Kenanga Research & Investment

Technology - Exploring Generative Artificial Intelligence

kiasutrader
Publish date: Wed, 03 Apr 2024, 11:02 AM

In our recent knowledge-sharing session, Exploring Generative Artificial Intelligence (AI), expert speakers Mr. Benedict Khoo (Supermicro), Mr. Ko Yun Hung (SNS Network) and Mr. Lim Kah Chun (AIMS Data Centre) delved into this latest cutting-edge technology and the opportunities it brings to local players. Supermicro showcased its significant contributions to advancing AI/HPC with its GPU server systems. The company's strategic focus on modularity and in-house manufacturing ensures flexibility, cost-effectiveness, and high-quality server solutions, making it a preferred server choice. SNS Network highlighted the transformative potential of AI across various sectors in Malaysia. The company acts as a bridge that connects server manufacturers and buyers, enabling AI adoption and integration in Malaysia. AIMS Data Centre provided valuable insights into the critical role of data centres in supporting AI and GPU servers. AIMS's Tier 3 certification ensures high availability and reliability, with redundant systems and fault-tolerant components. The company also explains its air-cooling configurations and innovative liquid cooling solutions, which positions it as a preferred data centre provider in Malaysia.

Understanding AI Supply Chain. In our recent knowledge-sharing session, Exploring Generative AI, expert speakers Mr. Benedict Khoo (Supermicro), Mr. Ko Yun Hung (SNS Network) and Mr. Lim Kah Chun (AIMS Data Centre) delved into this cutting-edge technology and the opportunities it brings to local players. The physical session drew over 80 attendees. The speakers shed light on various aspects of the hardware supply chain, starting with the pivotal role of Supermicro (SMCI:US), a leading server producer, in laying the groundwork for AI innovation. Supermicro's servers serve as the cornerstone for housing Nvidia's advanced computing cards, laying the foundation for transformative AI capabilities. Subsequently, attention turned to the indispensable role of data center installer, SNS Network (SNS:MK), tasked with the critical responsibility of setting up and configuring the intricate infrastructure required for seamless AI operations. Finally, the session by AIMS Data Center — an associate of Time dotCom (TDC:MK) — underscored the interconnectedness of various elements in the AI ecosystem by being the final destination that houses the servers. This is where server owners will be dependent on AIMS for uninterrupted supply of power and cooling, as well as tight security to ensure seamless operation and zero down-time.

Supermicro

Mr. Benedict Khoo (Senior Director in Field Application Engineering) spoke on “Accelerating Advance AI/HPC with Supermicro GPU Server System”. Supermicro is a leading global provider of high-performance, high-efficiency server technology and green computing solutions. Since the launch of ChatGPT on 30 Nov 2022 which sparked the AI bull run, its share price has skyrocketed >11x, outpacing Nvidia (>5x). To better service its customers worldwide, Supermicro operates across multiple continents, including its Silicon Valley headquarters in San Jose, California, as well as facilities in Taiwan, the Netherlands and other regions. Being at the forefront of the AI space which provides servers that enable AI functionality, the company is highly reliant on research and development and this is evident through its human resource allocation. It has over 6,000 employees globally, half of which are dedicated to technical research and development.

Over the past year, Nvidia’s graphical processing unit (GPU) has emerged as a staple in AI, becoming the go-to hardware preference among hyperscalers for high-performance computing (HPC), AI, and deep learning workloads in data centers. Nonetheless, insights from Khoo’s presentation highlighted that a functional server necessitates an array of components beyond Nvidia’s GPUs to constitute a comprehensive system. Khoo has more than 18 years of experience in different roles from hardware validation division, server platform engineering and is now the senior director of Supermicro’s field application engineering division. He shared that Supermicro's crucial role lies in providing a complete server system by seamlessly integrating components such as RAM, storage, processor, mainboard, and power supply to complement Nvidia’s GPUs.

