Intel's CPU history 1971 - 2008

This article describes Intel's processors to date. It provides an up-to-date overview of all Intel processors and a bit of history about the computing world.

This article was initially created in 2008 with my Wonderword and was the first to include my bookmarking system.

Below is a list of changes:

2008-10-09

Original version: covers from Intel Pre-x86 Architecture to end of Intel Core Microarchitecture.

2009-02-20

Added Intel Atom Microarchitecture and Intel Nehalem Microarchitecture.

2011-01-05

Added Intel Sandy Bridge Microarchitecture.

2015-01-18

Given a site modernisation to follow the site design

Intel Pre-x86 Architecture

1971: Intel 4004 Microprocessor

Designed originally for calculators, this processor was one of the major developments in the history of Intel. As Intel's first CPU, it was an impressive finish.

• Intel Pre-x86 Architecture
• 108KHz - 740KHz
• 2,300 transistors
• 10um
• No cache
• 4-bit

1972: Intel 8008 Microprocessor

Designed as the first home processing unit, the 8008 performed "twice as fast" as the 4004. The main difference is the clock speed and the number of transistors. At this point, Intel was not focused on improving the fabrication technology.

• Intel Pre-x86 Architecture
• 200KHz
• 3,500 transistors
• 10um
• No cache
• 8-bit

1974: Intel 8080 Microprocessor

The 8080 was the first computer processor for hobbyists. This CPU meant that people interested in this could easily build their own computers. The 8080 did not come cheap, however.

• Intel Pre-x86 Architecture
• 2MHz
• 6,000 transistors
• 6um
• No cache
• 8-bit

Intel x86-16 Architecture

1978: Intel 8086 - 8088 Microprocessor

The processor was sold to IBM as part of their new Personal System range. Its success was due to the success of the PS system. The problem was that Microsoft refused to program for the processor because of a lack of compatibility, eventually losing IBM and Microsoft's alliance.

• Intel x86 - 16 Architecture
• 5MHz - 10MHz
• 29,000 transistors
• 3.2um
• No cache
• 16-bit

1982: Intel 286 Microprocessor

The start of big success for Intel started here. This CPU meant that computer users had backward compatibility with previous software written for previous processing units. Intel was developing processors very quickly, and the size of a single feature in the CPU was halving every time.

• Intel x86 - 16 Architecture
• 6MHz - 13MHz
• 134,000 transistors
• 1.5um
• No cache
• 16-bit

Intel x86-32 Architecture

1985: Intel 386 Microprocessor

This CPU supported multitasking for the first time, allowing the processor to focus its power elsewhere whenever possible. It was also the first 32-bit CPU, a technology that is still very much used today. This is why the 32-bit technology used is often called x86.

• Intel x86 - 32
• 16MHz - 33MHz
• 275,000 transistors
• 1.5um - 1um
• 16kB - 64kB cache
• 32-bit

1989: Intel 486 Microprocessor

The 486 was a huge change in the architecture of previous processors - it could do mathematics in a separate section of the processor. This, therefore, reduces the workload on the main CPU. The 486 was the first to cope with high-end operating systems requiring coloured displays. It also was the first CPU to support the Graphical User Interface desktop in this sense. It could also address up to 4GB of memory, which can be used by most 32-bit operating systems of today.

• Intel x86 - 32
• 25MHz - 50 MHz
• 1.2 million transistors
• 1um - 0.6um
• 8KB cache
• 32-bit

Intel P5

1993: Intel Pentium Processor

The Pentium was the first step in the major Intel marketing.

The Pentium's P5 architecture was the fifth major development in microarchitectures and slightly led to the later development of the P6 architecture. The Pentium was compatible with Windows 3.1 and Windows 95 and 98. Its speed was a serious improvement on the previous processors.

The Pentium also established Intel's range of processors, which would continue for years to come until its abandonment in 2009.

This CPU also introduced the MMX instruction set, which made it competitive against VIA and (the smaller competitor of the time) AMD chips. The Pentium did not have a direct Celeron sibling; instead, it was a CPU in architecture alone. It was also thought that the Intel P5 architecture led to Intel's Atom architecture, as the energy-saving features in the Atom were also found in the Pentium. However, this does not make the Atom a direct derivative from the Pentium; it is more of a discrete derivation.

•  Intel P5 Architecture
• 60MHz - 300MHz
• 3.1 million transistors
• 0.8um - 0.25um
• 512KB L2 (off-die) cache
• 32-bit
• Codenames: Tillamook
• 50 - 66MHz Front Side Bus

Intel P6

1995: Intel Pentium Pro

The first in the server line for Intel, the Pentium Pro, had a redesigned instruction set known to most compilers as the Pentium Pro instruction set.

