For veterans of the industry who are versed in server technologies, there is sometimes a tendency to get bogged down in specs and benchmark figures: these CPUs contain an x number of cores and threads more than the previous generation, this interconnect architecture transfers data at a rate of so-and-so many gigabytes per second, etc. These figures are important, but only because they serve a greater purpose: the advancement of human society through high tech solutions and superb processing power. As crucial as it is to develop the most advanced server products, it is also vital to share these resources with talented individuals, so that the best and brightest among us may have access to tools that match their potential. 

Organized by the HPC-AI Advisory Council (HPCAIAC) and National Supercomputing Centre (NSCC) Singapore, and sponsored by such industry heavyweights as Nvidia Corporation, the APAC HPC-AI Competition is an annual contest designed to bridge the gap between the cultivation of talent in universities and research centers, and the application of AI and HPC in the real world. By pitting the boundless potential of young and eager minds against supercomputing tasks with real-life implications, the contest aims to help students prepare for the ever-growing demand for higher computation performance, as well as the increasing complexity of research problems.

“High performance computing and artificial Intelligence are the most essential tools fueling the advancement of science,” says Gilad Shainer, Chairman of the HPCAIAC. “[Our] mission is to foster the next generation of supercomputing leadership, and to help develop the next generation of platforms and knowledge.”

Cheng Kung University has participated in the APAC HPC-AI Competition since its conception. In 2020, the Department of Engineering Science at NCKU assembled two teams of around six students each to participate in the contest. The students were mentored by Chair Professor Chi-Chuan Hwang; they were also coached by Research Assistant Chao-Chin Li, whom the students called Michael.

As part of the contest in 2020, the HPCAIAC and NSCC assigned four challenging tasks. The first task was centered on AI applications: contestants were asked to surpass international NLP (natural language processing) records by using BERT, a machine learning technique developed by Google. The second task was focused on HPC: participants must attempt to break climate simulation world records by using NEMO, a modelling framework for research activities and forecasting services in ocean and climate sciences. The third and fourth tasks were geared towards bio-science simulations and innovations as part of the global effort to combat COVID-19. Students were asked to use NAMD, a parallel molecular dynamics code, to recreate the molecular structure of a common virus in the shortest time possible. They were also tasked with proposing an HPC or AI application that could potentially be used against COVID-19.

Learn More:
《Glossary: What is Natural Language Processing?》
《Glossary: What is Machine Learning?》
《Spain’s IFISC Tackles COVID-19 and Climate Change with GIGABYTE Servers》

Taking into account GIGABYTE’s dedication to CSR and ESG objectives, and GIGABYTE’s position as an industry-leading server solutions brand, it should come as no surprise that GIGABYTE was glad to support the NCKU’s supercomputing team. Four G482-Z50 GPU Servers were provided by GIGABYTE. With these top-tier server products, students were able to practice using the most advanced supercomputing techniques to tackle some of the world’s most critical issues.

“GIGABYTE provided us with server products on par with any server brand you could care to name, because they believed in our effort to nurture the next generation of HPC and AI experts,” says Michael Li, who worked closely with the students and developed a fraternal bond with many of them. “Winning awards in an international competition is about more than just the university’s prestige or our students’ academic careers. It is about cultivating supercomputing geniuses who can give back to society and contribute to the betterment of all humankind.”

The servers were delivered to the NCKU’s campus in Tainan, and set up in a lab. The supercomputing team connected them to each other to form a computing cluster—an interconnected network of servers utilizing distributed computing technology to deliver performance on the scale of supercomputers. The software tools designated by the HPCAIAC and NSCC were installed on the cluster. Then, under the patient tutorage of Professor Hwang and Coach Li, these plucky university students got to work trying to shatter world records.

Learn More:
《Glossary: What is Computing Cluster?》
《Glossary: What is Distributed Computing?》
《Tech Guide: What is Cluster Computing and How Can GIGABYTE Help?》

The students spent every moment of free time they had outside of class, including evenings, weekends, and holidays, to practice running the supercomputing programs on the GIGABYTE servers. First, the students would pore over theoretical computer science studies to look for the latest hypothetical methods that may help them achieve a breakthrough. They would then sound out the professor and the coach on their proposal. If they seemed to be on the right track, they would begin tweaking parameters in the software and running tests on the servers to see if their theory panned out. Sometimes they would achieve a quantum leap forward, and Coach Li would take them out for a night on the town to celebrate. Sometimes an experiment would end in abject failure, and it would be back to the drawing board.

In the weeks and months leading up to the big day of the competition, the students achieved breakthroughs that would not only propel them past existing records, but also outmatch all the other top universities competing in the contest. Throughout the entire grueling process, the GIGABYTE servers were the students’ steadfast, unwavering companions, always ready to help test the latest configuration of software parameters. Long before the students emerged as world champions in the 2020 APAC HPC-AI Competition, they achieved their greatest triumphs on the four GIGABYTE G482-Z50 servers in the computer lab on the NCKU campus.

The supercomputing team summarized that the GIGABYTE servers offered them three key benefits, which ultimately helped the students break world records and take home the championship:

1. Incredible parallel computing performance, delivered through a combination of advanced CPUs and GPUs.
2. Seamless connectivity between the servers, thanks to communications standards that offer high throughput and very low latency.
3. Unrivalled reliability thanks to high availability features, which allowed the servers to operate continuously during the months leading up to the contest.

