Amd: history, processor models and graphics cards

Advanced Micro Devices or also known as AMD is a semiconductor company based in Sunnyvale, California that is dedicated to the development of processors, motherboard chipsets, auxiliary integrated circuits, embedded processors, graphics cards, and related technology products for the consumption. AMD is the world's second-largest manufacturer of x86 processors, and the second-largest manufacturer of graphics cards for the professional and home industries.

Index of contents

The birth of AMD and the history of its processors

AMD was founded on May 1, 1969 by a group of Fairchild Semiconductor executives, including Jerry Sanders III, Edwin Turney, John Carey, Steven Simonsen, Jack Gifford, Frank Botte, Jim Giles, and Larry Stenger. AMD debuted in the logical integrated circuits market, to make the leap to RAM in 1975. AMD has always stood out for being Intel's eternal rival, currently they are the only two companies that sell x86 processors, although VIA is starting to put the leg back into this architecture.

We recommend reading our best PC hardware and component guides:

We also advise you to read our AMD zone:

• AMD Ryzen AMD Vega

AMD 9080, the beginning of the AMD adventure

Its first processor was the AMD 9080, a copy of the Intel 8080 that was created using reverse engineering techniques. Through it came other models such as the Am2901, Am29116, Am293xx used in various microcomputer designs. The next leap was represented by the AMD 29k, which sought to stand out for the inclusion of graphics, video and EPROM memory drives, and the AMD7910 and AMD7911, which were the first to support various standards both Bell and CCITT at 1200 baud half duplex or 300 / 300 full duplex. Following this, AMD decides to focus solely on Intel-compatible microprocessors, making the company a direct competitor.

AMD signed a contract with Intel in 1982 to license the manufacture of x86 processors, an architecture that is owned by Intel, so you need permission from it to be able to manufacture them. This allowed AMD to offer very competent processors and to compete directly with Intel, who canceled the contract in 1986, declining to reveal technical details of the i386. AMD appealed against Intel and won the legal battle, with the California Supreme Court forcing Intel to pay more than $ 1 billion in compensation for breach of contract. Legal disputes ensued and AMD was forced to develop clean versions of Intel's code, which meant that it could no longer clone Intel's processors, at least directly.

Following this, AMD had to put two independent teams to work, one gutting the secrets of AMD's chips, and the other creating its own equivalents. Am386 was the first processor of this new era of AMD, a model that arrived to fight the Intel 80386, and which managed to sell more than a million units in less than a year. After him came the 386DX-40 and Am486 which was used in numerous OEM equipment proving its popularity. AMD realized that it had to stop following in the footsteps of Intel or it would always be in its shadow, in addition to that it was increasingly complicated by the great complexity of the new models.

On December 30, 1994, the California Supreme Court denied AMD the right to use the i386 microcode. Following that, AMD was allowed to produce and sell Intel microcode 286, 386, and 486 microprocessors.

AMD K5 and K6, a new era for AMD

AMD K5 was the first processor created by the company from its foundations and without any Intel code inside. After this came the AMD K6 and the AMD K7, the first of the Athlon brand that hit the market on June 23, 1999. This AMD K7 needed new motherboards, since up to now it was possible to mount processors from both Intel and AMD on the same motherboard. This is the birth of Socket A, the first exclusive for AMD processors. On October 9, 2001, the Athlon XP and Athlon XP arrived on February 10, 2003.

AMD continued to innovate with its K8 processor, a major overhaul of the previous K7 architecture that adds 64-bit extensions to the x86 instruction set. This supposes an attempt on the part of AMD to define the x64 standard and to prevail to the standards marked by Intel. In other words, AMD is the mother of the x64 extension, which is used by all x86 processors today. AMD managed to turn the story around and Microsoft adopted the AMD instruction set, leaving Intel to reverse engineer the AMD spec. AMD managed for the first time to place itself ahead of Intel.

AMD scored the same against Intel with the introduction of the Athlon 64 X2 in 2005, the first dual-core PC processor. The main advantage of this processor is that it contains two K8-based cores, and can process multiple tasks at once, performing much better than single-core processors. This processor laid the foundation for the creation of current processors, with up to 32 cores inside. AMD Turion 64 is a low-power version intended for notebook computers, to compete against Intel's Centrino technology. Unfortunately for AMD, its leadership ended in 2006 with the arrival of the Intel Core 2 Duo.

