'Opportunity' Huang Renxun: GPU accelerated computing becomes the main mode of extending Moore's Law

1. Apple Samsung could not save the smartphone processor market shrink 2. NVIDIA Huang Renxun: GPU accelerated computing has become the main mode of extending Moore's Law 3. Qualcomm's first XR platform release, comparable to the Xiaolong 845 VR/AR experience? 4. MediaTek MediaTek Sensio upgrade, MT6381 will usher in applications? 5. NVIDIA releases HGX-2 AI computing performance to a higher level 6. Aiming for automatic driving! Intel and Tsinghua University, Chinese Academy of Sciences launched a five-year cooperation

1. Apple Samsung could not save the shrinking smartphone market

Gathering micro-network news, market research organization Strategy Analytics recently released a statistical report, the main content for the 2017 global smart phone processor shipments, revenue and distribution.

Smartphone processor market shrinks

Strategy Analytics report shows that in 2017, the global smartphone processor market shrank by 5%, and total sales fell to 20.2 billion US dollars.

In fact, this is not an unimaginable thing. Since the beginning of last year, global smart phone shipments have been shrinking to a certain extent due to various factors such as the gradual maturity of major markets and slow development of emerging markets. The volume of shipments has continued to decline. Even if it is as strong as Apple, Samsung is unable to handle it.

As the core component of the smart phone, the shipment of the processor and the smart phone are accompanied by each other, and the decline is inevitable.

Judging from the specific processor brands, Qualcomm, Apple, MediaTek, Samsung and HiSile are ranked in the top five of 2017 global leading smartphone processor companies.

The top five vendors are not the same

Although the five major companies have not changed, but from the perspective of their respective companies, the situation is very different.

Due to the lack of development focus of MediaTek's smartphone processor in 2017, Qualcomm’s market share has been further improved to 42%.

The second Apple market share reached 22%.

MediaTek ranks third, with a 15% market share in the smartphone processor market.

The Strategy Analytics report points out that in the past year, Apple, Hass, Qualcomm and Samsung’s processors have all experienced an increase. However, Mediatek and Zengrui’s processor shipments have declined to a certain extent.

MediaTek is also due to a sharp decline in its share and shipments in the Chinese market, and for the first time in five years, its revenue declined. Qualcomm got orders from Chinese manufacturers such as Oppo, Vivo, and Xiaomi.

The specific reason is that in the past five years, the demand and development of 3G smart phone processors have been very slow, and the market has gradually transformed into the mainstream market of 4G smart phones. MediaTek and Zengrui, which were mainly played by 3G, were affected by the market. Shares naturally decline gradually.

Sravan Kundojjala, an analyst at Strategy Analytics, said: 'In 2018, it will be a crucial year for MediaTek and Zeng Rui.'

AI and 64-bit trend

In addition, the Strategy Analytics report also pointed out that there are currently more than 250 million smartphones that have begun adopting AI engines to implement machine learning and support new applications such as 3D face recognition and image recognition.

In the future, we believe that with the increasing popularity of artificial intelligence applications in smartphones, smart phone processors equipped with related technologies will become mainstream choices in the market.

On the other hand, Strategy Analytics also pointed out that after years of development, 64-bit smart phone processors have seen a significant increase in the last three years. The market share of 64-bit smart phone processors reached 88% in 2017, and this figure is In 2016 it was only 38%.

In addition, Strategy Analytics also stated that 10nm smartphone processors accounted for 14% of total processor shipments driven by Apple, Hass, Qualcomm and Samsung, while 8-core processors accounted for total shipments. 40%.

At present, the global mobile phone market is in a weak state, which is related to the mobile phone's differential reduction and lack of innovation. For example, Apple's decade-old iPhone X has not gained any benefit in the market. With the artificial intelligence and 64-bit popularity in the future, What happens to smart phone processors? (Proofreading/Fan Rong)

2.NVIDIA Huang Renxun: GPU accelerated computing becomes the main mode of extending Moore's Law

In this GTC Taiwan, NVIDIA CEO Huang Renxun will increase the demand for computing by 100 times every year in the next 10 years. At the same time, it is expected that under the gradual decay of Moore's Law, the GPU computing volume of the world's top 50 supercomputers will be in the future. It will grow 15 times in 5 years. At the same time, GPU-accelerated computing will become the main mode of extending Moore's Law.