Glossary:
《What is Parallel Computing?》
《What is GPU?》
《What is High Availability?》

In the quest for world-first scientific breakthroughs, using the right tools for the job can make all the difference. Some computing workloads are better suited for the complex, polymathic capabilities of the central processing unit; others run much more quickly and efficiently on general purpose graphics processing units (GPGPUs). The organizers of the APAC HPC-AI Competition realized this, and so they stipulated that certain problems must be solved with certain types of processing units. The ideal server solution, then, would need to support both advanced CPUs and GPGPU accelerators; it should also be able to run the same task on multiple processors to achieve parallel computing.《Glossary: What is GPGPU?》

GIGABYTE servers are designed to do just that. They support the most advanced CPU options, such as Intel® Xeon® Scalable, AMD EPYC™, and the Ampere® Altra® series, which is based on the ARM architecture. In the case of the G482-Z50, there is also support for a highly dense configuration of up to ten PCIe Gen 3.0 GPGPU cards. Each of the server’s dual CPUs can be connected to five GPGPU cards through a PCIe switch, which minimizes the communication latency between the GPGPUs. The supercomputing team at NCKU outfitted their GIGABYTE servers with NVIDIA accelerators. The G482-Z50 is also fully compatible with other options, such as the AMD Instinct™ MI100.

Learn More:
《GIGABYTE’s Complete List of Intel® Xeon® Scalable Servers》
《GIGABYTE’s Complete List of AMD EPYC™ Servers》
《GIGABYTE’s Complete List of Ampere® Altra® Servers》
《Glossary: What is PCIe?》

As is common in HPC applications, the four servers were connected to each other to form a cluster. To link the servers, the NCKU team used 100 Gbps InfiniBand (IB) switches, which are noted for their high throughput and low latency qualities. GIGABYTE servers are also compatible with other networking standard, such as Ethernet, which supports UTP and fiber optic cables.

The students used OpenZFS for file storage and Open MPI for communication. Since both of them are open-source, the NCKU team installed the Linux operating system on the GIGABYTE servers. OpenZFS functions like an advanced version of Network-Attached Storage (NAS), in that the files are distributed among all the servers in the cluster, rather than on a single server. This drastically improves the read/write speeds when storing or retrieving data. Open MPI was chosen because it allowed the students to access the entire cluster through any one of the four servers; it automatically distributed tasks to available computing resources to achieve optimal performance. The G482-Z50 is compatible with Red Hat Enterprise Linux, SUSE Linux Enterprise Server, and a wide range of other operating systems.

Learn More:
《Glossary: What is NAS?》
《Setting the Record Straight: What is HPC? A Tech Guide by GIGABYTE》

Throughout the many months before the competition, the team worked tirelessly to crack the supercomputing puzzles. They ran batches of tests on the GIGABYTE servers, working days, nights, weekdays, weekends—which meant the servers got no rest at all. Thankfully, this did not constitute a problem, because GIGABYTE’s server solutions come with a treasure trove of high availability (HA) features.

First and foremost is Smart Crises Management and Protection (SCMP), a GIGABYTE-patented feature that is deployed in servers without a fully redundant power supply unit (PSU) design. In the event of a faulty PSU or overheating, SCMP forces the CPU to enter ultra-low power mode. This reduces the power load, which protects the system from unexpected shutdowns, while avoiding component damage or data loss.

Smart Ride Through (SmaRT) is another feature that is implemented on all GIGABYTE servers. In the event of a power outage, the system will manage its power consumption (known as throttling) while reducing the power load. Capacitors within the PSU can provide power for ten to twenty milliseconds, which is enough time for a backup power source to come online.

Last but not least, the dual ROM architecture guarantees that, in the unfortunate event that the ROM storing the BIOS and BMC fails to boot, the system will reboot with the backup BIOS and/or BMC. Once the primary BMC is updated, the ROM of the backup version will automatically synchronize. The BIOS can be updated based on the firmware version.

On the big day of the competition—October 15th, 2020—everything that the students learned using the GIGABYTE servers was put to the test. In the end, hard work paid off. NCKU came in first, beating other prestigious research universities in the Asia-Pacific region, such as the University of New South Wales (UNSW) in Australia, Nanjing University (NJU) in China, and Singapore’s own Nanyang Technological University (NTU). Some of the NCKU team’s results even shattered world records at the time. For example, in the first part of the competition, which asked contestants to break NLP records using BERT, NCKU achieved an accuracy of 87.7%. This was higher than what had been achieved by the University of California, San Diego (87.2%) and Stanford University (87.16%).

“Our collective futures greatly depend on nurturing every student’s potential, especially in times of adversity such as these. The commitment and resilience demonstrated by this year’s competing teams reminds us all that we need to meet challenges head-on and be flexible in adapting to the new normal,” says Associate Professor Tan Tin Wee, Chief Executive at NSCC.

NCKU would go on to win big again in the 2021 APAC HPC-AI Competition. Professor Hwang says, “We would like to thank the College of Engineering and the Department of Engineering Science at NCKU, as well as GIGABYTE for their support.”

GIGABYTE Technology is glad to have provided the servers that helped to educate the next generation of HPC and AI experts in Taiwan. These world-class servers have not only proven capable of tackling problems related to some of the 21st century’s most pressing issues—namely artificial intelligence, climate change, and the COVID-19 pandemic—they have also shown that as long as we are willing to invest in the younger generation, humanity has a chance to triumph against adversity. This is in keeping with GIGABYTE Technology’s unwavering commitment to its CSR and ESG goals, as well as the GIGABYTE motto: “Upgrade Your Life”, which is a sincere belief that high tech solutions can help us build a better world.

Learn More: