Making the future of high-performance computing happen now
Published 24 November 2020
High-performance computing (HPC) is becoming more commercially accessible for businesses looking to solve large problems using advanced technologies like artificial intelligence, machine learning, big data and video special effects.
High-performance computing has enabled animation houses, fintechs, cloud-based tech providers, aerospace and oil and gas industries to power up their operations, scale up their business and release applications to customers with speed, agility and affordability.
“Enterprises need to think about scalability. CPUs are getting faster and GPU performance is more powerful. They can easily add one GPU card or change a CPU to upgrade to a high-performance computing system,” said Andy Lin, the Business Development Manager at GIGABYTE.
But as high-performance computing becomes more commercially accessible, enterprises will encounter the challenge of choosing the right hardware specifications for their needs, including servers, Graphics Processing Units (GPUs) and Central Processing Units (CPUs).
“For some enterprises, this will be the first time using a HPC, so they have no ability in choosing the right hardware specifications. They just want to get a solution,” added Mr. Lin.
To build a high-performance computing architecture, servers are networked together into a cluster, and for it to operate at peak performance, every piece of hardware must keep pace with each other.
“Enterprises need to work with a strong system integrator like GIGABYTE to know how to choose the best HPC solution for their individual needs,” advised Mr. Lin.
Selecting the most suitable hardware for HPC relies on what enterprises want to do and the digital architectures they have available.
For example, GIGABYTE’s new G-series servers are designed for scientific computing, data analytics and increasingly popular artificial intelligence applications. The servers are designed to improve latency, increase bandwidth and expand the number of connected GPUs, including the world’s most powerful GPU, NVIDIA’s A100.
“So many applications will be using a little bit of AI in the future. For the G-series, we are more focused on looking at high density AI servers to provide customers with strong and stable server solutions,” said Mr. Lin.
While HPC is becoming more popular in Southeast Asia, enterprises in the region are said to choose the safe option of selecting a standard ‘brand server’ like Intel, Dell or HP.
“They prefer to use a brand server that is easier to maintain. But GIGABYTE provides a longer term solution for the hardware portion of HPCs. This is the strategy difference from our competitors,” added Mr. Lin.
GIGABYTE provides servers supporting up to 32 GPU cards that are fully compatible with other computers, systems and servers, enabling enterprises to simply choose a solution that works for their needs.
Keeping it animated even during a pandemic!
Thanks to the rapid advancements in computing power, beautifully designed computer graphics are constantly becoming more widespread, from movies and online games, to apps you find on your phone and even on PowerPoint presentations.
Most computer graphic animation relies heavily on HPC to speed up image processing and rendering. As the leading animation production company in Taiwan, Moonshine Animation needed a server that could handle the high-performance processing power needed for AI computing.
They used AI for facial simulation, which is very intensive for a GPU that doesn’t have enough memory to achieve results in one day or less.
In the end, Moonshine Animation chose GIGABYTE’s G191 Series servers, which supports up to four professional grade graphics cards, each holding up to 48GB of memory!
And with the COVID-19 pandemic forcing people to work from home and delaying projects, Moonshine Animation found an additional bonus to using GIGABYTE’s servers: remote desktop applications.
This allowed their animators to connect to GIGABYTE’s server on the Internet using an entry-level Mini PC or Thin Client computer.
And for any enterprise looking to easily migrate to a new server and test out HPC, GIGABYTE has a standard model that can support a number of GPU cards, which is suitable for customers who don’t know how HPC applications could benefit them, said Mr. Lin.
On top of this, GIGABYTE has explored the combination of CPUs and GPUs to improve latency of complex applications like graphical animations.
Making the HPC future cool
As the future of HPC rapidly heats up, so might the hardware in the data centers containing many of these servers.
“Core density will be very high. Based on this very high density design, managing the CPU heat is challenging,” said Mr. Lin.
If a data center runs too hot, then the hardware could fail, leading to disastrous and costly downtime. Currently, many data centers are just using air cooling and fans to solve this problem, but this might not be a viable solution in the future, as HPC systems will be running too hot.
“GIGABYTE already has a liquid cooling solution for enterprises and partners. That’s why we are leading in the HPC servers, not only in the hardware portion,” said Mr. Lin.
“Microsoft is trying to do an immersion cooling solution with their underwater data center. So in the future you will see more liquid cooling,” he added.
One common application for HPC is aerospace, and as more of the universe is unlocked with science, data centers will need to expand.
This exactly what the German Aerospace Center looked to do, but their challenge was running a data center at an ambient temperature of 40°C without air conditioning.
This is where GIGABYTE came in to provide a Direct Liquid Cooling solution that absorbed the heat emitted by the CPU, GPU, and memory while an enormous amount of space-related research data was processed.
For data centers like the ones used by the German Aerospace Center, it is also crucial to have safe, stable, efficient and affordable spaces and solutions.
“When data centers want to support HPC services, the important thing is to ask ‘how can I support this in limited space for the highest performance and lowest cost?’,” said Mr. Lin.
At this stage, GIGABYTE has announced new 2U 4-node solutions and dual sockets that can support up to 128 cores in one node, which reduces the amount of space needed per server.
“Data centers need to choose the more density enhanced solution and GIGABYTE could do that. The trend will be that data centers will do more to support HPC working models for enterprises. That’s why we look to become more and more popular in this market,” said Mr. Lin with confidence.
In the near future, GIGABYTE will look to work even closer with data center operators and their end customers to provide solutions that power the future of HPC, especially as we will see edge computing, 5G and AI make their presence known in our digitally advancing world.
By Stuart Crowley, Editor, W.Media