In the GTC Taiwan, Huang Renxun once again emphasized that NVIDIA's CUDA computing model has accelerated the benefits brought by the past. It also shows that the model for GPU-accelerated computing will continue to expand in the future. It is expected that the global computing demand in 2028 will be equivalent to that of 10 million Volta-based GPUs. , If the traditional supercomputer-level computing capability is formed by multiple CPU stacks, it will occupy large-scale space and high power consumption. If the GPU is replaced, it will save more space and power consumption, and at the same time bring about higher acceleration. effect.

At present, Supercomputer Fundamentals has become an important tool for the development of modern science, playing an important role in molecular construction, quantum chemistry, quantum mechanics, weather forecasting, meteorological research, energy exploration, physical simulation, data analysis and artificial intelligence technology development, and providing 100 Tera- or billion-scale operational performance. According to OpenAI statistics, the artificial intelligence computing model in the next five years will grow 300,000 times, which is expected to grow 30,000 times faster than Moore's Law. It will allow GPU acceleration to allow Data, computational complexity is greatly increased, to solve the past manpower can not solve the computing needs.

Last year, it announced the launch of a Volta-based GPU that integrates the Tensor Core design with 32GB of HBM2 memory. This corresponds to 125 Tensor TFLOPS computing performance, corresponding to 7.5 FP64 TFLOPS or 15 FP32 TFLOPS budget performance, and 10% faster than the previous GPU-accelerated operation mode. Multiple efficiency, while further reducing the occupied space and power loss.

In order to break through the limitations of hardware architecture, NVIDIA further announced the launch of NVSwitch in this year's GTC 2018, enabling 16 sets of Volta GPUs to share up to 512GB of HBM2 memory (32GB x 16), a total of 81920 CUDA cores, and 2000 Tensor Core TFLOPS operations. Performance, constitute the world's highest performance GPU, and is not affected by the traditional CPU architecture to limit the GPU access memory capacity. With the design of NVSwitch, NVIDIA has announced the launch of the world's largest (and game-playable) DGX-2 GPU. Up to 2PFLOPS computing performance, and special porous fiber design allows the operating power up to 10000W of the box to maintain low-temperature operation, compared to the DGX-1 officially launched six months ago, the computing performance increased by 10 times.

Compared to the past, 300 pairs of dual-core CPUs must be used. The server must consume 180,000 watts of power consumption. The single DGX-2 GPU can handle the same computing performance, but the replacement price is only 1/8 and 1/18. Power consumption, at the same time compared to Alex Alex Krizhevsky, who used two NVIDIA GTX 580 GPUs, spent 6 days to complete training AlexNet, by DGX-2 GPU in just 18 minutes. At the same time DGX-2 GPU also broke Analyze 1075 images per second, become the fastest single-chip computing speed, and can process 15500 images per second at each node, and can complete the expansion in 14 minutes. The inference delay time is only 1.1 milliseconds, which is more inferable per second. Calculate 6250 images.

Through DGX-2's computing power and NVSwitch inline technology, NVIDIA also announced the launch of the HGX-2 server platform built with DGX-2, and with Quanta, Yunda, Foxconn, Inventec, Wistron, Wei Ying, ASUS, Gigabyte, ASRock, Taian, Acer and other local Taiwanese manufacturers cooperated, while emphasizing that about 90% of the world's servers originate in Taiwan, and NVIDIA is also continuing to cooperate with more local manufacturers in Taiwan.

With GPU computing capabilities, in conjunction with image processing technology collaborating with software vendors such as Adobe, it will be possible to instantly modify unnecessary objects in the image, or to reconstruct the lack of content in the image, and even to further present the 'beauty' effect. The cooperation of the kubernetes container cluster management system proposed by Google will enable more artificial intelligence systems to dynamically adjust computing performance in response to different computing needs, thereby allowing GPU speed computing performance to have more flexible configuration benefits. Will join Alibaba, Baidu, and eBay , HIKVISION, IBM, Xiaomi and other manufacturers cooperation.

