One of the remaining companies is very unique: it does not have a high market value, and its competitors are far stronger than themselves, but they can continue to develop under such strong pressure, and ultimately not only survive, but also in these dozens In the years, the powerful competitors felt the pressure and created a business miracle, and this company is today's protagonist - AMD.

In the late 1960s and early 1970s, "Silicon Valley" was not yet named "Silicon Valley," and computers were still scarce, after decades of rival Intel developed 3101 memory chips, followed by DRAM dynamic memory. Declared the core memory invented by An Wang to be history. Jerry Sanders (Jerry Sanders, with the $50000 raised by an old colleague of Xiantong Semiconductor, founded Advanced Micro Device, also known as AMD, in Sunnywell, Calif., on May 1, 1969. Since then opened a period worth remembering, can be called the history of business history of the development of the process.

If a company is established 50 years ago, it will be enough for everyone to praise it. Looking back at AMD's 50-year development, he has challenged opponents very strong. Since the 1980s, he has competed with Intel in the CPU field. After competing with NVIDIA in 2006, he is a full challenger. After the failure of Motorola, Transmeta, etc. to challenge the "Dragon" Intel, only AMD is still on this road, although AMD's market value of about 30 billion US dollars is far from Intel's more than 230 billion US dollars market value. However, AMD has not only survived, but in recent years, with Ryzen processors, GPUs and semi-customized services, it has gradually stepped out of the trough and its development has been on the right track.

In the past, AMD did not have the opportunity to continue to expand. For example, in the era of K8 architecture, Athlon Athlon 64 processor, and later Athon 64 X2 "true dual core" are AMD's brilliant moments, but then Intel took out Core 2 Duo. After that, AMD's R&D did not keep up with the pace of competitors, and once again fell into the point of comparison. Subsequent bulldozer architecture Although the CMT multi-cluster multi-threading technology idea is very good, but did not let AMD step into the glory of a few years ago. However, the Ryzen processor released in 2017 is gradually pulling AMD out of the trough, but not only in addition to the processor, but also AMD, the product line of the graphics card, it will take time to get out completely.

Now let's review AMD's 50-year history and see what it has gone from raising $50,000 to the current $30 billion market.

1. Leaving Fairchild to Cao Chuang AMD (1969-1976)

When it comes to the creation of AMD, you still have to put time forward. In 1947, William Shockley and John Barding, Walter Brighton invented transistors. William Shockley then worked to commercialize transistors and founded Shockley Semiconductor Laboratories when he left Bell Labs in 1955. Although his reputation and ability have attracted a group of talented scientists, strong scientific research skills do not mean strong management skills, because of his strange management skills and behavior, leading to the departure of eight outstanding scientists. And Shawkley also called them "eight rebellious" (Traitorous eight). Subsequently, the eight people received an investment from Xiantong Photography equipment Company to set up Xiantong Semiconductor Company. Because of these outstanding scientists and engineers, the development of Xiantong Semiconductor is very rapid, which attracts more talents to join.

The previous summary is coming to an end, when a young man named Jerry Sanders joined Xiantong Semiconductor. In the late 1960 s, but due to problems with Xiantong Semiconductor, talent gradually left, including the "eight rebellious" that once founded it. Best known is Robert Noyce, Gordon Moore pulled up Andy Grove to set up Intel. But it was not just these people who left. Jerry Sanders, then director of sales for Xiantong Semiconductor, also left in 1968.

Fairchild Semiconductor was also a long way to go, but it also contributed too much talent to the development of the industry.

Jerry Sanders then raised $50000 with several Xiantong Semiconductor employees to create Advanced Micro Device (AMD). At Sunnywell, where Xiantong Semiconductor is located.

Many of the employees who leave Xiantong Semiconductor are in semiconductors and integrated circuits, and Jerry Sanders is no exception. In the early days, AMD mainly produced logic chips, and in the same year AMD produced its first product, the Am9300 4-bit shift register, which began to be sold in 1970. It wasn't long before AMD completed the Am2501 logic counter for the first product with its own intellectual property rights, and was successful.

Image from AMD Social Network

Am2501 logic counter

In the past few years, AMD has introduced a memory stick, but that is an OEM. Actually, AMD has actually introduced memory in history. In 1971, AMD entered the RAM market and introduced the 64-bit bipolar RAM Am3101. This year was also a bumper year for AMD, and at the end of the year, AMD’s sales reached $4.6 million.

After more than three years of establishment, AMD went public. Since then, AMD has ushered in the first collaboration with Intel, as the second source of Intel's large-scale integrated circuit, providing dual 100-bit dynamic shift registers. AMD's product line has also expanded rapidly. By 1975, AMD had 212 products, 49 of which were AMD's proprietary products. Most of the products at that time were provided for telecommunications, banking, aerospace and even the military, so stability is very important for these integrated circuits. In June 1976, AMD's products became an integrated circuit company that was established in the past ten years and obtained military and aerospace certifications with its excellent quality and stability.