AMD Phenom, its first quad-core processor

In November 2006 AMD announced the development of its new Phenom processor, which would be released in mid-2007. This new processor is based on the improved K8L architecture, and comes as an attempt by AMD to catch up with an Intel that had been put ahead again with the arrival of the Core 2 Duo in 2006. Faced with the new Intel domain, AMD It had to redesign its technology and make the leap to 65nm and quad-core processors.

In 2008 the Athlon II and Phenom II made in 45nm arrived, which continued to make use of the same basic K8L architecture. The next step was taken with the Phenom II X6, launched in 2010 and with a six-core configuration to try to stand up to the quad-core models from Intel.

AMD Fusion, AMD Bulldozer, and AMD Vishera

The purchase of ATI by AMD put AMD in a privileged position, as it was the only company that had high-performance CPUs and GPUs. With this, the Fusion project was born, which had the intention of uniting the processor and the graphics card in a single chip. Fusion introduces the need to integrate more elements within the processor, such as a 16-lane PCI Express link to accommodate external peripherals, this completely eliminates the need for a northbridge on the motherboard.

AMD Llano was the product of the Fusion project, the first AMD processor with an integrated graphics core. Intel had made progress in integrating with its Westmere, but AMD's graphics were far superior, and the only ones that allowed advanced 3D games to be played. This processor is based on the same K8L cores as the previous ones, and was the premiere of AMD with the manufacturing process at 32 nm.

The replacement of the K8L core finally came from the Bulldozer in 2011, a new K10 architecture manufactured at 32nm, and focused on offering a high number of cores. Bulldozer makes cores share elements for each of them, which saves space on silicon, and offers a greater number of cores. Multi-core applications were the future, so AMD tried to make a major innovation to get ahead of Intel.

Unfortunately, the performance of Bulldozer a was as expected, as each of these cores was much weaker than Intel's Sandy Bridges, so despite the fact that AMD offered twice as many cores, Intel continued to dominate with increasing strength.. It also did not help that the software was still unable to efficiently take advantage of more than four cores, which was going to be the advantage of Bulldozer, it ended up being its greatest weakness. Vishera arrived in 2012 as an evolution of the Bulldozer, although Intel was further and further away.

AMD Zen and AMD Ryzen, the miracle that few believed and turned out to be real

AMD understood the Bulldozer's failure and they made a 180º turn with the design of their new architecture, dubbed Zen. AMD wanted to wrestle with Intel again, for which it took the services of Jim Keller, the CPU architect who had designed the K8 architecture and who led AMD into its long time with the Athlon 64.

Zen abandons the Bulldozer design and refocuses on offering powerful cores. AMD gave way to a manufacturing process at 14nm, which is a giant step forward compared to Bulldozer's 32nm. These 14nm allowed AMD to offer eight-core processors, just like the Bulldozer, but much more powerful and capable of embarrassing an Intel that had rested on its laurels.

AMD Zen arrived in the year 2017 and represents the future of AMD, this year 2018 the second generation AMD Ryzen processors have arrived, and next 2019 the third generation arrive, based on an evolved Zen 2 architecture manufactured at 7 nm. We really want to know how the story continues.

Current AMD processors

AMD's current processors are all based on the Zen microarchitecture and Global Foundries' 14nm and 12nm FinFET manufacturing processes. The name Zen is due to a Buddhist philosophy originated in China in the 6th century, this philosophy preaches meditation in order to achieve illumination that reveals the truth. After the failure of the Bulldozer architecture, AMD entered a period of meditation on what its next architecture should be, this was what led to the birth of Zen architecture. Ryzen is the brand name of processors based on this architecture, a name that refers to the resurgence of AMD. These processors were launched last year 2017, all of them work with the AM4 socket.

All Ryzen processors include SenseMI technology, which offers the following features:

• Pure Power - Optimizes energy use by taking into account the temperatures of hundreds of sensors, allowing you to spread the workload without sacrificing performance. Precision Boost: This technology increases the voltage and the clock speed precisely in 25 Mhz steps, this allows optimizing the amount of energy consumed and offering the highest possible frequencies. XFR (eXtended Frequency Range) - Works in conjunction with Precision Boost to increase voltage and speed above the maximum allowed by Precision Boost, provided the operating temperature does not exceed the critical threshold. Neural Net Prediction and Smart Prefetch: They use artificial intelligence techniques to optimize workflow and cache management with a preload of smart information data, this optimizes access to RAM.