Moreover, Supermicro further gained a competitive edge through its focus on modularity. Khoo highlighted one of Supermicro’s fundamental strategies in server construction, termed the "building-block solution", which offers customers the flexibility to mix and match components to meet server requirements at competitive prices. This approach has positioned Supermicro as a preferred choice among hyperscalers, as its flexibility and backward compatibility enable cost-effective upgrades without the need to replace the entire system when only a part requires upgrading. Additionally, this strategy significantly reduces e-waste, aligning well with Supermicro’s commitment to green computing.

Located in the center of Silicon Valley, Supermicro benefits from close proximity to its technology partners, including industry giants such as Nvidia, Intel, Broadcom, and AMD. This strategic location ensures that the group remains at the forefront of technological advancements, granting it a competitive edge in launching new server products. Supermicro's strong track record in delivering innovative server solutions, coupled with its swift time-to-market, has garnered recognition from technology leaders like Nvidia. As a result, Nvidia collaborates closely with Supermicro, designating it as a preferred partner for showcasing new GPUs. Notably, Supermicro was among the first to introduce Nvidia's MGX design in August 2023, featuring the NVIDIA Grace CPU Superchip, within its ARS-221GL-NR system.

Khoo emphasised another key factor driving Supermicro’s server ahead of its competition - stringent quality control. Unlike many others who opt for outsourcing to contract manufacturers, Supermicro takes pride in its in-house fabrication capability, designing and constructing its own metal enclosures for server racks and power supply units. This commitment to in-house manufacturing ensures the highest structural quality, particularly crucial given that a fully equipped 42U server rack can weigh up to 5,500 pounds (or 2,500 kg), with each AI-grade server valued at approximately USD250,000 (or RM1.2m). Additionally, Supermicro's proprietary power supply system enables further innovation in cooling solutions. For example, the company achieved a higher number of GPUs systems from 4 to 8 units within the same rack form factor by transitioning to its in-house liquid cooling technology, departing from the traditional air cooling systems that require space-consuming fans (see Exhibit 5). With liquid cooling, the previously occupied fan space is repurposed for extra GPU slots, effectively doubling computational power within the same form factor. This development holds significant importance, particularly as hyperscalers commonly assess computational power per square foot. In addition, Supermicro’s liquid cooling has been proven to consume less power compared to traditional air cooling, further enhancing the group’s appeal (see Exhibit 6).

One of Supermicro’s technology showcases is the successful implementation of its SuperBlade platform (see Exhibit 7) to develop a supercomputer, named Pagasus, for the Computational Science Center at the University of Tsukuba. Pagasus represents one of the pioneering systems globally to leverage the power of the 4th Gen Intel Xeon Scalable processors, Intel Optane persistent memory, and the NVIDIA H100 Tensor Core GPU. With a theoretical peak performance of 6.51 petaflops in double precision and a measured LINPACK performance of 3.48 petaflops, Pagasus ranks 190th on the TOP500 list of fastest supercomputers in the world. Even more impressively, as of June 2023, Pegasus holds the 12th position on the Green500 list, boasting an exceptional 40.448 GFlops/Watt efficiency, which is recognised as one of the world's most energy-efficient Green Supercomputers. It's worth noting that such processing power holds the potential for applications in big data analytics and ultra-large-scale AI models, including critical tasks like weather forecasting and predicting potential natural disasters, enabling the development of preemptive measures.

Another innovative development from Supermicro is its proprietary AI-powered platform named KODA, which is designed to revolutionise user interactions through state-of-the-art visual kiosk technology (see Exhibit 8). The KODA platform is a digital advisor and concierge system that uses generative and predictive AI to assist customers in physical retail stores in finding routes to a specific product or giving product recommendations. Customers interact with KODA just like they would a store employee and can ask questions and receive answers using natural language.

