Li Zhengyi, general manager of IBM's Hardware Systems Division in Taiwan, said that under the wave of AI, many companies are actively thinking about the direction of transformation and IT deployment strategy. IBM has actively accelerated the development of upper-level applications from hardware and expanded cooperation with Qunhuan Technologies to promote PowerAI's deep learning Architecture and other cognitive solutions; IBM also continued in the mainframe, server and storage services injected AI kinetic energy, such as the introduction of POWER9 and related data flow storage and management.
IBM recently released a POWER9 processor designed for AI workloads, helping to speed up the construction of AI applications.Wu Zhifeng, senior information technology consultant IBM hardware systems department pointed out that POWER9 processor with two IBM's unique technology, respectively, to support NVIDIA NVLINK for dramatically increasing data exchange between the GPU and the CPU, and built-in Distributed Deep Learning (DDL) software in the POWER9 processor to enhance deep learning performance.
Wu Zhifeng further explained that deep learning training will use the data resources of the GPU and the CPU. In order to make the data transmission speed between the GPU and the CPU also be the same as that of the GPU and the GPU, IBM will cooperate with NVIDIA to develop the NVLINK technology, Improve memory system data movement between GPU and CPU.
As DDL software is used to enhance the effectiveness of large-scale deep learning model, and can shorten the time to train a deep learning model.In the past to deploy large-scale deep learning often takes days or even weeks to train a deep learning model, The development of DDL IBM, dozens of servers can be optimized, a total of hundreds of GPU parallel computing, increase processing speed, reduce the depth of learning model training time.
In addition to DDL, IBM also has a software development kit for IBM Deep Learning Impact that provides complete life-cycle management as it develops and optimizes neural networks; it also provides 11 industry-related models that include hyper-parametric Excellent, can improve the accuracy of 5 to 10%, and parallel training, is now 85% speedup can be done to help users quickly converge to make a good model.
Li Zhengyi said that IBM has been able to provide AI kinetic energy that meets specific industry needs and accelerate corporate awareness by bringing industrial scenarios, data and professional experience into cognitive solutions, linking IBM's knowledge-based AI deep learning platform with Know-how across industries Transformation, promote the development of Taiwan AI ecosystem.