GPU Technology Conference: Heterogeneous technology to build cloud computing platform

The conference, hosted by NVIDIA (NVIDIA) in 2011, was opened by the Asian station of the GPU Technology Conference (GTC Asia), which focused on the hot "GPU computing", and this meeting invited top scientists in the field of High-performance computing to share the latest scientific research and commercial applications in GPU computing, In particular, GPU computing landing China's latest achievements. The following day, December 15, the reporter participated in a report by product director Dayrong on the theme "Building a cloud computing platform using heterogeneous technologies".

In this report, Dayrong focus on four points, the first is the heterogeneous computing and cloud computing introduction; second, the Dawning GPU technology: Server product line GPU monitoring, GPU scheduling, GPU portal, GPU Resource management introduction GPU optimization-Shuguang Nebula GPU application cooperation-physical and chemical biology, petroleum, communications, Power simulation, algorithm, and the third, GPU cloud computing solutions-China TOP100, dawning GPU case introduction.

The arrival of the "cloud era" has undoubtedly given the GPU unlimited opportunities, but also to the dawning company and other IT vendors to help the GPU a "cloud" opportunity. When introducing heterogeneous computing and cloud computing, Dayrong talked about the elastic cloud solution and demonstrated the elastic cloud topology. The elastic cloud solution includes the five modules of infrastructure, resource tier, delivery, security, management, etc. In the resource layer, elastic cloud utilizes resource pooling technology to construct resource pool, and the formation of resource pool is the basis of dynamic and elastic elastic cloud solution; The Cloudview management system realizes the organic integration of each module of the whole scheme, and it is also the core of enterprise cloud computing platform automation, and is the support system of enterprise-class cloud computing Platform Resource Service operation.

In introducing the dawning GPU technology, Dayrong focuses on the dawning "Nebula" high-performance computer system. The largest case of the dawning GPU is the "nebula", "Nebula" high-performance computer system with more than 5,000 nodes, supporting the Blade China's first petajoules super Computers Dawn "Nebula" high-performance Computing System, June 1 at the Beijing National Convention Center officially released, The tens of millions of billions of computing power once again refreshed the highest rate of Chinese high-performance computing. Shuguang "Nebula" High-performance computer system is the first real performance of the super million calculator, it has based on the independent research and development of "four high and low" technical highlights, namely high-performance, efficient, high reliability, high-density, low-power and low-cost six bright spots.

As the name suggests, the dawning "Nebula" High-performance computer system and the previous cluster is different, each of the clusters are very strong, so that together like a brilliant light ball. In addition, cloud computing is now a popular concept, and supercomputers are an important part of cloud computing, and the "nebula" will play an important role in the cloud computing hub after it is put into operation. This is the reason why the petajoules called the Nebula.

Dawn Godson After the limelight, will be exerting the GPU computing system. With its unique technical advantages, the GPU has quickly become the key technology to build the world's top supercomputers. In introducing the dawning GPU technology, Dayrong also mentions the videospeed. Videospeed is a high-performance video speed board, the use of 64 nuclear chips, can achieve high-speed video coding, decoding, transcoding, suitable for the need for a large number of fixed-point computing applications, network security, radio and television, video sites and so on.

In addition, with the growing popularity of multi-core processors, many of today's Intel commercial processors are dual-core and quad-core. Many authorities predict that future processors, most likely a multi-core processor, will have dozens of, hundreds of, or more processor cores integrated into the processor. Dawning has also provided a rich range of general-purpose processor solutions in the emerging development of general-purpose processors (/GPU).

(Responsible editor: Lu Guang)