In less than 10 years since its founding, AMD has achieved such results, and it is also excellent compared to Intel, born with the "golden key". Although less than a quarter of AMD's products are proprietary products, and some products are the second source of other manufacturers, in the next era of microprocessors, although initially produced for other manufacturers, But AMD is also gradually looking for its own unique product line.

2. Independent research and development of microprocessors and "second source" (1975-1985)

Intel introduced the 4004 processor in 1971. Four years later, in 1975, AMD entered the microprocessor market and introduced the Am2900 series of processors. This series of processors has also been widely used. It is modular in design, requiring more integrated circuits to work with a single processor IC. The Am2901 is a 4-bit ALU (Arithmetic Logic Unit) and the heart of the entire modular system. This series of processors is also used by many well-known manufacturers, such as the famous DEC, Xerox, and Atari has even used this processor to build an arcade.

However, the biggest impact on AMD since then was the reverse engineering of the Intel 8080 processor in 1975. In 1976, Intel began using microcode in its CPU. However, in the same year AMD and Intel signed a cross-licensing agreement, so since October 1976, AMD can use Intel's micro-code on its own microprocessor and peripheral devices.

After that, AMD was a relatively smooth one. In 1977, a joint venture with Siemens in Germany was established. However, the situation was not good. In 1979, the two sides had differences and AMD acquired the remaining shares of the joint venture. It then focused on producing Intel's x86 processors as a "second source."

However, from this time, the patent cross-disciplinary dispute between AMD and Intel has begun.

AMD and Intel signed the first cross-protocol in 1976. Since then, AMD has been able to use the microcode of Intel's current processor "justifiably". However, the cooperation with Siemens at that time injected a lot of money into AMD, so at that time AMD also had a self-developed microprocessor, but after the two separated, AMD completely shut down its advanced microcomputer subsidiary in 1981.

In 1978, Intel introduced the first 16-bit microprocessor 8086. In 1981, Blue Giant IBM began to manufacture its own PC, hoping to use x86 processors, but required the provision of x86 processors to have a "second source", so in 1982 In February, Intel and AMD formally signed an agreement, and each company has the right to become another supplier of semiconductor products to another company.

Intel's 8086 processor

At the same time, Intel and AMD also extended the cross-licensing agreement in 1976. At this point, AMD has signed two agreements with Intel, all of which are technology-related. While producing microprocessors for Intel, AMD has also accumulated a lot of manufacturing experience.

AMD's 8086 processor

In 1982, AMD began to produce products such as 8086 and 80186. In 1984, in the face of the rapidly growing IBM PC and compatible machine market, the 80286 chip was again "cloned" and named Am286. Although as a "second source", AMD did not give up research and development. In 1984, it also produced the world's first 512K EPROM (Erasable Programmable Read Only Memory) and the industry's first single-chip burst error processor. Advances in small disk drives and the development of the PC market. In 1985, it was also named the "Fortune 500" enterprise.

However, in 1985, the semiconductor industry changed a lot. Japan dumped the DRAM produced by its own company and quickly occupied the market. This has a great impact on AMD and Intel, which have DRAM production lines, so after that, AMD And Intel have both withdrawn from the DRAM market. In the 1980s, in addition to the continued production of x86 processors, AMD also developed its own processor and CMOS technology. In 1986, it introduced the first single-chip compression/expansion processor, which promoted office automation and then 1 million-bit EPROM. . In 1987, the industry's first graphics card with all computer display card interfaces and software at the time was introduced. The Am29000 (commonly known as 29k) produced in 1986 is a reflection of AMD's R&D capabilities. The Am29000 processor is a 32-bit processor that uses the RISC architecture, and this processor also affects the design of the AMD processor.

3. Self-developed x86 processor, competition unfolding (1984-1996)

In the late 1980s, although AMD strengthened its independent research and development for various reasons, it was mainly used as the “second source” in the x86 processor market. When it entered the 1990s, AMD changed quite a bit. On March 25, 1991, AMD first demonstrated the Am386 processor and joined the 32-bit 80386-compatible CPU market. Since then AMD is no longer a "second source", AMD began to compete with Intel for x86 processors.

In 1984, in order to consolidate its market advantage, Intel decided not to provide product information with AMD, and ultimately refused to provide AMD with the technical details of the 80386 processor. So in 1987, AMD filed an arbitration with the court, but at this time Intel canceled the 1982 extended technology exchange agreement. And this dispute has been going on for years, and finally ended in 1994, AMD received support.

