GPGPU | 通用圖形處理器

GPGPU stands for “general-purpose computing on graphics processing units”, or “general-purpose graphics processing units” for short. The idea is to leverage the power of GPUs, which are conventionally used for generating computer graphics, to carry out tasks that were traditionally done by central processing units (CPU). The combined use of these two kinds of processors is sometimes referred to as heterogeneous computing; it is a key factor in a lot of technological breakthroughs, such as the development of artificial intelligence through machine learning and deep learning.

GPGPUs excel at parallel computing due to their large number of cores, which operate at lower frequencies than CPUs, but are more suited for data that's in graphical form. Groundbreaking scientific research can benefit greatly from GPGPUs, and many high performance computing (HPC) servers utilize a large number of GPGPUs to reach the level of supercomputing.

It goes without saying that if the data you work with is already in graphical form—as would be the case in projects that involve computer vision, like the development of autonomous vehicles or facial recognition systems—you would do well to have GPGPUs in your servers. Other possible applications include:

Earth science: Whether it's meteorological or astronomical research, the addition of GPGPUs can speed up the process considerably. For example, Japan's Waseda University used GIGABYTE's G221-Z30 to assemble a computing cluster that can run simulations on how climate change will affect coastal regions. Lowell Observatory in Arizona, USA uses GIGABYTE's G482-Z50 to filter out "stellar noise" in the search for habitable exoplanets.

Scientific knowledge: Academic institutes and research centers are always pushing the envelope of human understanding. GPU Servers outfitted with GPGPUs can accelerate the effort. CERN, the European Organization for Nuclear Research, uses GIGABYTE's G482-Z51 to analyze data produced by the Large Hadron Collider (LHC) in search of the elusive "beauty" quark. The Institute of Theoretical and Computational Chemistry at the University of Barcelona expanded the power of its data center by about 40% with GIGABYTE's G292-Z42 and other servers.

A better tomorrow: As the world's brightest minds pit advanced processing power against problems faced by humanity, servers equipped with GPGPUs have a clear role to play. IFISC, Spain's Institute for Cross-Disciplinary Physics and Complex Systems, uses GIGABYTE's G482-Z54 for important research that covers climate change, green energy, and ways of combating COVID-19. The National Taiwan Normal University has built a Center for Cloud Computing with G190-H44 and other servers. As long as the data can be converted to graphical form, GPGPUs are a natural fit in the server solution.

With regards to GPGPUs that are used for servers, GIGABYTE recommends the G-Series of GPU Servers. Benefiting from GIGABYTE's expertise in thermal and mechanical design, the G-Series leads the field with some of the highest density GPU servers on the market, capable of housing multiple GPGPUs and using them to perform complex calculations. For example, the G481-HA1 can house up to ten NVIDIA Tesla® V100 PCIe GPGPUs, while the G492-ID0 supports NVIDIA HGX™ A100 with up to eight SXM4 GPGPUs. These servers have been rigorously tested to ensure their performance and stability in high stress scenarios.

Learn more:
《How to Build Your Data Center with GIGABYTE? A Free Downloadable Tech Guide》