The History of AMD CPUs: From Second Source to Chiplets

For decades, many people heard "PC processor" and thought of one name: Intel. But the x86 story is not a one-company story. Advanced Micro Devices, or AMD, began as a small firm that made chips for other companies. It spent decades in the shadow of a larger rival, won several legal and design contests, and in the late 2010s rebuilt itself around the Zen architecture and a chiplet way of making processors. This article follows that journey from the founding in 1969 to the processors now used in gaming PCs, servers and game consoles.

1969: Founding in Silicon Valley

AMD was founded in 1969 in Sunnyvale, California, by Jerry Sanders and a group of engineers and salespeople who had worked at Fairchild Semiconductor, the company that had helped develop the integrated circuit a few years earlier. In its early years the firm made logic chips and memory, many of them based on designs created by other companies. Competing with larger suppliers forced AMD to focus on price and on fast delivery, a habit that stayed with the company for decades.

The second-source era

In the 1970s, large electronics makers did not want to depend on a single supplier for an important chip. If that supplier ran short or raised prices, an entire product line could stop. The industry therefore used the second-source system: one company designed a chip, and another company was licensed to make the same chip so that customers had a choice. AMD became one of the most important second sources in the business.

In 1975 AMD began producing a version of the Intel 8080 processor called the Am9080. It had the same pinout and equivalent behaviour, so a system designed for the 8080 could use the AMD chip without changes to its circuit board. The work brought steady revenue and manufacturing experience at a time when AMD had no processor design of its own.

The IBM PC and the 8086 family

The Intel 8086 in 1978 and the cheaper 8088 that followed became the foundation of the x86 architecture. When IBM chose the 8088 for its PC in 1981, it also wanted a second source to protect its supply. Intel therefore licensed its designs to other manufacturers, and AMD was among the licensees. That placed AMD at the centre of the PC business, because the IBM PC became the model for an entire industry of compatible machines.

The licences were limited, however. Intel controlled the design, and the agreements specified which chips AMD could make and under what terms. This tension ran through the relationship for years. AMD needed the licences to grow, while Intel wanted to keep control of the architecture that generated the most profit.

The 286 and 386 disputes

In 1982 Intel introduced the 80286, a 16-bit processor with improved memory management. AMD obtained a licence and made its own version, the Am286. The agreement, however, contained terms about the use of microcode, the low-level instructions that tell a processor how to carry out its operations, and the two companies argued about those terms for years.

The Intel 386 of 1985 was a 32-bit processor that became the foundation of modern PC software. AMD eventually produced its own version, the Am386, after a long negotiation and a legal process that included litigation and arbitration. Both companies spent heavily on legal fees. The outcome was a cross-licensing settlement that set out which designs counted as independent work and which remained under the rights of the original owner.

The lesson AMD drew from these years was that intellectual property is central to the chip business. A company that does not own its core designs will always be exposed.

The 486 and the search for an identity

Intel's 80486 of 1989 combined the processor, a floating-point unit and a cache on one chip. It was clearly faster than the 386 and became the standard for high-performance desktops in the early 1990s. AMD produced its own compatible 486 chips, and its sales grew as the PC market expanded. Customers wanted faster machines at lower prices, and AMD offered an alternative that many system builders appreciated.

The K5: the first fully in-house x86 processor

In 1996 AMD released the K5, its first x86 processor designed entirely in-house. Until then, AMD had relied mostly on licensed designs and on reverse engineering of Intel products. The K5 proved that AMD's engineers could create a competitive chip from their own architecture. Its performance did not always match Intel's Pentium, but it offered good value and compatibility with the software of the time.

The K5 project faced several difficulties. Some design goals were ambitious, and the chip reached the market later than planned. The experience nevertheless taught AMD how to run large design projects, verify complex logic and coordinate with manufacturing partners.

NexGen and the K6

In early 1996 AMD acquired NexGen, a small company whose team had designed a processor called the Nx586. The NexGen engineers took a different approach: they translated x86 instructions into simpler internal operations, allowing the chip to process several operations in parallel. The Nx586 had strengths in some workloads, but it faced problems in manufacturing and in the market.

AMD used the NexGen team as the basis for the K6, released in 1997. The K6 was an efficient and inexpensive chip that competed directly with Intel's Pentium MMX. Its everyday performance was strong for its price, and it helped establish AMD's reputation as a maker of value processors.

The K6-2 and 3DNow!

The K6-2, released in 1998, added a set of instructions called 3DNow! to accelerate three-dimensional graphics and heavy floating-point work. The extensions were developed in response to the growing demand for 3D games and multimedia. Software had to be written to use them, and not every program did, but the K6-2 gave AMD a feature that Intel did not offer at the time.

The K6-2 also sold in large volumes in budget and mainstream markets, and it helped AMD take a larger share of the desktop business.

The Athlon: AMD becomes a performance leader

In 1999 AMD released the Athlon, designed by a team led by Dirk Meyer and other engineers who had worked on the K7 project. The Athlon was a major leap. It had a new architecture, a faster bus and a deeper pipeline, and it was the first AMD x86 processor to compete seriously with Intel's top products on raw performance.

In March 2000 AMD became the first company to ship a 1 GHz x86 processor to consumers, slightly ahead of Intel. The milestone received wide coverage and changed AMD's image from a value brand to a technical leader. Gamers and overclockers began to favour the Athlon.

Thunderbird, Palomino and the Athlon XP

The Athlon Thunderbird of 2000 added a level 2 cache on the processor die, improving performance, and it was followed by the Palomino core and the Athlon XP. The XP name was used to show that AMD's processors matched Intel's Pentium 4 in several tests even at lower clock speeds. Naming by performance rather than frequency was controversial, but it worked in the market.

