Largest Supercomputing Facility in Taiwan


High Performance Computing Largest Supercomputing Facility in Taiwan Print Page • Customer: Ubilink • Location: Taiwan • Industry: Supercomputer, HPC Project background In a groundbreaking move to advance the Taiwan AI revolution, Foxlink Group has joined forces with Ubitus to lead its subsidiary, Shinfox Energy, in founding a new company – Ubilink. Together, they are creating Taiwan’s largest computing power center with an impressive overall performance of 45.82 PFLOPS, surpassing NVIDIA's estimated theoretical value of 40 PFLOPS.[1] It will offer public cloud services, AI-computing-power rentals and subscription-based cloud services, including platform-as-a-service (PaaS) and software-as-a-service (SaaS), for training large language models. This new data center is also remarkable for handing customers the freedom to choose packages that draw only on renewable energy. The challenge Building such a vast and high-performance infrastructure from scratch posed significant challenges. Ensuring the seamless operation and maintenance of this AI supercomputing center was no small feat. Any instability could result in frequent interruptions, driving up operating costs and affecting overall efficiency. Our solution ASUS is fully committed to supporting Ubilink in constructing a global top AI computing center from the ground up. The center will house 128 NVIDIA® H100 servers, equipped with 1,024 GPU cards, making it the largest green-energy-powered, AI-computing service center in Taiwan. Remarkably, this ambitious project was completed in just three months. Infrastructure installation: ASUS expertly managed the construction of the supercomputing infrastructure, including the data center build-out, cabinet installation, rigorous testing, and onboarding. Each phase was meticulously executed to ensure optimal performance. Cooling and power management: Managing thousands of GPUs is no easy task, with ASUS refined the liquid-cooling system to ensure the power consumption meet the industry-standard acceptance level. Auto-detection and recovery tools: Given the complexity and potential failure scenarios in such a large-scale GPU cluster, traditional CPU clusters simply don’t compare. ASUS implemented advanced automated detection and recovery tools that minimize the impact of hardware failures, ensuring minimal downtime and maintaining system reliability. Software platform and services: ASUS provides an integrated development platform and software stack that streamlines the rapid deployment and optimization of AI applications, making it easier than ever to harness the full power of the center’s capabilities. Why ASUS? The status of ASUS as exclusive hardware supplier ensures that we deliver the fastest, highest-quality equipment, backed by a professional technical team. This makes us the strongest pillar of support for Ubilink’s supercomputing ambitions. ASUS AI servers, renowned for their unmatched computing power and stability, coupled with our integrated development platform, have garnered global acclaim. We empower customers to rapidly deploy and optimize AI applications, driving innovation and efficiency – ready to embrace the era of AI. Recommended Model ESC N8-E11 \| ASUS Servers [1] NVIDIA DGX results from the June 2023 summit, where the NVIDIA DGX H100 (128 units) achieved a performance of 40.66 PFLOPS, which can serve as a benchmark for this case.

High Performance Computing

Largest Supercomputing Facility in Taiwan

Print Page

• Customer: Ubilink

• Location: Taiwan

• Industry: Supercomputer, HPC

Project background

In a groundbreaking move to advance the Taiwan AI revolution, Foxlink Group has joined forces with Ubitus to lead its subsidiary, Shinfox Energy, in founding a new company – Ubilink. Together, they are creating Taiwan’s largest computing power center with an impressive overall performance of 45.82 PFLOPS, surpassing NVIDIA's estimated theoretical value of 40 PFLOPS.[1] It will offer public cloud services, AI-computing-power rentals and subscription-based cloud services, including platform-as-a-service (PaaS) and software-as-a-service (SaaS), for training large language models. This new data center is also remarkable for handing customers the freedom to choose packages that draw only on renewable energy.

The challenge

Building such a vast and high-performance infrastructure from scratch posed significant challenges. Ensuring the seamless operation and maintenance of this AI supercomputing center was no small feat. Any instability could result in frequent interruptions, driving up operating costs and affecting overall efficiency.

Our solution

ASUS is fully committed to supporting Ubilink in constructing a global top AI computing center from the ground up. The center will house 128 NVIDIA® H100 servers, equipped with 1,024 GPU cards, making it the largest green-energy-powered, AI-computing service center in Taiwan. Remarkably, this ambitious project was completed in just three months.

Infrastructure installation: ASUS expertly managed the construction of the supercomputing infrastructure, including the data center build-out, cabinet installation, rigorous testing, and onboarding. Each phase was meticulously executed to ensure optimal performance.

Cooling and power management: Managing thousands of GPUs is no easy task, with ASUS refined the liquid-cooling system to ensure the power consumption meet the industry-standard acceptance level.

Auto-detection and recovery tools: Given the complexity and potential failure scenarios in such a large-scale GPU cluster, traditional CPU clusters simply don’t compare. ASUS implemented advanced automated detection and recovery tools that minimize the impact of hardware failures, ensuring minimal downtime and maintaining system reliability.

Software platform and services: ASUS provides an integrated development platform and software stack that streamlines the rapid deployment and optimization of AI applications, making it easier than ever to harness the full power of the center’s capabilities.

Why ASUS?

The status of ASUS as exclusive hardware supplier ensures that we deliver the fastest, highest-quality equipment, backed by a professional technical team. This makes us the strongest pillar of support for Ubilink’s supercomputing ambitions.

ASUS AI servers, renowned for their unmatched computing power and stability, coupled with our integrated development platform, have garnered global acclaim. We empower customers to rapidly deploy and optimize AI applications, driving innovation and efficiency – ready to embrace the era of AI.

Recommended Model

ESC N8-E11 | ASUS Servers

[1] NVIDIA DGX results from the June 2023 summit, where the NVIDIA DGX H100 (128 units) achieved a performance of 40.66 PFLOPS, which can serve as a benchmark for this case.