|
In this presentation, I give an introduction into microprocessor trends, describe two distinct eras of computing usage in training AI systems and show the wide variety of computing architectures in computer science. I also describe our advanced computing and artificial intelligence product portfolio, which focuses on innovation, continuous dedication and backward compatibility. The central part of this talk is insights into our Da Vinci architecture, descriptions of all building block, the core architecture and its micro-architectural configurations. Last, I show the process of how we execute artificial intelligence projects and the challenges which are still ahead of us.
0 Comments
Today we announced terrific new products. Our new flagship smartphone delivers an outstanding photography experience by using the worlds first multi-reflection super periscope telephoto, 10x long-range optical zoom, 100x max and an ultra vision Leica Penta camera system. We also incorporate "Hey Celia", a new virtual private assistant that supports you in your daily life. The HUAWEI Sound X speaker, with its refined and elegant design, delivers a visible stunning shaking bass and 360 degrees surround sound with six of full-range speakers. The HUAWEI Watch GT2 products appear in new designs and editions. Finally, our HUAWEI Eyewear combines leading technology with fashion. Step into new products and rewatch our live stream.
 Artificial intelligence will shape our future like no other technology. Since the role of this technology is expanding, it will optimise and improve what people do. In the medical field, artificial intelligence will help support diagnostic processes and other related processes. To bring artificial intelligence to clinical relevance, the technology industry, the medical industry and physicians are required to solve complex challenges in an interdisciplinary approach. This week I had the honour and pleasure to give a talk at the "Emerging Technologies in Medicine" conference, which brings experienced physicians, engineers and computer scientists together to talk about topics related to the future and exchange opinions for the challenges ahead. In my talk, I presented how Huawei supports the health care industry with artificial intelligence and described the process of how we execute AI projects. I described the challenges of microprocessor trends and introduced computer architectural approaches to solve these challenges, and showed Huawei's rich AI product portfolio. I also showed where AI, specifically with our products, has been successfully used in medical research such as in retinal blood vessel segmentation in the eyeground and the prediction of protein subcellular localisation. There are still challenges ahead, but with collaborative approaches such as this, I think we will be better equipped in facing our future. We opened the doors yesterday to a range of new connective possibilities and unveiled some exciting new products. We announced the world's slimmest 5G foldable smartphone, a business tabled with the highest screen-to-body ration, new amazing laptops and a variety of powerful communication devices. Last, we announced our app store, which is supported by HUAWEI Mobile Services. Re-watch our virtual launch here.  The 2025 Huawei trend forecast gives you the lowdown on data centre facilities five years from now. From 2010 to 2019, the data centre industry experienced exponential growth. The move towards hybrid IT continues as the balance between on-premise facilities, colocation and cloud services become more pronounced. The rapid development of new technologies such as artificial intelligence, cloud computing, big data, and 5G brings further growth and drive ever-increasing market demand. To meet this demand, data centres must solve construction challenges to build at speed, scale, manage energy usage and costs in a sustainable way. Besides, data centres also face many challenges in terms of architecture, flexibility and operations and management. The following trends look at how we meet this challenge as an industry and drive towards this future. 1) High density CPU performance and server capacity keep increasing with the evolution of IT computing capacity. As the demand for AI applications also increases, the importance of AI computing power grows in parallel. Data centres development targeting high density to balance efficiency and costs. Currently, the average power capacity in a data centre is 6 to 8kW/rack. The power density of 15 to 20kW/rack is going to be predominant in data centres by 2025.
2) Scalable architecture Generally, the lifecycle of IT devices is 3 to 5 years, and the power density doubles every 5 years. The lifecycle of data centre infrastructure, however, is 10 to 15 years. The data centre facility undergoes IT evolution for the next 2-3 generations. Scalable expansion and phased investment for optimal CAPEX for the lifecycle of the data centre is required. Also, the data centre must support the hybrid deployment of IT devices with different power densities because of the execution of diversified IT services. 3) Green Currently, the power consumption of data centres accounts for three per cent of the world’s total power consumption. The total power consumption of the data centre reaches more than 1,000TWh by 2025. Energy-saving, emission reduction and operating expense (OPEX) reduction are significant challenges. Increasing power usage effectiveness (PUE) of data centres and building sustainably is imperative and inevitable. It is an irresistible trend to use clean energy and waste heat to save resources (such as energy, land, water, and materials) throughout the lifecycle of the data centre. The average PUE of a new data centre in China is dropping to 1.1 in the next five years. 4) Quick deployment Internet services usually spike in a short period; data and traffic demands on the service side increase sharply. Therefore, data centres must deploy out quickly. On the other hand, the data centre is changing from a support system to a production system. A faster rollout, therefore, means faster benefits. The typical trailing twelvemonth (TTM) of a data centre is 9 to 12 months, which decreases to less than 6 months in the future.
5) Full digitisation and AI-enablement The software-defined, intelligent data centre is upon us. With the continuous improvement of IoT and AI technologies, data centres are gradually evolving from single-domain digitisation in terms of operation and management, energy-saving and operation, to full-lifecycle digitisation and automatic driving in terms of planning, construction, optimisation operation and manafement. AI is going to be widely adopted and applied. 6) Full modularisation More data centers will be constructed in full modular mode to address the problems of slow construction and high initial investment costs. Modular designs evolve from component modularisation to architecture modularisation; finally achieving full modularisation of the data centre. The fully modular design enables fast deployment, flexible capacity expansion, high energy efficiency and simple operation and management. 7) Simplified power supply architecture-lithium batteries becoming the norm The power supply and distribution system of a traditional data centre is complex and fragmented; it occupies an enormous footprint, and it is challenging to locate faults. A simplified power supply architecture reduces power conversion times, shorten the power supply distance and footprint, improve the space utilisation rate and enhance system energy efficiency. Compared with lead-acid batteries, lithium batteries have advantages in terms of footprint and service life. As the cost of lithium batteries decreases, lithium batteries seeing a widely used in data centres in the future.
8) The convergence of liquid cooling and air cooling; more indirect evaporative cooling and less chilled water cooling GPU and NPU applications generate an increasing demand in high-density scenarios, and liquid cooling systems are becoming more and more popular. Some storage and computing services however, are still in low-density scenarios. To quickly adapt to uncertain IT service requirements in the future, the cooling solution must be compatible with the air cooling system and liquid cooling system. Besides, the complex architecture of the chilled water cooling system hinders quick deployment— and smooth operation and management. An indirect evaporative cooling system, with a modular architecture, shortens the deployment time and simplifies operation and management. Also, by fully utilising the natural cooling resources, the power consumption of the cooling system is going to be significantly reduced. In areas with a suitable climate, the chilled water system is replaced gradually by an indirect evaporative cooling system. 9) Dynamic linkage between bits and watts Reducing PUE doesn’t mean that the overall energy consumption of the data centre is optimal. Instead of focusing on the data centre energy facilities, the energy consumption of the data centre needs to be evaluated and optimised as a whole. Through full-stack innovation among facility, IT, chipsets, data, and cloud, bits and watts are going to work collaboratively to achieve dynamic energy-saving and optimal energy efficiency of the entire system. 10) Trustworthiness As the data centre facility becomes more intelligent; the network security threats are getting more complicated. The data centre must have six features: resilience, security, privacy, safety, reliability, and availability to prevent attacks and threats from environments and malicious personnel, including network intrusion threats.
|
ArchivesCategories
All
|
RSS Feed