CONNECT WITH US

Pegatron reveals new AI server product equipped with Blackwell; set to ship in 4Q24

Chloe Liao, Taipei; Jack Wu, DIGITIMES Asia 0

Photo: Pegatron chairman T.H. Tung (R) and CTO Yean-Jen Shue (L). Credit: DIGITIMES

Computex 2024 kicks off this week with Nvidia CEO Jensen Huang firmly in the center of the spotlight.

The AI competition between various companies is expected to intensify. On June 3, Pegatron Chairman T.H. Tung and CTO Yean-Jen Shue showcased the latest AI server solutions and AI smart manufacturing at their Guandu headquarters in Taipei. Huang has repeatedly cited Pegatron as a crucial Taiwanese supply chain partner, hinting that despite Pegatron being a latecomer to the AI server field, it has not fallen behind.

The most notable of Pegatron's server achievements revealed at this event are the latest GB200 NVL72 and NVL36 enterprise-level AI server solutions, jointly developed with Nvidia. Shue also mentioned that products featuring Nvidia's latest Blackwell architecture could begin shipping in the fourth quarter of 2024.

At Computex 2024, Pegatron will also display the MGX 4U server, developed in collaboration with Intel and featuring both CPU and GPU.

Regarding the closely followed server market progress, Pegatron began shipping a small number of AI server products in February 2024. Shue revealed that the company is currently targeting regional nation-level data center orders and not just focusing on large CSP customers.

Specifically, data centers in regions such as the Middle East or Morocco in North Africa are not dominated by large CSPs. On top of that, there are small to mid-sized CSP customers in the North American market in negotiations as well. Shue humorously noted, "Even if the scale isn't as big as a football field, it can still be the size of a basketball court," indicating a sizeable business opportunity.

There is considerable interest in how Pegatron will perform in the AI server market, mainly because Pegatron previously missed out on opportunities in the server sector due to its decision to focus on smartphones. Its return to the server sector has led to concerns that it will be a latecomer struggling to catch up.

Tung reiterated that recent killer applications in AI will redefine all products and services in the future, and no single entity will monopolize this vast market. He likened the current stage of AI to the early morning sun at 08:00 am, when it's just beginning to shine, and urged the public to relax. The real concern shouldn't be where the business is coming from, but whether one's foundation is stable enough.

Shue admitted that Pegatron indeed missed out on traditional server markets before. However, thanks to AI, the gap between everyone has been reduced since the addition of CPU and GPU is irreversible. Pegatron has been preparing in the server industry for a long time, and now is the best time to "break out of the shell."

With the advent of GPUs, in addition to large CSP customers, many small and mid-sized customers are also gradually making requests. Pegatron's extensive ecosystem, including partnerships with Nvidia, Intel, and AMD, allows it to provide ready-made solutions based on customer needs.

Currently, Pegatron can supply everything from motherboards (L6) to complete server racks (L11). The highlights of the announcement on June 3 are the latest GB200 NVL72 and NVL36 enterprise-level AI server solutions developed with Nvidia. The former, in particular, is a liquid-cooled rack platform with up to 72 GPU nodes, designed specifically for processing computation-intensive workloads.


In addition, Pegatron will showcase the MGX 4U server, developed in collaboration with Intel, at Computex 2024. It features CPU and GPU architectures and can accelerate AI training and inference tasks.

Moreover, with cooling playing an increasingly important role in AI server design, Pegatron showcased its current L11 liquid cooling solution, Holley, which can enhance performance while saving energy by 40%. Through its system monitoring software and RPU flow control cooling mechanism, the solution supports a cooling capacity of 65KW.

Besides server-related solutions, Pegatron has frequently been named by Nvidia as a smart factory partner. In this regard, Pegatron emphasized that in addition to AI advancements, 5G and AIoT are also crucial. Pegatron uses the Nvidia Omniverse platform with its self-developed PEGAAi platform to create a digital twin factory solution. This solution allows for the planning, designing, and optimizing production processes in a virtual environment, enabling remote collaboration in the future.