The Athlon XP and Athlon MP were successful in desktops and servers. AMD enjoyed strong profitability during this period, but the rivalry with Intel remained intense.

The Athlon 64 and x86-64

The most important turning point of this era came in 2003, when AMD released the Opteron for servers and the Athlon 64 for consumers. Both implemented an extension to x86 called x86-64, officially named AMD64, which added 64-bit addressing and more registers. Intel had been developing a different 64-bit architecture with Hewlett-Packard, called IA-64 or Itanium, which was not compatible with existing x86 software. AMD's approach let a processor run both 32-bit and 64-bit programs, so customers could upgrade without rewriting their software.

The decision paid off. Intel eventually adopted the AMD64 instruction set under the name Intel 64, which confirmed that AMD had set the direction of mainstream x86 for the following decade.

The memory controller moves onto the chip

The Athlon 64 also moved the memory controller into the processor itself. Previously it sat on the motherboard as a separate chip, adding delay to every memory access. Once integrated, the processor fetched data from RAM more quickly, and performance improved in many applications. The idea became standard across the industry, and it also laid the groundwork for Infinity Fabric and the chiplet designs that came later.

The Athlon 64 X2 and the dual-core era

In 2005 AMD released the Athlon 64 X2, a dual-core processor with two cores on one chip. Intel had dual-core products at about the same time, but the Athlon 64 X2 was praised for its performance, its low power use and its compatibility with existing motherboards. It gave AMD a strong position in the consumer market during the early years of multi-core computing.

The ATI acquisition in 2006

In 2006 AMD completed the purchase of ATI Technologies, a major graphics chip company, for about 5.4 billion US dollars. The deal changed the company's direction. It brought a graphics team with decades of experience and a valuable patent portfolio. AMD believed that combining central and graphics processing would define the future of computing.

The acquisition was also costly. AMD took on significant debt and had to merge two engineering organisations with different cultures, and its financial position became more fragile in the following years. The idea of combining CPU and GPU, however, led to the accelerated processing units (APUs) that AMD began shipping in 2011, and later to the processors inside game consoles.

Phenom and an expensive mistake

The Phenom of 2007 was AMD's first native quad-core processor, with all four cores built on a single die rather than two dual-core chips joined together. The design was ambitious, but the first versions had a bug in the translation lookaside buffer (TLB) that affected stability and performance in some situations. AMD issued a patch and later revisions, but the company's reputation suffered.

The Phenom's clock speeds also trailed Intel's Core 2, launched in 2006 with a large performance lead. AMD's market share weakened, and the company concentrated on cost reduction and on its graphics business. The period taught AMD the value of thorough verification, realistic schedules and the cost of design errors.

Bulldozer and the APU

In 2011 AMD introduced the Bulldozer architecture, which grouped cores into modules in which two cores shared a floating-point unit. The idea was to add cores at a reasonable silicon cost, but performance per core came in below expectations, especially in floating-point heavy workloads. The FX line was widely criticised.

In the same period AMD shipped the Llano APU, which combined a CPU and a GPU on one chip for notebooks and budget desktops. The APU approach was the right direction in the long run, but the company had very limited funds at the time.

Hard years and the return

From 2012 to 2016 AMD cut costs sharply and reduced its workforce. Lisa Su became CEO in October 2014 and focused the company on core products and a new architecture. Jim Keller, a designer with a respected record on several earlier chips, joined in 2012 and led the development of the Zen architecture.

Zen was designed to deliver much better instructions per clock, to use a structure that scaled well to more cores, and to take advantage of TSMC's smaller manufacturing processes.

Zen and Ryzen

In March 2017 AMD launched the first Ryzen processors, based on Zen. They offered many more cores than competing chips at much lower prices, and they made AMD an attractive choice for gamers and content creators again. In June 2017 the EPYC server line followed, offering more cores and memory capacity than Intel's server chips at competitive prices.

Chiplets change the economics

Zen 2, released in 2019, was another major turning point. AMD split the processor into several small dies called Core Complex Dies (CCDs) and placed them around a central I/O die that contained the memory controller. The dies were linked by Infinity Fabric.

The benefit was yield. On TSMC's 7 nm process, small dies had a much higher share of defect-free chips than one large die would. AMD could therefore lower the cost per processor and reuse the same CCD across many products, from Ryzen to high-core-count EPYC. This gave the company a stronger position on price and core count at the same time.

Zen 3, 3D V-Cache and Zen 4

Zen 3, in 2020, reorganised the cache of the CCD so that each core could reach a larger pool of level 3 cache, which improved gaming performance noticeably. In 2022 AMD released the Ryzen 7 5800X3D, which stacked 3D V-Cache on top of the CCD and greatly increased the L3 cache. It became a favourite among gamers for its gains in games.

Zen 4, also in 2022, introduced the AM5 socket and DDR5 memory, a major platform change. Zen 5, released in 2024, improved performance per clock and computing throughput.

Game consoles and handheld devices

AMD has supplied chips for several game consoles. The PlayStation 4 and Xbox One of 2013 used the Jaguar architecture, and the PlayStation 5 and Xbox Series consoles of 2020 used Zen 2. These long-term contracts gave AMD a steady income. The Steam Deck, released in 2022, also uses an AMD APU that combines a Zen 2 CPU with an RDNA GPU, so AMD has a place in handheld gaming too.

Lessons from AMD's history

Conclusion

From a licensed copy of the 8080 in 1975 to the Athlon that reached 1 GHz, and on to the Ryzen and EPYC processors built from chiplets, AMD's history is a story of competing against much larger rivals and rebuilding itself several times. Each time you turn on a computer with an AMD processor, you benefit from those engineering decisions.