In cooperation with Taiwan, NVIDIA stated that Foxconn will use artificial intelligence technology to detect production efficiency. The Chinese University of Medicine’s attached hospital assists physicians in analyzing and predicting cancer metastasis through the use of artificial intelligence technology. Taiwan University differentiates nasopharyngeal cancer through artificial intelligence. Organs, and Taiwan’s artificial intelligence laboratories have also assisted Tainan City Government in monitoring bridge structures to prevent typhoon damage through artificial intelligence. The Taoyuan City Government plans to allow 30% of fixed-route buses to be equipped with Level 3 autopilot by 2020.

As at the end of the keynote speech in the GTC 2018 with 'PLASTER', Jen-Hsun Huang also emphasized programability, low latency, high accuracy, size, and data throughput, respectively. Throughput), Energy Efficiency, which in turn promotes the rate of learning (Rate of Learning), allowing artificial intelligence to grow faster. Economic Daily

3.Qualcomm's first XR platform released, comparable to the Xiaolong 845 VR/AR experience?

Gathering micro-messages (text/Xiaobei) Qualcomm released the first XR platform on the Xilong XR1 platform during the Augmented Reality Expo (AWE) yesterday.

The XR is a new concept proposed by Qualcomm in 2017. The technology covers AR (Augmented Reality), VR (Virtual Reality) and MR (Mixed Reality). Qualcomm once predicted that in 2021 XR will create a market of 1080 billion US dollars.

It is reported that Qualcomm's preliminary strategy for the development of XR technology is: Chip-level: Qualcomm's top-level mobile chips will support the application of mobile platforms built on XR; Software-level: Provides dedicated VR SDK for developers; HMD Accelerator Program: HMD Reference Design Help OEMs to commercialize products as quickly as possible; Eco: Work with partners to build ecosystems.

Prior to the launch of the XR1, Qualcomm had launched more than 20 XR devices with its partners, covering all-in-one machines and XR-compatible smartphones. Once, the XR was integrated as a technology into Qualcomm Snapdragon high-end processor chips, such as the XR technology. Dragon 845 promoted a major selling point, namely the ability to locate in the indoor space, six degrees of freedom and instant positioning and map building system, can produce a more realistic immersive experience.

It is reported that before the XR concept came forward, Qualcomm Snapdragon processor has begun to integrate this technology, Snapdragon 820 is the first generation XR hardware platform, Snapdragon 835 is the second generation XR hardware platform, Snapdragon 845 is the third generation XR hardware platform.

XR1 as a dedicated extended reality platform integrates Qualcomm Technologies heterogeneous computing architecture, including ARM-based multi-core CPU (Kryo), vector processor, graphics processor (Adreno) and Qualcomm artificial intelligence engine AI Engine.

XR1 also includes XR software service layer, machine learning, Xiaolong XR Software Development Kit (SDK), and Qualcomm Technologies' connection and security technology.

It is reported that Xiaolong XR1 has three grades, respectively, for the entry cardboard, 3DoF (3 degrees of freedom) mainstream products and 6DoF (6 degrees of freedom) flagship product.

According to informed sources, the XR1 platform simplifies the design of a large number of handsets optimized for mobile phones in the previous Opteron processors, reducing baseband integration, resulting in a significant reduction in costs, while focusing on AI and security. Previously, Vive Focus and other integrated devices use the Snapdragon mobile processor chip. Nowadays, XR-specific platforms can be used, which can achieve greater cost advantages.

Can guess, future high-pass XR technology development path at the chip level will be divided into carrying XR technology high-end Opteron processor and XR chip.

Qualcomm expects that the first VR-equipped ones equipped with XR1 chips will be available in the early 2019s, and predicts that the VR/AR all-in-one will reach 186 million units by 2023. Qualcomm revealed that Meta, Vive, Vuzix and Pico etc. Manufacturers have been based on the first dedicated XR1 platform for development. According to foreign media reports, the chip will be the first to apply in the new Vuzix Blade smart glasses and HTC's AR / VR head display.

