HPC | 高效能運算
What is it?
High Performance Computing generally refers to the ability to process data and perform calculations at high speed, by aggregating computing power – combining the power of individual computers (referred to as "nodes"), into a large group, or "cluster". The nodes of a HPC cluster could be made of commodity parts (such as an enterprise grade CPU, memory, flash drives etc.) similar or even the same as a typical desktop or workstation, or they could be highly customized.
The combined computing power of a cluster can deliver much higher performance than a single computer, as individual nodes work together ("talking" to each other via a shared interconnect ) to solve a problem larger than any one computer can easily solve.
Why you need it?
A HPC computing cluster is typically used to solve large problems in science, engineering or business. Examples include:
Science: HPC is used by scientists at universities and research institutions to understand the evolution of our universe, conduct research into new medicines, or simulate and predict climate and weather patterns.
Media & Entertainment: HPC is used to render animation and special effects, edit feature films, or live stream special events globally.
Artificial Intelligence: A particular subset of HPC which is currently very popular is machine learning, which is being used to develop a myriad of artificial intelligent applications, such as self-driving vehicles, facial recognition software, speech recognition and translation, or drone technology.
Oil and gas: HPC is used to process and simulate information such as satellite images, ocean floor sonar readings etc. in order to identify potential new deposits of oil or gas.
Financial services: HPC is used to track stock trends and automate trading, or analyze patterns to detect fraudulent activity.
Medicine: HPC is used to help develop cures for diseases like diabetes, or to enable faster and more accurate diagnosis such as cancer screening techniques.
How is GIGABYTE helpful?
GIGABYTE's H-Series (2U 4 Node Server) and G-Series (GPU Server) systems are designed especially to be used in HPC clusters, combining a dense amount of compute and / or GPU compute power into a 1U, 2U or 4U server chassis, which can be linked together via interconnects such as Ethernet, Infiniband or Omni-Path. A particular example of a server platform ideal to be used for HPC is our recently released H262 Series AMD EPYC 7002 Series "Rome" 2U 4 Node server, which can feature up to 512 cores / 1024 threads of compute power (two 64 core AMD EPYC CPUs per node, four nodes per system) into a single 2U chassis. By populating a full 42U server rack of these systems (leaving some room for networking switches), a customer will be able to utilize up to 10,240 cores and 20,480 threads – a massive amount of computing power.