In this dispute, Intel also responded to AMD. In 1990, Intel negotiated with AMD about AMD's right to use Intel processors. So in the absence of intellectual property rights, AMD re-developed Intel's x386 and x486 processors. This is also the Am386 processor mentioned earlier. In 1993, AMD immediately released the Am486 processor, which offered higher performance at the same price and was favored by companies such as Compaq (Compaq, TIm Cook).

AMD Am386

AMD Am486

In the end, the mutual litigation between the two sides lasted for many years, and by 1996 it was over. In the agreement, AMD obtained the right to use Intel x386 and x486 microcode many years ago, but did not get the right to use the next generation processor microcode. So far, the litigation concerning the exchange of microcode intellectual property between the two parties has also ended. However, AMD and Intel, which are opponents of each other, have also embarked on the road of independent research and development of micro-architecture.

4. Competition on the road of independent research and development (1996-2003)

Intel retired from the first Pentium processor in 1993. As the successor to the 486 processor, Pentium adopted the new P5 microarchitecture, which is also the first generation of superscalar IA-32 architecture. Although AMD's Am386 and subsequent Am486 processors are also widely welcomed at this time, performance is still far behind.

Three years later, in 1996, AMD introduced the K5 processor. Although the delay due to research and development difficulties, this processor still brings a lot of innovation.

AMD K5 processor Die photo

As mentioned before, AMD introduced the Am29000 processor in 1986, using the RISC architecture. The K5 processor is a RISC architecture with a highly parallel Am29000 processor and an x86 decoding front end. So the K5 processor also provides good x86 compatibility. All models have 4.3 million transistors, five out of order integer arithmetic units and one floating point unit. The branch target cache is 4 times larger than Intel's Pentium processor. In terms of caching, it has a 16KB four-way set of connected instruction cache and 8KB of data cache. At the same time, the floating point division and the square root are mechanically verified, and the operation result is loyal to the real mathematical result. These indicators are more advanced and more accurate than Intel's Pentium processor, but due to the low clock frequency, the actual performance is lower than the Intel Pentium processor, which also causes the K5 processor not to be the same as the previous Am486 processor. Widely acclaimed.

Although AMD improved the performance of the K5 processor by increasing the frequency later, AMD subsequently introduced the K6 processor in February 1997. AMD's K6 processor has made even more updates than the K5 processor. Usually the processor design will be much earlier than the release time. In 1996, AMD acquired NextGen, another x86 processor manufacturer at that time. NextGen's R&D team also joined the K6 processor R&D team. So although the new processor is named K6, the product is very different from the previous generation K5. AMD incorporates NextGen's technology, which includes feedback dynamic instruction reordering, the MMX instruction set, and a floating-point arithmetic unit (FPU) in the K6 family of processors. In terms of pins, the K6 processor is the same as Intel's Pentium processor and can be used on motherboards with Socket 7 slots.

AMD K6 processor

The K6 processor's initial frequency is 166MHz and 200MHz, but later introduced a version of 233MHz or even 300MHz frequency, 300MHz frequency version was released in 1998, after which AMD released an updated version of the K6-2 processor.

The K6-2 processor is an improved version based on the K6 processor. It has a higher frequency and uses a 0.25 micron process to upgrade the interface to Super Scoket 7, but is compatible with the previous Socket 7. The K6-2 processor also has a larger 32KB instruction and 32KB of data.4KB's level 1 cache. But the biggest improvement is the addition of the 3DNow! SIMD (Single Instruction Multiple Streaming) instruction set, which significantly improves performance. As a result, the number of transistors in the K6-2 processor has increased dramatically to 9.3 million.

At that time, the processor design was not so complicated compared to the present, so the processor update was very fast. In August 1999, AMD took out the K7 processor, and this time it launched a brand for the K7 processor. Athlon (Athlon) is still in use today. The K7 architecture design team was led by Dirk Mayer, who later became the CEO of AMD. At that time, AMD competed directly with Intel in the processor market, but AMD still uses the same processor socket as Intel. However, this time AMD had an authorization problem on the processor socket, so AMD could no longer use the SLOT-1 slot used by the Intel Pentium Pro and Pentium III processors, and finally used a slot called SLOT A. This slot is based on the EV6 bus. It was originally developed by DEC for the Alpha processor and has a higher bandwidth than the Scoket 7 sockets of the previous K5 and K6 eras.

Along with the K7 processor, there is also a big event in the computer industry, which is to break through GHz. Through cooperation with Motorola, AMD has completed the copper interconnect process one year ahead of schedule, and can be modified to use the 180nm process. According to Moore's Law, the smaller the linewidth of the chip, the lower the power consumption, which gives the processor the ability to hit higher frequencies. So in March 2000, AMD announced the launch of the 1GHz Athlon 1000 processor, which was earlier than the rival Intel and won the GHz battle.