The major differences meant it could not run older non-Pentium Pro compiled programs and vice versa. This server line featured high cache levels of up to 1MB. However, the most important feature of this CPU was the new architecture based on the P6 architecture. The P6 architecture also led to the development of many future microarchitectures. To get a more detailed analysis of how successful the P6 architecture was, look at the roadmap at the end of the document.

• Intel P6 Architecture
• 150MHz - 200MHz
• 5.5 million transistors
• 0.50um - 0.35um
• 1MB L2 (off-die) cache
• 32-bit
• Codenames: Tillamook
• 60 - 66MHz Front Side Bus

1997: Intel Pentium II Processor

The Pentium II also featured the MMX technology in the later Pentium I processors. It was the first processor capable of high-quality (320 x 240) video decoding.

The processor was also a mainstream processor which brought high-end multimedia to the home computer. The Pentium II also had a brother - its high-end derivative, the Xeon counterpart.

The Pentium II had heat issues that were well and truly solved while keeping noise and power levels down. There was also a later mobile version of this CPU called Tonga and later Dixon. Both of these CPUs tried to maintain long battery life in laptop computers using them. My first computer family computer was a Pentium II.

• Intel P6 Architecture
• 200MHz - 300MHz
• 7.5 million transistors
• 0.35um - 0.25um
• 512KB L2 (off-die) cache
• 32-bit
• Codenames: Klamath, Deschutes, Tonga and Dixon
• 66 - 100MHz Front Side Bus

1997: Intel Celeron (based on Pentium II)

The initial release of the Celeron in 1997 was based on the Pentium II processor.

The first Celeron CPU was slightly more clocked than the Pentium seen previously but without any cache memory, and it was named Covington. Despite being higher clocked than the Pentium, it was still outperformed by it due to its lack of L2 cache.

The Celeron quickly gained a very bad name for the brand, yet it still sold very well since it could easily be overclocked and because it was cheap to buy Celeron-based computer systems. Intel released its next Celeron-based CPU called Mendocino, the next Pentium II-based Celeron. This CPU was a very powerful performer, mainly because it had an on-die Level 2 cache, which no other CPU had - including its "higher-end" brother, the Pentium II. The main problem with this processor for Intel was that it could easily be overclocked very well and outperform their Pentium II, which lost sales for the Pentium II.

The final CPU released in this family would be a 533MHz clock speed with a 66MHz front-side bus.

• Intel P6 Architecture
• 200MHz - 533MHz
• 7.5 million transistors
• 0.35um - 0.25um
• 256KB L2 cache
• 32-bit
• Codenames: Covington and Mendocino
• 66 - 100MHz Front Side Bus

1997: Intel Pentium II Xeon

The Pentium II Xeon was a high-end CPU with 512KB up to 2MB of the cache. The Xeon also brought along more improvements than the Pentium Pro could not, such as the energy efficiency breakthrough with the new fabrication techniques.

The new Xeon range would continue for years and is Intel's longest-reigning range. As much as this Xeon would replace the Pentium Pro, it did not do so for quite some time, but it did bring the advantage that when compiling for a Xeon-based machine, the code would not need the Pentium Pro-based compiler to create the code functionally.

• Intel P6 Architecture
• 200MHz - 300MHz
• 7.5 million transistors
• 0.35um
• 512KB - 2MB L2 cache
• 32-bit
• Codenames: Deschutes
• 100MHz Front Side Bus

1999: Intel Pentium III Processor

The development of the Pentium III was a big shock to the market. This processor seriously increased the clock rate, initially to 650MHz and later to 1.4GHz.

Once the Gigahertz range of this processor came along, the MHz range seemed old-fashioned. The Pentium III was also the first major commercial success on television, with its wide range of adverts featuring the "blue men." Also to come along with the Pentium III was the large 512KB cache seen previously on the Pentium and Pentium II processors, but also a breakthrough in energy efficiency, allowing just 32W to the highest-end CPU.

The CPU was manufactured with the latest 0.25um manufacturing technology but improved to 130nm. It was also able to run in both Slot 1 and Socket 370, using an adaptor that converts Slot 1 to Socket 370, which makes it one of the only CPUs to run in both a Slot and a Socket. The Pentium III was also added to the SSE instruction set seen in most current CPUs. The Pentium III also had a mobile and server counterpart.

However, by this point, the P6 architecture looked to be stagnating regarding clock speed, and the Tualatin core was very difficult to improve in terms of performance.

• Intel P6 Architecture
• 650MHz - 1400MHz (1.4GHz)
• 9.5 million transistors
• 0.25um - 0.13um (130nm)
• 256KB - 512KB cache
• 32-bit
• Codenames: Katmai, Coppermine, Coppermine-T and Tualatin
• 100 - 133MHz Front Side Bus

1999: Intel Celeron (based on Pentium III)

The next Celeron in line was the Pentium III edition. This Celeron would be the final P6 architecture-based desktop CPU ever released by Intel. It once again had half the L2 cache of its Pentium counterpart (which made room for the mistake again that if when a Pentium III was produced and half the cache or more failed to function, Intel could claim it was a Celeron). It was initially released with a 128KB cache, half of the original Pentium III processors in the low end, making it a quarter of the maximum cache found on the high-end Pentium IIIs.