AMD Ryzen and AMD Ryzen Threadripper, AMD wants to fight Intel on an equal footing

The first processors to launch were the Ryzen 7 1700, 1700X, and 1800X in early March 2017. Zen was AMD's first new architecture in five years and demonstrated great performance from the start, even though the software was not optimized for its unique design. These early processors were highly proficient in gaming today, and exceptionally good at workloads that make use of a large number of cores. Zen represents an increase in the CPI of 52% compared to Excavator, the latest evolution of Bulldozer architecture. The IPC represents the performance of a processor for each core and for each MHz of frequency, the improvement of Zen in this aspect exceeded everything that had been seen over the last decade.

This massive improvement in the IPC allowed Ryzen's performance when using Blender or other software prepared to take advantage of all its cores out of about four times the performance of the FX-8370, AMD's previous top-of-the-range processor. Despite this huge improvement, Intel continued and continues to dominate in games, although the distance with AMD has been drastically reduced and is not important for the average player. This lower gaming performance is due to the internal design of Ryzen processors and their Zen architecture.

The Zen architecture is made up of what are called the CCX, they are quad-core complexes that share an 8 MB L3 cache. Most Ryzen processors are made up of two CCX complexes, from there AMD deactivates cores to be able to sell processors of four, six and eight cores. Zen has SMT (simultaneous multithreading), a technology that allows each core to handle two threads of execution. SMT makes Ryzen processors offer four to sixteen threads of execution.

The two CCX complexes of a Ryzen processor communicate with each other using Infinity Fabric, an internal bus that also communicates with each other the elements inside each CCX. Infinity Fabric is a highly versatile bus that can be used both to communicate elements of the same silicon pickup and to communicate two different silicon pickups with each other. Infinity Fabric has considerably higher latency than the bus used by Intel in its processors, this higher latency is the main cause of Ryzen's lower performance in video games, along with higher latency of cache and access to RAM compared to Intel.

Ryzen Threadripper processors were introduced in mid-2017, monsters that offer up to 16 cores and 32 processing threads. Each Ryzen Threadripper processor is made up of four silicon pads that also communicate through Infinity Fabric, that is, they are four Ryzen processors together, although two of them are deactivated and only serve as a support for the IHS. This turns Ryzen Threadrippers into processors with four CCX complexes. Ryzen Threadripper works with socket TR4 and has a four channel DDR4 memory controller.

The following table summarizes the characteristics of all first generation Ryzen processors, all manufactured at 14nm FinFET:

Segment	Cores (threads)	Mark and CPU model	Clock speed (GHz)	Cache	TDP	Plinth	Memory supported
Base	Turbo	XFR	L2	L3
Enthusiastic	16 (32)	Ryzen Threadripper	1950X	3.4	4.0	4.2	512 KB by nucleus	32 MB	180 W	TR4	DDR4 quad channel
12 (24)	1920X	3.5	32 MB
8 (16)	1900X	3.8	16 MB
performance	8 (16)	Ryzen 7	1800X	3.6	4.0	4.1	95 W	AM4	DDR4-2666 dual-channel
1700X	3.4	3.8	3.9
1700	3.0	3.7	3.75	65 W
Principal	6 (12)	Ryzen 5	1600X	3.6	4.0	4.1	95 W
1600	3.2	3.6	3.7	65 W
4 (8)	1500X	3.5	3.7	3.9
1400	3.2	3.4	3.45	8 MB
Basic	4 (4)	Ryzen 3	1300X	3.5	3.7	3.9
1200	3.1	3.4	3.45

This year 2018 the second generation AMD Ryzen processors have been launched, manufactured at 12 nm FinFET. These new processors introduce improvements focused on increasing the operating frequency and reducing latency. The new Precision Boost 2 algorithm and XFR 2.0 technology allow the operating frequency to be higher when more than one physical core is in use. AMD has reduced L1 cache latency by 13%, L2 cache latency by 24%, and L3 cache latency by 16%, causing the IPC of these processors to have increased by approximately 3% versus the first generation. In addition, support for the JEDEC DDR4-2933 memory standard has been added.