Then in June 2000, AMD released the second generation of Thunderbird Athlon processor. Unlike the original SLOT A slot, this generation of Athlon processors also used PGA (pin grid array) for the first time. In the form of a slot, this form of CPU socket has also become a slot form that AMD has continued to date, but AMD still provides a CPU in the form of a SLOT A slot. Relative to the first-generation Athlon Classic architecture, the "Thunderbird" architecture has a new cache architecture designed to further enhance processor performance. The Thunderbird Athlon also became the most successful product of AMD at that time.

AMD also insisted on an updated rhythm every year. In 2001, it launched the third-generation Athlon processor code-named "Palomino". This generation of processors AMD also named it "Athlon XP", and in the same year,MicrosoftReleased the most classic desktop operating systemWindowsXP. This generation of processors AMD has also made many updates. The first is the first time that Intel Pentium III's complete SSE instruction set has been inherited from its own processor, and this instruction set is Intel's 3DNow! instruction set that was previously released by AMD. What's more, AMD also introduced the "desktop dual-channel" technology at that time, and the Athlon supporting "dual processor" is called Athlon MP. However, the third-generation "Palomino" Athlon processor was still manufactured in the 180nm process, so the higher frequency caused the processor to be hotter.

Just entering the new millennium, the notebook market has gradually entered the field of vision due to factors such as technology level, so AMD also launched Palomino's core mobile Athlon 4 processor in 2001.

Later in 2002, AMD introduced the fourth-generation Thoroughbred core Athlon processor. At this time, AMD also used the 130nm process for the first time. Compared with the previous 180nm process processor, the core area was significantly reduced. At this point, the better process technology also allows the Athlon processor to have a higher frequency and better overclocking performance.

Then the fifth and last generation Athlon processors were introduced in 2003. AMD continues to improve processor performance by increasing L2 cache and increasing front-side bus speeds, but at this time AMD is no longer facing the "Coppermine" Pentium III or "Willamette" Pentium 4, but the updated "Northwood" Pentium 4 processing. Although "Barton" Athlon is highly praised in the DIY circle for its superior overclocking performance, the architectural disadvantage also allows AMD to prepare a new processor to cope with this situation.

Early computers were still very expensive, and AMD also introduced lower-cost processors by cutting core caches and other methods. So after AMD launched the K7 Althon processor, it also launched a brand with a lower positioning, called Duron, the processor of this series. The core of the Duron processor comes from the "Thunderbird" Thunderbird core of the Althon processor. The biggest difference with the Thunderbird core is the reduction of the L2 cache, which is reduced from 256K in Althon to 64KB, which also makes Duron's L2 cache. The size is not as good as the L2 cache size of the Intel Celeron processor competing with it. However, because the Duron processor core still inherits the excellent cache design of the Althon "Thunderbird" core, the Duron processor actually does not run much slower. Since then, the Duron processor has been launched with Athlon for three generations until 2003. In 2001, the Mobile Duron processor was also introduced for the notebook market. The Duron processor is also popular with its super-overclocking performance. At that time, the "Dragon Dragon Pencil Method" cracked multiplier became a talk of many DIY enthusiasts.

However, the Duron series was replaced by another series, Semporn.

In this completely self-research processor time, AMD and Intel many times, each other win or defeat, from this also can see that AMD is really good strength. But between 2002 and 2003, several things happened to AMD, one of which was that the performance of Athlon processors began to lag behind that of Intel;, and the other was to work with IBM to develop SOI (silicon on insulators) technology. And the last one was Jerry Sanders, the founder of AMD, who announced that he would no longer be CEO., who founded the company himself.

5. The founder left, 64 and "true and false dual core" (2002-2006)

After entering the new millennium, the competition between AMD and Intel is becoming more and more fierce. But at this time, Jerry Sanders, founder of AMD, announced in 2002 that he would no longer hold the position of CEO, given to Hector Ruiz, and better known as Lu Zhiyi.

In 2003, AMD, after a year of coaching, continued its path of innovation and launched the first 64-bit processor based on the x86 architecture, Opteron. As a new processor, Opteron processing uses the newly designed K8 architecture, but the Opteron processor is geared towardserverMarketing. On September 23 of the same year, AMD officially brought 64-bit computing into the PC field and launched the Athlon 64 FX processor, which mainly competes with several generations of Pentium 4 processors. Although it is a 64-bit processor, it is also compatible with 32-bit applications. At the same time, the K8 architecture has also experienced many years of development, from 130nm to 65nm process technology, from the core to the dual core, and there are many CPU sockets, which is the "veteran" in AMD products.

AMD Opteron Opteron processor

AMD Athlon FX processor

In 2003, when people were still attacking higher frequencies, AMD first came out of a different path, updating 32-bit processors that have been in use for many years, and providing 64-bit computing for the first time in the PC field.

