ISC 2015 presented the concept of "big computing" for the first time and associated it with Big Data. The seminars about "big computing" were held on the last day of the conference.
What is "big computing"? Firstly, computing performance will enter the times for exascale, and the computing scale will embrace a greater era. Secondly, the trend of integration of high performance computing, Cloud computing and Big Data will be deepened, so that the boundary of computing will become vague, and the definition about computing will become more extensive. In some sense, "big computing" is perfectly in harmony with the idea of "Computing+" presented by Inspur.
Undoubtedly, artificial intelligence research involved in the Human Brain Project is the most typical field of "big computing". It integrates high-performance computing, Big Data and Cloud computing very well, and puts forward new requirements towards real-time computing capacity and power control. As is known to all, as the important technology of artificial intelligence, deep learning is one of the most popular IT technologies at present. The world top-level IT companies including Google, Alibaba, Baidu, etc. are dedicated to the development and application of such technology.
However, the problem upsetting numerous scientists is not just the compilation of intelligent software. What's more, excessive computed amount leads to the failure to simulate the speed of real-time information processing like the human brain does. Besides, compared to the human brain, there is a great difference in machine power consumption. Before then, scientists in Japan and Germany made use of "K Computer" supercomputer to complete the largest neural network simulation experiment, in which it took 40 minutes to process data that came from the human brain for 1 second. A greater difference still than processing speed is the power consumption of both. The total power consumption of "K Computer" is 9898.56kW, while the power for the conversion from human brain activity to electric power is just equivalent to the power of a 20W bulb.
It can be seen that computing speed and power consumption have become the problems to be solved in artificial intelligence research. In view of this, Inspur introduced SmartRack Co-processing Acceleration Scale Server focusing on deep learning and artificial intelligence research. This product is an innovative product with higher computing density and faster deployment speed. It can realize the deployment of four co-processing accelerators in 1U space, more reasonable distribution of computing resources, and more outstanding performance in parallel computing. It should be known that the strength of the human brain exactly lies in large-scale parallel processing. This also means that SmartRack Co-processing Acceleration Scale Serve can better suit various kinds of "brain projects" when compared to conventional servers.
Meanwhile, actual operational data from users is also a powerful proof of outstanding for the performance of SmartRack in energy consumption control, space optimization, and computing density. The actually measured data of the largest and the most complicated online ticket booking system 12306 indicate that energy consumption of SmartRack reduces by 15% compared to conventional products, and reduces by 6% compared to the conventional blade. Meanwhile, SmartRack takes into consideration the overall system design, breaks through the standard restriction of conventional rack server and cabinet, the height of node increases by 5%, the depth of cabinet increases by 19%; the design of the cabinet jambe is cancelled, the utilization rate of the overall space increases by 31%, and the deployment density can be improved by two times at most. The increase of inner space also provides more possibilities for node design. If SmartRack can provide a more flexible GPU node scheme, computing density can be improved by more than two times compared to the conventional blade.
As one of the first manufacturers engaged in R&D for the whole cabinet globally, Inspur took the lead in introducing the first co-processing Scale Server globally. At present, Inspur and Baidu cooperate to set the highest record of SmartRack in the deployment of 5000 nodes on a single day in the industry. Meanwhile, in-depth customized cooperation is performed in terms of whole cabinet server by natural air cooling featuring high temperature anti-corrosion and Big Data cold storage.
Undoubtedly, the advent of the "big computing" era will bring a reform of ideas to the existing data centre. Both the high-performance computing centre and data centre at enterprise level need to take into account IT infrastructure with fusion architecture to cater to the change of the times. This will result in the fact that a whole cabinet server with representative products for fusion architecture will become the favourite solution of the new era.
Inspur owns the most complete whole cabinet server product array globally at present, and has the capacity to provide node forms including co-processing acceleration, high temperature anti-corrosion, cold storage, high-density computing, high-performance storage, etc. This may take SmartRack from China to the world, and obtain approval from more global clients.