Computing power architecture
The types of chips that provide computing power for AI include GPU, FPGA and ASIC
GPU is a kind of microprocessor specialized in image operation on personal computers, workstations, game machines and some mobile devices (such as tablet computers, smart phones, etc.). It is similar to Cu, except that GPU is designed to perform complex mathematical and geometric calculations, which are necessary for graphics rendering
FPGA can complete any digital device function chip, even high-performance CPU can be implemented with FPGA. In 2015, Intel acquired the FPGA long alter head with us $16.1 billion. One of its purposes is to focus on the development of FPGA's special computing power in the field of artificial intelligence in the future
ASIC refers to the integrated circuits designed and manufactured according to the requirements of specific users or the needs of specific electronic systems. Strictly speaking, ASIC is a special chip, which is different from the traditional general chip. It's a chip specially designed for a specific need. The TPU that Google recently exposed for AI deep learning computing is also an ASIC
extended data:
chips are also called integrated circuits. According to different functions, they can be divided into many kinds, including those responsible for power supply voltage output control, audio and video processing, and complex operation processing. The algorithm can only run with the help of chips, and because each chip has different computing power in different scenarios, the processing speed and energy consumption of the algorithm are also different. Today, with the rapid development of the artificial intelligence market, people are looking for chips that can make the deep learning algorithm perform faster and with lower energy consumption
to practice guitar: most of them are basswood panels, some even basswood back panels, which are less than 300 yuan, also known as firesticks (because they are made of the same material as firewood, and the intonation is poor)
entry guitar: most of them are spruce panels, rosewood finger boards, and the back panels are not necessarily. The price is about 1000 yuan, and most of them are 500 yuan or 600 yuan, About 1000 yuan, there are also some elementary Miandan Qin
medium Guitar: most of them are spruce panel, rosewood fingerboard, back side panel, mahogany and fire rosewood. Generally, they are middle end single piano of various series, and some of them are full single (very few), with the price around 3000
high end: all single guitars and spruce veneers are the most popular. The price is four or five thousand or more. There is no upper limit. There are tens of thousands of them, but most people don't need them
tips on guitar selection: http://www.jitatang.com/guitar It is recommended to refer to
no matter where you buy it, you should know some experience to avoid being trapped. Generally, you will be trapped when you go to a piano shop for the first time. The most important two points (you can't hear the timbre and so on):
1. Look at the appearance and choose the log color or black directly Sunset color (the classic three color system), never white, Pink Guitar, that is the above said practice piano grade
2. Check the handle. When the guitar is tuned to the standard pitch, the distance between the string and the fingerboard should be about 3mm (that is, the thickness of a coin) at the 12th point. If you press the string too high, you will feel tired, otherwise it will cause the phenomenon of playing. A good hand feeling is that you can easily press any tone with your left finger without any proct or noise, and you don't feel any effort when you press it horizontally
3. Performance (in fact, you can't see it. You can get what you pay for. It's right to choose the most expensive one in your budget.)