However, although the initial Athlon 64 processor is very competitive in terms of predecessors and rivals, AMD quickly released a revised version of the processor in 2004, and these processors use the 130nm SOI process. Also replace the CPU socket with Scoket 939. Compared to the unrevisioned Athlon 64 processor, the new version of the processor has been updated on the memory controller and front side bus. The memory controller adopts a dual-channel design, and the HT (HyperTransport) bus frequency is increased from 800MHz to 1000MHz. At the same time, under the impetus of "Moore's Law", the AMD processor has also crossed the 100nm mark and started to adopt the 90nm process. Manufacturing the CPU and launching the "Winchester" core. Since then, the "Venice" and "San Diego" cores have been inherited in 2005, and the cache has been improved. For example, the SSE3 instruction set added in the Intel Pentium 4 "Prescott" core has been added, and the memory has been greatly modified. Controller.

Obviously, AMD can release the processor so quickly, which shows that AMD is really well prepared, and it is amazing that AMD can innovate in processor design in such many generations of processors. The current CPU is a built-in memory controller, and in that era, AMD already had a built-in memory controller in the CPU. At that time, due to the gradual enhancement of processor performance, there was a higher demand for heat dissipation, and AMD also developed the Cool'n'Quiet function, which reduces the CPU's operation and voltage, and reduces power consumption and temperature when the load is small.

However, when AMD triumphed, there was a problem with Intel. In 2004, Intel's Pentium 4 processor reached a higher frequency through an ultra-long pipeline design, but did not lead in performance. At this time, due to manufacturing processes and other reasons, the 4GHz of the Pentium 4 processor did not arrive as expected, so the then Intel CEO Barrett sneered at an industry conference, apologizing for the cancellation of Intel's new generation Pentium 4 processor. . This matter also points out the future direction of CPU development, that is, multi-core.

In 2005, AMD announced the next-generation Athlon 64 processor, the Athlon 64 X2. This model of processor is also the first "native dual core" processor. But not long ago, Intel announced the dual-core Pentium D processor, using the same socket LGA 775 as the last Pentium 4, using two Pentium 4 "Prescott" cores, which also led to a huge controversy afterwards. "true and false dual core."

AMD Introduces Athlon 64 X2 Dual-Core Processor

After AMD introduced the Athlon 64 X2 processor in 2005, it said that its dual-core processor is a "true dual-core". AMD's dual-core has an advantage in architecture, and later said in the promotion that "the two cores on a chip are The real dual-core, and Intel's two chips on a processor, is a fake dual-core." Then there was a lot of battle between AMD and Intel for manufacturing and software operation. However, the conclusions are also known. Later, Intel's Core 2 Duo processor introduced in 2006 also turned to a core design of two chips.

At this time, AMD's limelight is no more than two. When the integrity is at its peak, the opponent launched a new processor architecture in 2006, which was swept away. At this time, AMD is also brewing a very influential acquisition.

6. The opponent's counterattack is pressing hard, and the acquisition of ATI has gradually declined from the glory (2006-2009)

The "true and false dual-core" battle has made AMD's reputation more and more high, and in fact, during that time, AMD is indeed more dominant in performance; at this time, AMD is brewing an acquisition plan, which is to acquire the world's two major GPUs. One of the company's ATI. In July 2006, AMD announced the acquisition of ATI for $5.6 billion, and AMD became the first vendor to have both high-performance CPUs and GPUs. However, this acquisition decision also caused AMD to have some problems in the finances for the next period of time. Although Lu Zhiyi was not dismissed by the board at the time, the subsequent financial situation eventually led him to resign in July 2008. This transaction has had an impact on both AMD and ATI. Since the acquisition, ATI's share of the GPU sector has begun to decline, and AMD is also not good.

AMD acquires ATI

In 2006, Intel introduced the Core 2 family of processors, and its newly designed core architecture allows it to achieve greater performance without the need for higher frequencies. All of a sudden, the Athlon processor has the advantage of the Pentium 4 series processor. At the end of 2006, Intel introduced a quad-core processor, which also put AMD at a disadvantage again. Fortunately, AMD's response is still fast. In September 2007, the first native 4-core third-generation Opteron processor was introduced, followed by a new series of Phenom processors in November to meet the challenge. But I don't know why, because of the reason, the TLB error was found in the Phenom processor, and AMD subsequently fixed the bug and introduced a new stepping processor to solve the problem.

AMD launches Opteron 64 native quad core processor

The first generation Phenom processor uses a 65nm process, the CPU socket is AM2+, supports DDR2 memory, has quad-core, dual-core and very unique three-core version.