XR1 features in visual, audio and interactive: up to QHD+ resolution display, 6 head degrees of freedom + 6 hand degrees of freedom, 4K 60FPS video playback, Qualcomm Aqstic/aptxHD audio technology, AR capture latency Within 20ms, etc. (Proofreading / Maocao)

4. MediaTek Sensio MediaTek upgrade, MT6381 will usher in the application?

Gathering micro-messages (Wen/Xiaobei) In December last year, MediaTek released the smart health solution MediaTek Sensio. The mobile phone or mobile phone accessories that carry this program can measure six physiological data including electrocardiogram in about 60 seconds. The core of MediaTek Sensio is a smart health chip MT6381.

A MT6381 can measure the user's heart rate, heart rate variability, blood pressure trend, blood oxygen saturation (SpO2), electrocardiogram (ECG), light volume pulse wave (PPG) 6 physiological data, so it is called six-in-one smart health The chip can be integrated into smart phones, wearable devices and other terminals. It is reported that this chip is by far the most complete intelligent health program.

According to MediaTek’s production plan, the MediaTek Sensio Smart Health Solution will be available early this year. However, today, no cell phone manufacturer has explicitly announced that it will carry the chip.

Does MT6381 really have no market prospects?

According to news recently, MediaTek is working with Google and Indian mobile phone manufacturers to import smart health chips into Android Go platform smartphones, and it is expected to begin volume production this year.

MediaTek said that it has added the algorithm for predicting blood pressure trends to MediaTek Sensio Smart Health Solutions. With MediaTek Sensio's development, MediaTek Sensio and MT6381 are confident in the market prospects of MediaTek Sensio.

It is reported that Taiwan University, National Taiwan University Hospital and MediaTek's research team stated that they will continue to deepen cooperation and optimize and verify algorithms for simultaneous clinical research. With the advancement and upgrade of MediaTek Sensio, in the future, manufacturers are expected to rely on MediaTek Sensio. Development of medication reminders, blood pressure warnings, emergency calls and other applications for smart devices and accessories.

MediaTek, Google ties closely, or becomes MediaTek Sensio landing key

Since the end of 2017, MediaTek has been working more closely with Google. In November 2017, MediaTek announced that it officially entered Google’s GMS Express project and became a first-party SoC member. The software update approval time has been reduced to four weeks; December 2017, MediaTek announced that it has become Google's latest Android Oreo (Go version) partner and will work closely with Google to solidify the entry-level smartphone market.

Android Go can be seen as a compact version of Android 8.1, mainly composed of lightweight Android, Google Go series comes with applications and Google Play Store designed for low-end devices, designed for 512MB or 1GB memory mobile phone.

Android Go is targeted at the Indian market through a single detail: Gboard made an update in April 2017, adding support for six native Indian languages, which has increased Gboard's support for native languages ​​in India to 22 types. Although Google never mentioned that Android Go is aimed at the Indian market, according to the global configuration of smart machines, it can be speculated that a major target market for Google Android Go is the Indian market.

It is reported that MediaTek MT6739, MT6737 for 4G mobile phones, MT6580 for 3G mobile phones and other chips have fully supported Android Android Go.

Given MediaTek's close cooperation with Google on the Android Go smartphone, MediaTek Sensio's 'homeopath' import into the Android Go smartphone is also reasonable. And for emerging markets such as India, smart health will become a major selling point for mobile phones. The introduction of MT6381 will not bring much cost increase. After all, compared with the full screen, the MT6381 is very low cost.

Therefore, Google may become the key to MediaTek Sensio landing.

5.NVIDIA releases HGX-2 AI computing performance to a higher level

In order to accelerate the development of artificial intelligence (AI) and respond to increasingly heavy computational demands in the future, NVIDIA announced at the "GTC Taiwan 2018" conference a single integrated computing platform built specifically for AI and high-performance computing - NVIDIA HGX-2. The product improves AI computing performance through 2 Petaflops computing power and has been adopted by many computer/server manufacturers such as Asustek, Acer, Yunda, Lenovo, and others.

NVIDIA founder and CEO Huang Renxun said in a keynote speech that as the computing demand increases rapidly, the method of expanding the CPU has become outdated, and the server complexity of building high-performance computing (HPC) and AI keeps rising, and it has almost reached the system. The limit of design; NVIDIA HGX-2 combined with GPU equipped with Tensor Core, through the integration of high-performance computing and AI, can provide a powerful multi-purpose computing platform in the industry to overcome the above challenges.

