[Editor's Note] People say that it has gone beyond global warming to tropicalization. Along with the hot earth, the IT industry is also heating up, and the ChatGPT and various AI service crazes are also heating up data center servers. The demand for data, which was already increasing, has pushed up demand forecasts with the AI service boom, and the server market has triggered strong demands for power usage efficiency (PUE) in terms of ESG and long-term operating costs.

In this interview, we asked Vic Mallyalah, Senior Vice President of Business Development at Supermicro, a global leader in server solutions, about Supermicro's server market trends and server deployment solutions for the growing needs of AI. In addition, Kim Seong-min, Director of Business Development at Supermicro Korea, answered questions and shared domestic news.

■ Please introduce yourself.

Big Mallyalah: As Supermicro’s Field Applications Engineer and Senior Vice President of Business Development, I lead the worldwide field engineering team. Our team helps enterprises, service providers, and other customers understand their requirements and business needs and build optimized solutions.

Kim Sung-min: As Supermicro FAE and Business Development Director, I manage the Korean FAE team. Like Big, I understand customer requirements and needs and support the construction of solutions optimized for unique applications required in the Korean market.

■ Telecommunication companies are also building cloud infrastructure, and many enterprises are putting effort into deploying their own clouds and data centers. What is the background behind this increased data center construction?

Big Mallya: As many people have experienced, artificial intelligence (AI) is becoming more advanced day by day in terms of intelligence and usability. AI, which provides convenience, productivity, help and fun, and is becoming more powerful and advanced day by day, has infinite potential and opportunities for service and monetization in algorithms, semiconductors, software platforms and applications.

As AI grows in scale, data centers are being assigned more work, and this is increasing every day. For AI training, it is ideal to deploy cutting-edge servers equipped with the latest CPUs and GPUs, taking into account power and cooling requirements. Even with AI at the edge, there is still a need for data centers to be closely involved in AI technology.

Supermicro is proud to offer Green Computing, an AI hardware platform that minimizes energy consumption by optimizing hardware. Supermicro’s system solutions are optimized to accelerate a variety of AI solutions in autonomous driving, chat GPT, medical apps, scientific research, weather forecasting, and Level 4 factories.

■ The problem of excessive power consumption in data centers is emerging as a social and environmental issue. Supermicro is highlighting a new cooling solution as an answer to this issue. Please introduce it.

Big Mallyalah: Supermicro recently unveiled the industry’s first liquid-cooled server featuring NVIDIA HGX H100 (Delta-Next) GPUs and NVIDIA HGX 4-H100 (Redstone-Next). The product is a rack plug-and-play liquid-cooling solution that improves performance and reduces operating costs while minimizing environmental impact.

The new liquid-cooled NVIDIA H100 system reduces server cooling costs by up to 80% and overall data center power consumption by up to 30%. NVIDIA is one of our key partners and we will be sharing more news on this soon.

Supermicro liquid cooling solutions also mean customers can get the latest liquid cooling technology directly from Supermicro. A Cooling Distribution Unit (CDU) with dual pumps and power supplies can efficiently cool an entire rack up to 80kW via cold plates and cooling distribution manifolds (CDMs).

Kim Sung-min: Supermicro is currently working with Korean partners to meet the growing demand for water cooling solutions and explore growth opportunities in this market.

■ How much of an effect can we see in terms of total cost of ownership (TCO) compared to air cooling? Also, I wonder if there is any concern about system damage due to coolant leakage.

BIG: The biggest advantage of water-cooled data centers is that it reduces the power usage of the data center. According to customers, it can reduce the PUE (Power Usage Effectiveness) of the data center from the industry standard of 1.6 to about 1.06.

For example, the power required to drive CDU pumps for liquid circulation is so small that it can easily be offset by reduced cooling costs. Computing can be deployed at a higher density in a single rack, which also reduces the physical space in the data center.

The latest connectors can eliminate concerns about coolant leaks. In years of testing direct-to-chip water-cooled servers for both CPUs and GPUs, there has never been a single coolant leak.

■ I mentioned a rack-type water-cooled AI development system suitable for office and home office environments. I am curious about the current status of local adoption of Supermicro's AI development system and future market outlook.

BIG: Supermicro is very interested in liquid-cooled AI development systems. For the workstation form factor, it has dual 4th generation Intel Xeon Scalable CPUs and four NVIDIA A100 GPUs, all liquid-cooled. As a result, system noise levels are also reduced to acceptable levels for offices and homes.

Kim Sung-min: With this type of closed-loop water-cooled workstation, researchers and developers can develop AI prototypes more quickly and conveniently right at their desks without compromising hardware performance. There is no need to compromise performance for a quiet working environment, and demand for closed-loop water-cooled workstation form factors is also increasing in domestic academia and research institutes.

Korea is one of the fastest growing AI markets in both hardware and software, and Supermicro opened an office in Korea in 2021 to meet the market demand. As the most open hardware platform provider, we look forward to working with existing Korean customers as well as new opportunities with Korean AI startups.