AMD then introduced an upgraded version of the Phenom II processor at the end of 2008, upgrading the process to 45nm SOI and supporting DDR3 memory. The new generation of Phenom II processors are also available in quad-core, dual-core and triple-core versions, while the 6-core version is also available later.

However, the performance of this generation of Phenom processors is still weaker than that of competitors. However, the second-generation Phenom processor has the same fun as the "Dragon" and "Barton", which is a lot of fun for players. core". In the second-generation Phemon processor, such as the three cores are actually shielded by the 4-core processor, and the player can open the shielded core through some means to achieve the purpose of improving performance. And this also allows many players to choose a processor that can open the core. Although opening the core may lead to unstable operation, the players are still enjoying it.

The K10 core is not only used on the Phenom processor, AMD also introduced the K10 Athlon and Sempron processors, and these processors also have the Athlon II X4 640T, which can satisfy the player's heart-breaking artifact. .

After the acquisition, ATI was a tough time. In 2008, the Radeon HD 4000 series graphics card was introduced, and the HD 4850 became the first graphics card with floating point performance exceeding 1TFLOPS, while the HD4890 was the first core frequency to exceed 1GHz graphics card.

Although the players are having fun, AMD has not been good at this time. After Lu Zhiyi left, Dirk Meyer, who led the development of the K7 architecture, took over the role of AMD CEO in 2008. In 2009, AMD also divested its chip manufacturing plant business into a fabless semiconductor company.

In 2009, AMD officially spun off its chip manufacturing division, founded GlobalFoundries.AMD after 40 years to become a Fabless company, and Jerry Sanders, who once said "Real men have Fabs," would never have thought AMD would make such a decision. But since then, it has become a good chip manufacturer in the industry through a series of acquisitions of GlobalFoundries, and continues to work with AMD.

7. Regrouping but with little success, APU frustrated PC market but spent elsewhere (2008-2013)

After Lu Zhiyi's exit in 2008, Dirk Meyer took over the role of AMD CEO, but due to miscalculation of the market, the betting PC market was eventually cancelled as CEO in 2011. Then Rory Read took over.

Replacing Lu Zhiyi's Dirk Mayer

After Intel introduced Core 2, Intel Core processor was launched at the end of 2008, and Intel's new "Nehalem" architecture once again improved the performance of the processor. AMD can only attract consumers to buy products and more cores through cost performance. To attract consumers.

But AMD is not sitting still, Dirk Meyer is focusing on the PC and data center market, developing a notebook platform, but his subsequent strategy has also led AMD to lose its subsequent rise in intelligence.Mobile phoneWhen the market, he said that the mobile and consumer electronics market will not impact the current PC market, resulting in a reduction in PC market share, but then the market proved that he was wrong, and he was canceled as CEO in January 2011 by AMD's board of directors. Then Rory Read joined AMD as CEO. During his tenure, he reorganized corporate debt and introduced new cooperation measures to keep AMD back on track and earn income healthier.

Dirk Mayer's successor, Rory Meyer

During Dirk Meyer's tenure, due to the impact of Intel Core 2 processor, AMD had to start developing a new generation of processors. In October 2011, AMD launched a re-engineering work, using AMD FX processing with "Bulldozer" architecture. Device. AMD also used a lot of new technologies in the "Bulldozer" core. It is the biggest innovation after AMD K8. He changed the traditional CPU design ideas, modularized the CPU, and each module has a class segmentation. For the two cores, the two microkernels are independent of each other and have a highly shared floating point unit, L2 cache, and the like.

First back in 2007, when AMD first released the SSE5 instruction set, but the opponent Intel did not passively follow, but launched the AVX instruction set, and because of Intel's strength, software developers turned to Intel, and finally AMD Seeing that the trend has gone, it was forced to turn to the AVX instruction set in 2009, which also led to the danger of "Bulldozer" not born, and then AMD's innovation and innovation of this architecture has just begun.

AMD uses a modular design in the "Bulldozer" core, and each module is divided into two microkernels, which results in a functional unit no longer a traditional integer + floating point design, two in each module Integer units, and "one" shared floating point units, and the integer units in the module are each equipped with a scheduler, so that two threads can be executed. AMD turns this multi-threading technology into "CMT" (Cluster Multithreading).

CMT Hyper-Threading Solution for "Bulldozer" Architecture

This is the biggest feature of AMD's new generation of processors, and this feature has caused many problems. For example, in later practical applications, AMD's processor performance was not as good as the publicity, even because this incident caused a class action lawsuit, saying that AMD falsely promoted multi-core processors, and by 2019, the case still had no results.

However, although the idea of ​​modular design is not so satisfactory in practical applications, it reflects AMD's next product idea, which is to improve the floating-point performance of the processor through graphics cards. This is AMD's APU product line.

