentioning that, in addition to the previously released 4U4 card, 4U8 card, 7U8 card AI servers and other star products, at the exhibition site of this conference, the TTY flagship-level training and inference integrated AI server TG988V3 made its global debut and attracted much attention! This product supports 8 high-performance OAM GPU modules that comply with UBB2.0 specifications, and supports the fourth or fifth generation Intel® Xeon® Scalable processors. It features outstanding performance, strong scalability, and high reliability, and is suitable for large-scale distributed training and inference, high-performance computing, and large model all-in-one machine applications.
What technological innovations has the company made to address the current pain points of large models?
In recent years, with the rapid development of large-scale artificial intelligence models, higher requirements have been imposed on the computing power, cooling systems, and communication capabilities of servers.
Regarding the current pain points faced by large models, what technical innovations and improvements has TTY made?
Ma Ze elaborated that, first of all, a server based on the OAM 2.0 specification was developed. Unlike the plug-in card method of traditional module server architecture, the current TTY module adopts a decoupled design. It has 8 modules, and the 8 GPUs use a fully interconnected architecture. Each GPU has unique characteristics, thereby significantly improving the overall operational efficiency. This architecture not only supports NVIDIA's HGX architecture and other various architectures, but also can be compatible with domestic GPU cards, effectively reducing reliance on specific GPU brands and greatly enhancing the flexibility of the computing power center.
Secondly, in terms of cooling technology, TTY has adopted two solutions: cold plate cooling and immersion liquid cooling. Cold plate cooling quickly removes heat by direct contact, while immersion liquid cooling involves completely submerging the machine in a liquid to absorb heat. Through these technical means, TTY aims to reduce the PUE value, improve the energy efficiency ratio of the entire data center, and thereby effectively lower the overall TCO cost.
Ma Ze further pointed out that although the cost of converting to cold plates is approximately 1.2 times that of traditional air-cooling, in the long run, the cold plate technology can significantly reduce energy consumption, lowering the PUE from the original 1.4 to 1.2 or even lower, thereby saving a large amount of electricity and other funds for the data center. And liquid cooling technology is an important future technological direction. Liquid cooling technology not only significantly reduces energy consumption but also enables the performance of GPUs or CPUs to be maximized.
However, there is a major drawback to liquid cooling technology - the cost of customer renovations is too high. If liquid cooling technology is adopted, not only the servers in the customer's old data center need to be upgraded, but also the power supply facilities, ventilation ducts, and structural support of the data center. "Most newly-built data centers currently adopt the liquid cooling technology route. We believe that liquid cooling will become a standard feature for data centers in the future," Ma Ze mentioned. Liquid cooling is the company's strength and is also a key research topic at present.
With the digitalization wave sweeping across the globe, the application scenarios in various industries have become increasingly complex and diverse. From data-intensive scientific research projects to highly real-time industrial automation production, the demand for computing and storage resources has seen an explosive growth. To precisely match the diverse demands of different customers, TTY has implemented a full-scenario product layout and rigorous testing coverage strategy. The products cover the product system of four core categories: general servers, storage servers, artificial intelligence servers, and edge computing servers.
In the field of general servers, TTY covers both X86 and the innovation-driven routes. It has built a high-performance and highly reliable general computing power platform, enabling it to efficiently undertake various general computing tasks and support the digital transformation of key industries such as finance, government affairs, and telecommunications.
The storage server achieves the goal of "being able to store, retrieve and utilize books efficiently" through intelligent layering and global balancing technologies.
In the field of artificial intelligence servers, we are building a complete stack of AI infrastructure from training to inference, to support the implementation of large-scale models with trillions of parameters.
In terms of edge computing servers, in certain specific scenarios, it can perform local data processing without relying on cloud operations. After completing local testing, it can accelerate the overall data processing speed and reduce latency, while alleviating the pressure on cloud servers and forming an efficient mode of cloud-edge-end collaboration. In summary, we have developed corresponding mechanisms for different business scenarios, and with the layout of four products, we meet market demands.
According to the statement of Ma Ze , currently the revenue from the company general servers and AI servers accounts for the highest proportion. "With the rise of the AI era, the proportion of AI servers is constantly increasing and growing at a fast rate. Moreover, AI servers require GPU cards, which are more valuable than traditional servers, further driving the growth of revenue."
It is reported that previously, TTY also set up server clusters for research institutions. The successful establishment of this cluster has significantly improved the research efficiency, reducing the working hours required for the calculation and reasoning of the research personnel. Additionally, a cluster was set up for the local police, and the hyper-converged architecture integrates computing, storage and networking, which can significantly enhance the efficiency of data processing, accelerating the case analysis and decision-making process. The server cluster set up for an Internet company optimized its computing architecture to solve the CPU performance bottleneck and the low GPU utilization problem. The establishment of this cluster has increased the performance of the customer's GPU architecture by 2 times and saved 30% of the cost.
Looking to the future, Ma Ze stated that the company will also keep up with the technological trends and continuously plan and iterate its products based on customer demands.
Breaking the "bottleneck" problem: The rise of domestic products is underway.
During the interview, Ma Ze also shared his views on market opportunities, challenges and future development trends.
He believes that the current booming development of the information technology innovation industry has brought unprecedented opportunities to the company. With the continuous intensification of the Sino-US technology war, the information technology innovation industry has become an inevitable path for national development. Whether for strategic national considerations or for other reasons, solving the problem of "technical bottlenecks" in key core technologies is extremely urgent, which provides a broad market space for domestic manufacturers.
It is reported that TTY has been proactively developing information technology innovation products since 2019. Currently, its information technology innovation product technology routes have comprehensively covered products from Phytium, Hygon, and Kunpeng and can meet the needs of different industries for information technology innovation.
However, opportunities and challenges coexist. Ma Ze pointed out that there is a significant gap between domestic GPUs and NVIDIA in terms of performance and ecological integrity. This requires domestic manufacturers to invest more effort in the research and development of domestic GPUs. He mentioned that, taking the NVIDIA H20 chip incident as an example, Ma Ze believes that this is only a buffer measure taken by NVIDIA to clear inventory and reduce losses. "In the long run, we still need to adhere to the development route of self-control and autonomy, and continue to make efforts in the research and development of domestic GPUs to make up for the gap in performance and ecological construction with NVIDIA."
Ma Ze also mentioned that the current infrastructure of state-owned enterprises is facing numerous challenges in terms of computing power. The main issues are as follows: On one hand, the computing power centers built in various regions do not match the demand of industry applications, resulting in underutilization of computing power resources. The utilization rate has dropped to around 30% on many occasions. On the other hand, the process of domestic substitution is slow. Most data centers still rely on foreign brands such as NVIDIA. The domestic substitution of high-end GPU chips and high-speed network chips is extremely difficult. The domestic chip technology is not yet mature, and the adaptability and maturity of the industrial chain also need to be improved. Moreover, in terms of green and low-carbon aspects, the promotion of liquid cooling technology is limited by costs. Most regions still rely on traditional air cooling construction, and the application of software in various industries is still in its infancy.
Looking to the future, Ma Ze stated that TTY will continue to closely follow technological trends and constantly plan and iterate products based on customer needs. Whether it is expanding domestic and international markets or adapting to NVIDIA and domestic GPUs, TTY will be committed to providing customers with stable and high-performance hardware underlying platforms. On this basis, it will also conduct overall adaptation and optimization of models and software in accordance with customer needs, helping customers solve business pain points and facilitating the smooth transformation of their businesses.