The following second-generation Ryzen processors have been released for now:

Model	CPU		Memory supported
Cores (threads)	Clock speed (GHz)	Cache	TDP
Base	Boost	XFR	L2	L3
Ryzen 7 2700X	8 (16)	3.7	4.2	4.3	4 MB	16 MB	105W	DDR4-2933 (Dual-channel)
Ryzen 7 2700	8 (16)	3.2	4	4.1	4 MB	16 MB	65W
Ryzen 5 2600X	6 (12)	3.6	4.1		3 MB	16 MB	65W
4.2 GHz
Ryzen 5 2600	6 (12)	3.4	3.8		3MB	16 MB	65W
3.9

Second-generation Ryzen Threadripper processors are expected to be announced this summer, offering up to 32 cores and 64 threads, unprecedented power in the home sector. For now only the Threadripper 2990X, the 32-core top of the range, is known. Its full features are still a mystery, although we can expect a maximum of 64MB of L3 cache as it will have all four silicon pads and eight active CCX complexes.

AMD Raven Ridge, the new generation of APUs with Zen and Vega

To these we must add the Raven Ridge series processors, also manufactured at 14 nm, and which stand out for including an integrated graphics core based on the AMD Vega graphics architecture. These processors include a single CCX complex in their silicon chip, so they offer a quad-core configuration all of them. Raven Ridge is AMD's most advanced family of APUs, it has come to replace the previous Bristol Ridge, which relied on excavator cores and a 28nm manufacturing process.

Processor	Cores / threads	Base / turbo frequency	L2 cache	L3 cache	Graphic core	Shaders	Graphics Frequency	TDP	RAM
Ryzen 5 2400G	4/8	3.6 / 3.9 GHz	2 MB	4 MB	Vega 11	768	1250 MHz	65W	DDR4 2667
Ryzen 3 2200G	4/4	3.5 / 3.7 GHz	2 MB	4MB	Vega 8	512	1100 MHz	65W	DDR4 2667

EPYC, AMD's new assault on servers

EPYC is AMD's current server platform, these processors are actually the same as Threadrippers, although they come with some improved features to meet the demands of servers and data centers. The main differences between EPYC and Threadripper, is that the former have eight memory channels and 128 PCI Express lanes, compared to Threadripper's four channels and 64 lanes. All EPYC processors are made up of four silicon pads inside, just like Threadripper, although here they are all activated.

AMD EYC is capable of outperforming Intel Xeon in cases where cores can operate independently, such as high-performance computing and big data applications. Instead, EPYC lags behind in database tasks due to increased cache latency and the Infinity Fabric bus.

AMD has the following EPYC processors:

Model	Socket Configuration	Cores / threads	Frequency	Cache	Memory	TDP (W)
Base	Boost	L2 (kB)	L3 (MB)
All Core	Max
Epyc 7351P	1 P	16 (32)	2.4	2.9	16 x 512	64	DDR4-2666 8 Channels	155/170
Epyc 7401P	24 (48)	2.0	2.8	3.0	24 x 512	64	155/170
Epyc 7551P	32 (64)	2.0	2.55	3.0	32 x 512	64	180
Epyc 7251	2 P	8 (16)	2.1	2.9	8 x 512	32	DDR4-2400 8 Channels	120
Epyc 7281	16 (32)	2.1	2.7	2.7	16 x 512	32	DDR4-2666 8 Channels	155/170
Epyc 7301	2.2	2.7	2.7	16 x 512	64
Epyc 7351	2.4	2.9	16 x 512	64
Epyc 7401	24 (48)	2.0	2.8	3.0	24 x 512	64	DDR4-2666 8 Channels	155/170
Epyc 7451	2.3	2.9	3.2	24 x 512	180
Epyc 7501	32 (64)	2.0	2.6	3.0	32 x 512	64	DDR4-2666 8 Channels	155/170
Epyc 7551	2.0	2.55	3.0	32 x 512	180
Epyc 7601	2.2	2.7	3.2	32 x 512	180

The adventure with graphics cards Is it up to Nvidia?

AMD's adventure in the graphics card market begins in 2006 with the purchase of ATI. During the early years, AMD used designs created by ATI based on the TeraScale architecture. Within this architecture we find the Radeon HD 2000, 3000, 4000, 5000 and 6000. All of them were making small improvements continuously to improve their capabilities.

In 2006 AMD took a big step forward with the purchase of ATI, the world's second largest graphics card maker, and a direct rival to Nvidia for many years. AMD paid $ 4.3 billion in cash and $ 58 million in shares for a total of $ 5.4 billion, completing the action on October 25, 2006. This operation put AMD's accounts in red numbers, so The company announced in 2008 that it was selling its silicon chip manufacturing technology to a multi-billion dollar joint venture formed by the Abu Dhabi government, this sale is what led to the birth of the current GlobalFoundries. With this operation, AMD ditched 10% of its workforce, and was left as a chip designer, with no manufacturing capacity of its own.