AMD has long been rumored to add graphics to the processor, but the real product was officially launched in 2011. However, the original product still uses the K10 architecture CPU core, but it still gives people a certain surprise. The real APU is the second-generation APU code-named "Trinity". Because this generation, it really shows AMD's idea, that is, through the graphics card to improve the processor's floating-point computing power.

In modern computer systems, in addition to hardware, there is also a need for strong software support, so in order to achieve this "heterogeneous computing" goal, AMD supports technologies such as OpenCL C++, which can be easily used for CPU and GPU. Programming, so that they work better together and handle concurrent loads.

Then the APU, code-named "Kaveri", launched in 2013, went one step further and realized that the GPU directly uses the virtual address of the CPU to access the system memory and allows the "pointer-is-a-pointer" between the CPU and the GPU. The unified custom space makes it difficult to write heterogeneous code, and this is the goal of AMD, let the CPU and GPU really do their jobs.

After that, the APU update is more minor repairs.

The ability to achieve this, and AMD's other product line, is the GPU. At the end of 2011, AMD officially released a new GCN architecture GPU, and the current GCN architecture is still the basis of the current AMD graphics card, arguably the longest-lived GPU architecture. This architecture is very outstanding in terms of graphics and computing performance, so as the DX11 era (later support DX12, but DX12 Feature 11_1) graphics card, the general computing power is outstanding, sweeping the VLIE4 and VLIW5 architecture computing performance is low Happening. This also allows the APU from the third generation to realize the idea of ​​HSA heterogeneous computing. But unfortunately, although AMD's ideas are very good, and in the recent AI tide, heterogeneous computing is also the trend of the times, but in 2013, AMD has not been as brilliant as before and after 2005, the appeal is far less than before, so At this point, the software developer did not provide more support for the APU, so everyone's impression of him only stayed in the CPU and GPU integration, for most PC consumers, "throat" only exists in the slogan .

However, there is always a "village of the village", and the loss of the PC field does not mean that it has encountered difficulties in other fields. AMD has developed a "new skill" - a semi-custom chip, and this is the home console market. Around 2013, Sony's Playstation 3 and Microsoft'sXbox360 has been launched for many years, it is time to launch a new generation of game consoles, and both eyes have shifted from Power architecture to x86 architecture. At this time, AMD has launched the "Bulldozer" architecture, and APU products have also been launched. The price is also cheaper. The initial price of the PS3 and Xbox360 exceeds $399, and the APU is a good choice for devices that require a lot of control over home consoles.

Sony has found AMD to develop a semi-custom processor. The processor uses AMD's "Jaguar" low-power core, and the CPU part uses a total of 8 cores with dual modules. The GPU is also developed by AMD, which shares 8GB of GDDR5 memory between the CPU and the GPU. On the Microsoft side, the Xbox One game console has the same CPU portion as the Sony PS4, but the frequency is slightly lower, while the GPU portion is also customized by AMD. The CPU+GPU shares 8GB of DDR3 memory, but the Xbox One also has 32MB of high-speed ESRAM.

Playstation 4

The new version of Xbox One, which is also AMD semi-customized chip inside

In the PC field, although AMD strives to promote HSA heterogeneous computing, it has had little effect. But the idea of ​​heterogeneous computing has been successfully implemented in the field of home game consoles. Recently, the launch time of a new generation of home game consoles is coming soon, and there are rumors that Sony and Microsoft will continue to adopt AMD custom processors.

With the sales of the PS4 and Xbox One, AMD has made a lot of money and has the funds to develop the next generation of processors. At this time, AMD changed coaches again. In October 2014, Dr. Su Zifeng served as the fourth CEO of AMD.

8. Design God returns to regroup and gain competitive opportunities again after many years (2014-)

Familiar with this industry knows that the development of a new architecture takes many years, and the "Bulldozer" architecture has been going on for many years, and AMD needs a new processor architecture to save the tide. Soon after the chip design god Jim Keller returned to AMD in 2012, he started to develop the Zen micro-architecture, but it was not long before the "Bulldozer" architecture was released.

From 2012 to early 2015, AMD is quietly developing this new architecture. At this time, AMD relied on the "non-PC" field market to gradually alleviate the problems caused by the lack of product power in the PC field. In 2014, AMD chose the direction, let AMD return to the right Rory Read to leave AMD, then Dr. Su Zifeng served as the president of AMD and served as its fourth CEO. By the beginning of 2015, the actual development of the Zen architecture has been around for 3 years, so when the company announced its earnings report in early 2015, AMD revealed its product roadmap, and the long-rumored Zen architecture also surfaced.

Dr. Su Zifeng is the President and CEO of AMD

However, Zen micro-architecture is still under development, and AMD's CPU product line is still supported by old products. As the release of the Zen micro-architecture increased during that period, good and bad rumors were rampant. Although AMD's share price has begun to rise at that time, people still have doubts about the strength of their products.