For example in Exhibit 9, a customer enters a grocery store and is welcomed by Koda's digital avatar on a kiosk. Assuming the customer inquires about the beverage aisle, they are then promptly guided to the precise location with a real-time overview of the store layout. Unlike traditional wayfinding solutions dependent on legacy WIFI or GPS technology, KODA utilises the advanced capabilities of NVIDIA Omniverse to reinvent product-level navigation in three dimensions: X, Y, and Z. This brings about better accuracy and efficiency of locating products within an intricate environment while also eliminating the necessity for planogram compliance. Planogram compliance refers to a set of rules governing product placement and space layout used by FMCG brands for optimal customer targeting.

To harness the growing potential in the South East Asia (SEA) region, Khoo shared that Supermicro has plans to build a facility in Johor, Malaysia. The choice of Johor stems from the availability of land, a crucial factor as the group aims to establish a comprehensive plant, encompassing its in-house manufacturing facility, rather than relying on outsourcing to contract manufacturers in the region. However, Supermicro presence in the region will likely create an abundance of job opportunities for the residents in the area. In addition to space availability and growing push for solar energy, identifying Johor as its next expansion base aligns well with the growing data center campuses being built in Johor by the likes of ST Telemedia Global Data Centres, YTL Power International and Equinix. Other data centre operators that are planning expansions in Johor include Princeton Digital Group (PDG), AirTrunk, Keppel, ChinData's Bridge DC, and Yondr.

SNS Network

The event was followed by a presentation titled “Unlocking AI Experience” by Mr Ko Yun Hung, co-founder and Managing Director of SNS Network Technology (SNS Network). Established in 1998, SNS Network Technology (SNS Network) is primarily engaged in the provision of Information Communication Technology (ICT) solutions across Malaysia. SNS Network’s ICT division services a diverse group of clientele, ranging from individual end-users and small to medium-sized enterprises (SMEs) to esteemed educational institutions, government agencies, and multinational corporations. The group differentiates itself from the competition by having a dedicated team that focuses on innovative offerings and customised solutions to cater specific client’s needs, ensuring maximum efficiency, productivity, and profitability regardless of customer’s scale or sector.

Beyond its customized ICT solutions, SNS Network has a growing brand presence across Malaysia. Under the leadership of Ko, the group has developed a range of brand portfolio which includes: (i) iTworld, a renowned multi-brand Apple authorized reseller, (ii) GLOO, a leading multi-brand ICT retail store, and (iii) JOI, which delivers affordable consumer devices. Recognised for quality, reliability, and affordability, their brands have earned the trust of the Malaysian market. SNS was listed on Bursa Malaysia’s Ace Market in September 2022.

During Ko’s presentation, he opined that Generative AI marks the beginning of a potential new supercycle in the technological landscape and is set to supersede the previous two supercycles led by mobile internet in the 2010s and cloud computing which peaked during 2015-2020. AI has the potential to reshape the trajectory of future technological advancements across various sectors such as smart city and urban development, education and e-learning, healthcare and biotechnology, cybersecurity, manufacturing, retail outlets and even agriculture. An example provided by Ko was the usage of AI for optimising traffic flow patterns based on real-time weather forecasts, effectively mitigating congestion. This could be further enhanced with AI-driven predictive policing and emergency response systems which have the potential to proactively anticipate safety concerns, thereby fostering a safer and more secure urban environment.


Ko highlighted that the start of the AI cycle has presented various opportunities in Malaysia, such as data centre construction, data centre fit-out, testing and commissioning, and operation and management. While SNS is not involved in construction, it has the expertise to cater to the other three segments (see Exhibit 17), filling in the gap between server manufacturers (i.e., Supermicro) and server buyers. For instance, a customer who intends to procure Supermicro servers will engage SNS Network to take on the role of sourcing and planning, followed by integration into the selected data centre location where SNS Network will configure power distribution, cooling, and network connectivity. Subsequently, SNS Network will conduct necessary testing and commissioning procedures and offer monitoring and performance optimisation services. Going the extra mile, SNS Network also offers AI enterprise services, which include a range of solutions such as cloud computing, data analytics, and AI workload optimisation, catering to diverse business needs across various sectors in Malaysia. Notably, SNS Network was recently appointed as Supermicro’s partner in Malaysia, which is timely with Supermicro’s upcoming facility expansion in Johor.