-
don't understand the meaning of "assembly" in the title, is it assembly of prefabricated structure
-
"connected layer" refers to which layer? There may be a reference to "foundation layer" at the budget entrance. There is no "foundation layer" or "connected layer" in the building construction drawings and structural construction drawings. Either it is the ground floor (first floor), or it is the basement (negative 1 floor or negative 1 floor) × From the top of foundation to ± 0.000 shall not form a layer unless an overhead layer with ventilation and moisture insulation is designed
-
please explain the question correctly, ask clearly, don't be specious, and make clear what to answer< br />
The design and naming of Huawei's previous chips are basically based on the more grand names in ancient times, which also shows the extensive and profound culture of ancient China. It's inspiring to hear that. This time, the name of a foreign scientist has attracted more attention. In the final analysis, it's just a name
The R & D codes of scientific and technological procts and the use of foreign terms are common. The American air defense system "Aegis" is a typical example. What's wrong? Some people ridicule Huawei for nearly registering all the terms in the book of mountains and seas. Huawei uses a lot of "more patriotic" terms. Now it registers a "Da Vinci architecture". As a result, a small group of people's hearts suddenly can't stand it, so they make a farce. What happened to Da Vinci? What happened to Vienna? They are all very famous international terms. I'm very happy to register them. Why can't Huawei use the word "world" when its procts are sold in the world< br />The core micro architecture of NVIDIA graphics card has experienced Tesla, Fermi, Kepler, Maxwell, Pascal and Turing
CPU architecture is a specification given by CPU manufacturers to CPU procts belonging to the same series. The main purpose is to distinguish different types of CPU. At present, the CPU instruction set classification on the market is mainly divided into two camps, one is the complex instruction set CPU led by Intel and AMD, the other is the reced instruction set CPU led by IBM and arm
The details of NVIDIA graphics card architecture are as follows:
2000 - acquired 3dfx, the pioneer of graphics technology; 2001 - enter the integrated graphics market; 2002 - rated by Fortune magazine as the fastest growing company in the United States; 2003 - acquired mediaq; 2004 - SLI released, which greatly improved the graphics processing capacity of a single PC; 2005 - developed processor for Sony game console; 2006 - revolutionary CUDA architecture appeared
2007 - selected by Forbes as the best enterprise of the year; 2008 - Tegra mobile processor comes out; 2009 - the first GPU technology conference, launched Fermi architecture; 2010 - help the world's fastest supercomputer; 2011 - acquisition of icera, the baseband leader; 2012 - Launch GPU based on Kepler architecture; 2013 - Tegra 4 series processor launched
2014 - release tegrak1shield tablet, Android game fire; 2015 - deep cultivation and deep learning; 2016 - driving AI revolution; 2017 - Volta architecture comes out to further promote modern AI; 2018 - turing architecture came out and redefined computer graphics; In 2019, AI computing power will continue to innovate in all walks of life
2. The internal force of the unit is calculated by the moment distribution method, because the columns on other floors are elastic connection except the bottom of the bottom column, in order to rece errors, the linear stiffness of the columns on other floors is multiplied by a rection factor of 0.9, The corresponding transfer coefficient is also changed to 1 / 3, and the bottom column is still 1 / 2
3
Reconfigurable computing architecture (cgra) is a computing mode that can dynamically configure the software and hardware structure according to the transformed data flow or control flow
the biggest advantage of cgra is that it can organize computing resources with different granularity and functions through the airspace hardware structure, adjust hardware functions through hardware configuration ring operation, and interconnect the hardware resources with good function configuration to form a relatively fixed computing path according to the characteristics of data flow, Thus, the data-driven calculation is carried out in a way close to the "special circuit" (as shown in the figure). When the algorithm and application are transformed, the hardware is reconstituted into different computing paths by configuration again. On the one hand, because there is no delay and energy consumption of fetching and decoding operations in the traditional instruction driven computing architecture, and on the other hand, it is executed in a way close to "special circuit" in the computing process, which is an efficient and flexible computing architecture, so it can achieve a good balance between computing power and energy consumption, and achieve a higher energy efficiency ratio
to add a comparison with traditional chips:
CPU architecture is based on general instructions, and is very flexible in time domain execution mode. However, it costs a lot to take the finger, to decode, and to control if the conditional branch prediction is encountered. At the same time, in the time domain execution mode, the data reuse rate is low, and the energy consumption of frequent memory access is high
TheGPU architecture adopts the SIMD execution mode based on instructions, taking into account the general requirements such as 3D graphics processing. In addition to the cost of fetching and decoding, it also has the design of high throughput and high bandwidth. In the actual AI processing process, the bandwidth and resource utilization change greatly, the utilization efficiency is low, and the energy consumption is high
TheNPU computing architecture has customized the specific neural network, which is still based on instruction execution, so the overhead of instruction reading and decoding can not be avoided. At the same time, for the layer without customization, it can not be handled well, and the flexibility is limited. Compared with NPU, cgra has more than 10 times performance improvement
FPGA computing architecture is fine-grained bit level, based on look-up table execution mode, hardware programmable, very flexible. However, a large number of fine-grained LUTS make the internal wiring very complex, and the existence of a large number of storage based LUTS and interconnection lines will also consume a lot of power consumption