It is reported that the HGX-2 cloud server platform is capable of multiple precision operations and provides strong flexibility to support future computing needs. It also uses FP64 and FP32 for high-precision operations for scientific and analog computing, and can also use FP16. The Int8 format performs AI training and inference, in response to more and more high-performance computing combined with application requirements derived from AI.

In addition, the product can achieve an AI training speed of 15,500 frames per second through the ResNet-50. The performance is more than sufficient to replace the server with up to 300 CPUs. It also has an NVIDIA NVSwitchTM mesh interconnect architecture that enables smooth chaining of 16 units. The NVIDIA Tesla V100 GPU with built-in Tensor Core functions as a huge GPU, and the system built with HGX-2 is the NVIDIA DGX-2 introduced recently.

Huang Renxun pointed out that HGX-2 is part of the NVIDIA GPU acceleration server platform series. This series of products is connected to the entire data center server system and is suitable for each large market. It can recommend the most suitable GPU for different AI, high-performance computing and speeding up operations. Combined with CPU configuration. Such as HGX-T for hyperscale training and HPC; HGX-1 for large-scale inference and intelligent image analysis (IVA); and SCX-E for data center, HPC, IVA, virtual desktop infrastructure ( VDI) etc. New Electronics

6. Aiming for Automatic Driving! Intel and Tsinghua University, Chinese Academy of Sciences Launches a 5-year Cooperation

Micronet Collection May 30 Report

This afternoon, Intel announced in Beijing that it will set up the Intel Intelligent Network University Automotive University Collaborative Research Center to conduct in-depth research on autonomous driving with China's top universities and research institutes. The first universities to sign cooperation agreements are Tsinghua University, Chinese Academy of Sciences. Institute of Automation, the cooperation for a period of 5 years.

It is understood that the Intel Intelligent Network University Automotive University Cooperative Research Center will aim at five major topics, covering self-driving cars, car networking technologies, and large-scale intelligent infrastructure supporting autonomous driving; targeting China and the unique culture and geographical features of the Asia Pacific region, Respond to the unique challenges that auto-driving cars face when they are widely used and deployed.

Specific research directions include: 1) Auto-driving car safety; 2) Autopilot man-machine interface, and challenges brought by new supporting laws and regulations; 3) Open data sets and benchmarks; 4) Car networking based on cellular wireless networks (C-V2X), Mobile Edge Computing, Smart Transportation, and Smart Infrastructure Research; 5) Advanced Algorithm Research Based on New Hardware Devices and Architectures (eg Intel 3D Xpoint).

The parties involved with Tsinghua University and the Institute of Automation of the Chinese Academy of Sciences will conduct joint research on innovative networking applications and parallel driving.

Director of the State Key Laboratory of Complex Systems Management and Control, Vice Chairman and Secretary-General of the Chinese Society of Automation Wang Feiyue said: 'Through this cooperation, we believe that our research results in parallel driving, parallel intelligent systems and other fields, combined with Intel's automatic driving The accumulation of technology in the field will help auto-driving the landing in China.

Professor Li Keqiang of the Department of Automotive Engineering at Tsinghua University stated: 'At present, we have completed a series of basic theoretical and applied technological researches in the area of ​​intelligent networked autos. This cooperation with Intel will help us focus on a global perspective and link global resources. , Together to promote China's autopilot, intelligent network of automotive industry innovative application research. '

In the past few years, Intel has continued to increase its layout in the field of automated driving. Last year, they spent a large sum of $ 15.3 billion to acquire Mobileye. According to data, Mobileye is a well-known technology company headquartered in Israel, this 18-year-old technology The company was listed on the New York Stock Exchange in 2014. At that time, the market value had reached 5 billion U.S. dollars. Currently, there are about 27 million cars from 25 manufacturers on the market that have driving assistance functions, and Mobileye products account for 70% of the market share. .

Just this month, Mobileye also announced that it has reached a cooperation agreement with European automakers on the autopilot technology of 8 million vehicles. Intel expects to start providing automotive manufacturers with a fifth-generation system chip based on Mobileye EyeQ5 in 2021. Designed to provide visually-centric computer sensor fusion for fully self-driving cars.

2016 GoodChinaBrand | ICP: 12011751 | China Exports