AIMS Data Centre

Lastly, the event was wrapped up with a sharing session with Mr Lim Kah Chun from AIMS Data Center (AIMS), presenting on “How Data Centres Supports AI & GPU Servers”. Lim has more than 16 years in the data center industry and is currently overseeing three departments: Data Center Product, Data Center Presales & Solution and Data Center Services.

With many of us still trying to grasp what a data center is, Lim clarified that a data centre is a designated space within a building or a group of buildings specifically designed to accommodate computer systems and associated components, including telecommunications and storage systems. Data centres essentially charge rental of space, and the charges will also fluctuate depending on customer’s requirement on factors such as location, facility features, tier certification, power density and operational cost. Given the critical role of IT operations in ensuring business continuity, data centres typically feature redundant or backup components and infrastructure for power supply, cooling, data communications connections, as well as environmental controls such as air conditioning and fire suppression. Additionally, various security measures are implemented to safeguard sensitive information. Notably, large data centres operate on an industrial scale, consuming as much electricity as a small town to support their extensive operations and infrastructure.

Lim shared that all of AIMS’s data centres are Tier 3 certified. The tiering system is a classification system developed by the Uptime Institute to measure the reliability and redundancy of a company’s data centre. To qualify as a Tier 3 data centre, the facility has to ensure high availability, offering both redundancy and concurrent maintainability. Unlike Tier 1 and 2 data centers, a Tier 3 facility does not require a total shutdown during maintenance or equipment replacement. In a Tier 3 data centre, all IT equipment is dual-powered, and critical IT components are designed to be fully fault-tolerant. This means that in the case of a component failure, the system is still able to continue to operate seamlessly. AIMS achieved this by having uninterruptible power supply (UPS) systems which doesn’t just provide backup power but also filters out anomalies (ie. surges or spikes) to ensure stable power supply to critical equipment.

Additionally, Tier 3 data centres feature multiple power and cooling distribution paths, allowing for maintenance and upgrades to be performed without disrupting operations. From a layman perspective, it's commonly believed that power is the primary concern in a data centre. However, Lim's insights shed light on the equally critical role of cooling. He revealed that cooling is equally vital due to the continuous operation of servers, which generate substantial heat from their intense computational tasks. Without efficient dispersal of this heat, there's a risk of equipment overheating, potentially leading to performance issues or even damages. Therefore, AIMS has a variety of air cooling configurations (ie. room based, row based, and rack based), allowing customers to achieve optimal air cooling while maintaining a high efficiency in terms of space utilisation with ease of maintenance. The group also offers its in-house designed liquid cooling for customers who do not have liquid cooling fitted in their servers initially but intends to adopt such methods. This is done by submerging partial or the entire server in a dielectric, non-conductive coolant or liquid to quickly disperse heat away from the device.

These comprehensive offerings have positioned AIMS as one of the preferred data centres in Malaysia. AIMS is currently Malaysia’s leading interconnection hub, hosting 100% domestic and 80% foreign network services providers in its premises. In addition, AIMS data centres serves as a key cloud exchange in Malaysia offering direct cloud access to the likes of Azure, Google Cloud, AWS, Alibaba Cloud, Time Cloud Services and AWM Cloud. The group is currently building a new Tier 3 data center in Cyberjaya which will be done in three phases on a 3.13 acre land size. Despite the target completion date set at 2QCY25, Lim said that phase 1 and 2 are already fully sold out due to overwhelming demand.

Source: Kenanga Research - 3 Apr 2024

Discussions
Be the first to like this. Showing 0 of 0 comments

Post a Comment