Semiconductor dual "British" to reveal the future trend of the chip industry Industry News Silver Wing

Recently, the CPU-based Intel and GPU-based NVIDIA has released its first quarter earnings this year, which both profits were up 45% and 48%, arguably the growth of both profits are considerable, especially for Intel, in the quarter when the PC market is still declining, still based on PC chip Intel can achieve such a profit growth is valuable.

  But the fact is far from it looks so simple, reflected in the capital market, Intel has made a seemingly good earnings, its stock did not rise or fall, down as high as about 6%, compared to the stock price Rose about 14%.

So the question comes, why is the same profit growth, and the growth rate is almost the same, but the performance of the two sides in the stock is quite different? What is the real reason behind this? But also indicates the chip industry, how the trend?

  In fact, as long as a little attention to both the earnings and industry analysis is not difficult to find the reason. As cloud service provider data center demand for the chip is very strong, the first quarter of the data center business revenue of $ 409 million, more than doubled, easily defeated analysts expected $ 318.2 million. In contrast, Intel's data center business (DCG) revenue of $ 4.2 billion, despite an increase of 6%, but less than expected, leading to Intel shares fell 3%. That is the two sides in the data center (chip) market performance is the key to determine the stock price.

Intel's internal business has also played a key role in saying that Intel's share price has fallen as a result of external (mainly from NVIDIA) competitive pressures. From the main business of Intel's revenue and profit, according to Intel's first quarter results show: PC chip-based client computing business unit (CCG) operating profit of 3.031 billion US dollars, an increase of 60%, while Data center server chip-based DCG operating profit fell 16% year on year to $ 1,487 million.

  And the well-known fact is that CCG business in the PC industry is declining, the data center server chip-based DCG is Intel's current and future growth momentum is the industry (including Intel's own) consensus, not to mention the quarter of CCG operations Profit growth is not sustainable, this is still down from the PC market, Intel CEO Division odds in the earnings release this year, the business will still appear single-digit decline in the forecast is evident, and then once again proved To the data center server chip-based DCG business performance for Intel's pivotal.

In fact, the analyst for the quarter of Intel's DCG operating profit decline are puzzled, have to ask during the conference call. And Intel is given because the future of the process node as a preferred, due to higher costs. Is that really the case?

  "Now, the world's major Internet and cloud service providers are using NVIDIA's GPU chip." NVIDIA Chief Financial Officer Colette - Kress (Colette Kress) after the end of the earnings conference call said.

  I do not know what the industry to see this sentence? What we see is that Craig said these big companies include social networking Facebook, Alphabet's Google, IBM, Microsoft and Alibaba Group. Even the Amazon AWS cloud services are using NVIDIA chips. And these are the main customers of Intel's DCG business.

In this, some people may be questioned, in the current data center server chip market share of more than 90% market share of Intel, NVIDIA's impact will really cause a substantial impact on Intel? Here we may wish to look at NVIDIA's GPU in the data center server chip market performance.

  With its last year's Tesla P100, the DGX-1, built by eight Tesla P100, has a GPU throughput of 250 legacy servers. In addition, the DGX-1 system also includes a set of depth learning software, the Deep Learning GPU Training System (DIGITS), which can be used to design the Deep Neural Network (DNN). It is understood that DGX-1 can speed up the training 75 times, the CPU performance by 56 times, that is, Intel dual Xeon system requires more than 250 nodes and 150 hours to train Alexnet, and DGX-1 only need a node 2 hours, the latter in the performance and node total bandwidth Have a clear advantage.

In fact, in the Google Brain project, NVIDIA's GPU has been measured relative to the advantages of Intel CPU, the project, the three machines in the configuration of the 12 GPU, the performance reached 1000 nodes containing the level of the CPU cluster.

  Into this year, NVIDIA has introduced a new Tesla V100, this AI chip has 21 billion transistors, a year ago than the Tesla P100 even stronger, while Tesla V100 has 5120 CUDA processing kernel. In this regard, NVIDIA CEO Huang Renxun said that the chip's ability and energy efficiency will help cloud service providers to significantly enhance their ability to use AI (cloud services and data center to provide business development trends), can improve the ability of the data center 15 times instead of 400 traditional CPU servers, and even no need to create a new data center.

See here the industry should be clear that the British Weida for the impact of Intel, and now and the future, NVIDIA's main force of the GPU will be in varying degrees, the impact of Intel in the data center chip market CPU sales, revenue and profits, in fact, this The impact of the quarter should be on the show, but Intel to the future process node as the first choice, the higher the cost of its operating profit due to decline in the light away.

  What worries the industry is that, on the GTC 2017, which was hosted by NVIDIA today, NVIDIA announced that the number of GTCs has tripled in five years, reaching 7,000 this year, and GPU developers have grown 11 times to more than 50 million, CUDA drivers and SDK downloads are more than a million.

  In addition, NVIDIA also introduced the GPU cloud platform and open source Xavier DLA, we see is not only in the GPU itself, but on the basis of this is to build to meet the AI needs of the future data center of the new ecology, this is the Intel Especially after the traditional data center and cloud services market, Intel addition to the CPU, the largest moat is also built around the x86 CPU ecology.

In view of the above cloud services and data center providers based on their own and provide a business function with AI changes and the rapid development of NVIDIA, Intel is not without awareness and action, but with the NVIDIA as early as around 2011 to start AI chip research and layout In May, the acquisition of Computing Visual Software Inc. Itseez; Silverwing June acquisition of FPGA manufacturers Altera; August acquisition of the depth of learning start-up company Nervana Systems; September Acquisition of machine vision start-up company Movidius and a series of concentrated outbreaks are almost all related to the so-called AI.

  In theory, Intel's acquisition of these enterprises in the AI chip has its own advantages, for example, although the GPU compared to the CPU has some advantages, but compared with Altera's FPGA is still a lot less (a researcher test, compared to GPU, FPGA architecture is more flexible, the unit energy consumption under the more powerful), the depth of learning algorithms in the FPGA can be faster and more efficient operation, and power consumption can be even lower; Nervana Systems study of the depth of learning chips with Cost-effective than the GPU, the processing speed is 10 times the GPU and so on.

For Intel's various layouts in the AI chip, Huang Renxun on behalf of the British Weida questioned: If the strong financial core (Xeon Phi) processor (Intel released last year, the so-called AI server chip, and thus triggered compared with the NVIDIA GPU, who Who is very suitable for AI, then why buy Altera?

Since buying Altera, Altera is also very suitable for AI, why buy Nervada Systems? If Nervada Systems is the real AI technology, to carry out the development and product launch, then the strong nuclear coprocessor how to do? If these three are suitable for AI, it does not mean that the strong nuclear coprocessor processor is not suitable for AI?

Although the above remarks do not rule out Huang Renxun favor their own AI chip suspected, but also indirectly that Intel is currently in the AI chip is still in the choice of path and integration stage and slow action, still need the final product and market to prove. But for now, the NVIDIA GPU is the only one to achieve a large-scale application program, and after the market test, which also explains why the same profit growth, but the British Weida shares skyrocketed, and Intel shares fell behind the real reason.

website：http://www.silverwing.com.hk/news_view_135_153.html

Company

Industry

Product Update..

Product Inform..

AE

Amotech

Central

Harvatek

Pocons

JoYin

Damai

iA

ECS

ITW

SANICO

Sangshin

EVE

Carling

RN2

ABCO

C² Battery

Amphenol

Contact us

Semiconductor dual "British" to reveal the future trend of the chip industry

Related News