The following years followed AMD's financial problems, with further downsizing to avoid bankruptcy. AMD announced in October 2012 that they planned to lay off an additional 15% of its workforce to reduce costs in the face of declining sales revenue. AMD acquired the low-power server maker SeaMicro in 2012 to regain lost market share in the server chip market.

Graphics Core Next, the first 100% AMD graphics architecture

The first graphics architecture developed from the ground up by AMD is the current Graphics Core Next (GCN). Graphics Core Next is the code name for a series of microarchitectures and a set of instructions. This architecture is the successor to the previous TeraScale created by ATI. The first GCN-based product, the Radeon HD 7970 was released in 2011.

GCN is a RISC SIMD microarchitecture that contrasts with TeraScale's VLIW SIMD architecture. GCN requires many more transistors than TeraScale, but offers advantages for GPGPU calculation, makes the compiler simpler, and should also lead to better resource utilization. GCN is manufactured in the 28 and 14nm processes, available on select models from the Radeon HD 7000, HD 8000, R 200, R 300, RX 400 and RX 500 series of AMD Radeon graphics cards. The GCN architecture is also used in the APU graphics core of PlayStation 4 and Xbox One.

To date, the family of microarchitectures that implement the instruction set called Graphics Core Next has seen five iterations. The differences between them are quite minimal and do not differ too much from each other. One exception is the fifth-generation GCN architecture, which has greatly modified stream processors to improve performance and supports simultaneous processing of two lower precision numbers instead of a single higher precision number.

The GCN architecture is organized into compute units (CU), each of which combines 64 shader processors or shaders with 4 TMUs. The computing unit is separate from, but is powered by, the Processing Output Units (ROPs). Each Compute Unit consists of a Scheduler CU, a Branch & Message Unit, 4 SIMD Vector Units, 4 64KiB VGPR files, 1 scalar unit, a 4 KiB GPR file, a local data quota of 64 KiB, 4 texture filter units, 16 texture recovery load / storage units and a 16 kB L1 cache.

AMD Polaris and AMD Vega the newest from GCN

The last two iterations of GCN are the current Polaris and Vega, both manufactured at 14nm, although Vega is already making the leap to 7nm, with no commercial versions yet for sale. GPUs from the Polaris family were introduced in the second quarter of 2016 with AMD Radeon 400 series graphics cards. Architectural improvements include new hardware programmers, a new primitive discard accelerator, a new display driver, and an updated UVD that can decode HEVC at 4K resolutions at 60 frames per second with 10 bits per color channel.

AMD began releasing details of its next generation of GCN architecture, called Vega, in January 2017. This new design increases the instructions per clock, achieves higher clock speeds, offers support for HBM2 memory and a larger memory address space. Discrete graphics chipsets also include a high bandwidth cache controller, but not when they are integrated into APUs. Shaders are heavily modified from previous generations to support Rapid Pack Math technology to improve efficiency when working in 16-bit operations. With this, there is a significant performance advantage when lower precision is accepted, for example, processing two medium precision numbers at the same speed as a single high precision number.

Vega also adds support for new Primitive Shaders technology that provide more flexible geometry processing and replace vertex and geometry shaders in a render pipe.

The following table lists the characteristics of current AMD graphics cards:

CURRENT AMD GRAPHICS CARDS
Graphic card	Compute Units / Shaders	Base / Turbo Clock Frequency	Amount of memory	Memory interface	Memory type	Memory bandwidth	TDP
AMD Radeon RX Vega 56	56 / 3, 584	1156/1471 MHz	8 GB	2, 048 bits	HBM2	410 GB / s	210W
AMD Radeon RX Vega 64	64 / 4, 096	1247/1546 MHz	8 GB	2, 048 bits	HBM2	483.8 GB / s	295W
AMD Radeon RX 550	8/512	1183 MHz	4GB	128 bit	GDDR5	112 GB / s	50W
AMD Radeon RX 560	16 / 1, 024	1175/1275 MHz	4GB	128 bit	GDDR5	112 GB / s	80W
AMD Radeon RX 570	32 / 2, 048	1168/1244 MHz	4GB	256 bits	GDDR5	224 GB / s	150W
AMDRadeon RX 580	36/2304	1257/1340 MHz	8 GB	256 bits	GDDR5	256 GB / s	180W

So far our post about everything you need to know about AMD and its main products today, you can leave a comment if you have something else to add. What do you think of all this information? You need help to mount your new PC, we help you in our hardware forum.