However, at this time, AMD did not fall on the graphics card. In 2016, AMD introduced the Radeon R9 Fury series of graphics cards, the first use of HBM (high bandwidth memory), so that AMD Radreon graphics performance has been greatly improved. However, in the GPU field, although GCN is a very good architecture, compared to NVDIA, a competitor in the GPU field, the performance of the Radeon graphics card is not bad, but it is gradually falling behind.

For AMD, the processor is obviously more important. In 2015, rumors about the Zen micro-architecture processor continued, and at the end of 2015, it was reported that the Zen micro-architecture processor had passed the test. This also makes people's expectations stronger.

In June 2016, AMD CEO Dr. Su Zifeng officially demonstrated the Zen micro-architecture processor, using 8 cores and 16 threads, and IPC increased by 40% compared with the previous generation. However, "good things" still need to wait. In March 2017, AMD officially released the Ryzen processor using Zen micro-architecture.

AMD Zen processor physical exposure

The new processing still uses modular design, but more mainstream, the smallest CPU Complex module (CCX) has four x86 cores, each core has independent L1 and L2 cache, a single module shares 8MB L3 cache, the core changes The previous generation of CMT multi-threading technology, using more mainstream SMT multi-threading. The first generation of Ryzen processors used a dual CCX module design. Each CCX module was connected via AMD's newly designed Infinity Fabric summary/interface. The single processor consisted of 8 cores.

Ryzen processor Die

Ryzen processor switched to SMT Hyper-Threading Technology

In addition, AMD has added SenseMI technology to the Ryzen processor. With the Infinity Fabric bus, the AMD Ryzen processor has 100 sensors inside and on the motherboard to monitor system health in real time. Precision Boost provides precise frequency control. AMD also uses the Nueural Net Prediction neural network prediction and Smart Prefetch intelligent data read-ahead function to select the optimal processing path and reduce the operation delay.

Control voltage more accurately

Neural Network Prediction

Intelligent data pre-reading

These highlights bring AMD back to the same starting line as Intel. At present, Ryzen processors have been released for two generations. The second generation also has some performance improvement on the basis of the first generation, using a 12-nm process.

What is even more surprising is that AMD has entered the HEDT market since then. ThreadRipper processors have been introduced, raising the number of cores to 16. By the second generation launched last year, even the largest number of cores was 32 cores. This also shows consumers the great potential of Zen microarchitecture and Infinity Fabric.

The orange plastic shell outside is not used to protect the CPU. This is the CPU mounting slide, without which the CPU can not be fixed.

On the GPU side, AMD did not fall. In 2016, the Radeon 400 series GPU code-named "Polaris" was launched, but AMD pays more attention to efficiency in this GPU update, with 2.8 times the performance per watt of its predecessor. But it's worth noting that AMD does not position Polaris at the flagship level, it's just a mid-end graphics card. In terms of core improvements, AMD also improves geometric units, improves renderer efficiency, color algorithm and so on. Compared with R9 290X graphics card, its CU computing unit performance improves by 15%. Later, AMD launched a new process version of the Radeon RX500 Series graphics cards. But the bigger update is the Radeon RX Vega graphics card with GCN 5.0 architecture. In addition to conventional graphics performance improvements, there is also support for compact semi-precision computing. At the same time, in order to compensate for the bandwidth problem of HBM2 memory, AMD also developed HBCC high bandwidth buffer controller.

The latest work of AMD RTG is Radeon VII. Although Vega core has been used, the process has been upgraded to 7nm.

But the bad news is that Raja Koduri, head of AMD RTG, and others have left AMD. Although AMD has planned products for the next few generations, it is hoped that the development of AMD GPU will not be affected thereafter.

Conclusion:

In the past 50 years, AMD has left many excellent products for consumers, from the 80x86 era to K7, K8, and today's Ryzen, which are fully competitive in the market. The introduction of x86-64 architecture, APU heterogeneous computing, also let everyone see the vision of AMD. Despite the mistakes, we still insist on making changes and improving our products.

Throughout the development of AMD in the past 50 years, there is no gorgeous origin. Starting from dozens of employees, AMD has been active in the semiconductor and chip design industry through hard work. There are many talented people in this industry, but many of them have fallen. What remains must be the elite among the elites. AMD has also been developing continuously through 50 years of accumulation and innovation, gradually improving its competitiveness. Although there are fewer and fewer competitors around it, AMD has not only persisted in the cruel competition, but also has not retreated in front of its powerful competitors. Instead, AMD has constantly challenged and demonstrated its strong competitiveness.

This year is the 50th anniversary of AMD, a company can go through 50 years of development, enough to illustrate the company's heritage. I hope AMD can continue to do well in the future and bring better and excellent products to consumers.