The first Celeron was based on the Coppermine-based Pentium III CPU. Although the energy efficiency in comparison to performance in the early Coppermine Pentium IIIs was a problem, the Coppermine was still a very good performer. The Coppermine processor was later replaced with a double-cache system, equivalent to a lower-end Tualatin. This would now feature 256KB of cache, half of that found on the Pentium III model, but still enough to make it a solid performer.

Intel made a serious movement towards energy efficiency at this point in time with the Celeron Tualatin.

• Intel P6 Architecture
• 650MHz - 1400MHz (1.4GHz)
•  9.5 million transistors
• 0.25um - 0.13um (130nm)
• 128KB - 256KB cache
• 32-bit
• Codenames: Coppermine-128 and Tualatin-256
• 100 - 133MHz Front Side Bus

1999: Intel Pentium III - M

The Pentium III-M was the final notebook-based CPU based on the P6 architecture. It had some major differences from the desktop version. The first was the feature of Intel SpeedStep. This technology allowed the dynamic changing of the voltage (VCore) and the CPU frequency to reduce power usage and, in turn, make the battery last longer.

The CPU could only support up to 1.33GHz, unlike the later Tualatin desktop chip, which could go as far as 1.4GHz. The mobile edition also featured a slower front-side bus of only 100MHz instead of 133MHz until it was finally upgraded to the Pentium III-M, which featured a 133MHz front-side bus and 512KB cache.

• Intel P6 Architecture
• 650MHz - 1330MHz (1.33GHz)
• 9.5 million transistors
• 0.25um - 0.13um (130nm)
• 256KB - 512KB cache
• 32-bit
• Codenames: Coppermine and Tualatin
• 100 - 133MHz Front Side Bus

1999: Intel Pentium III Xeon

The second Xeon in the family was the Pentium III edition of it.

Based on the Katmai and later the Coppermine cores, it had a single pumped 100MT/s front side speed.

However, the P6 architecture could not compete with its opposition in AMD's line (now quite a big threat to Intel) with their new K6 processors.

Also, the Pentium III Xeon may have been a good performer for workstations due to its low power consumption, but its performance was still not improving directly. Times were changing, however, and competitors were using the Megahertz Myth to draw more customers, which Intel responded to with the later NetBurst architecture.

• Intel P6 Architecture
• 650MHz - 1400MHz (1.4GHz)
• 9.5 million transistors
• 0.25um - 0 .18um (180nm)
• 1MB and 2MB cache
• 32-bit
• Codenames: Tanner and Cascades
• 100MHz Front Side Bus

Intel NetBurst

2000: Intel Pentium 4

With the latest and greatest milestone achievements being achieved mainly by Intel with the Pentium series up to the Pentium III, the Pentium 4 aimed to do precisely the same.

The Pentium 4 radically changed the architecture, removing the old P6 architecture, which looked to be going no further regarding clock speed.

The new architecture was to be called NetBurst, and it was not in the slightest bit related to any other architecture made by Intel—it was entirely new. This architecture was so different from all other architectures that it was one of the most radical changes Intel ever brought to its CPUs. The idea was that performance would increase if they pushed for a higher pipeline. This is precisely what happened in Pentium 4.

The first of these CPUs was released in 2000 with the Willamette core, which has a 180um fabrication process and speeds up from 1.5GHz to 2GHz with base speeds of 100MHz. The early Pentium 4 CPUs had 15-stage pipelines, whilst later models featured as long as 31-stage pipelines. The higher pipelining means that the processors can perform more calculations and process more instructions per clock cycle, which gives the system a higher clock rate. However, it can also reduce the performance as if an instruction has been incorrectly predicted using the branch predictor, and the whole pipeline must be flushed, causing it to work from the top to the bottom again. Intel set the milestone with this CPU again by breaking the 2GHz barrier.

The later versions of the Pentium 4 were the Northwood core and the later Prescott core, which was available in both socket 478 and socket LGA775.

• Intel NetBurst Architecture
• 1.5GHz - 3.4GHz
• 42 million transistors
• 0.18um - 90nm
• 256KB - 2MB cache
• 32-bit or 64-bit
• Codenames: Willamette, Northwood and Prescott
• 400-800MHz Front Side Bus

2001: Intel Xeon (180nm - 130nm)

The NetBurst Pentium 4 showed that the increase in clock speed was not necessarily making it faster but more popular.

The Xeon was still based on the P6 architecture, which could not last much longer. Therefore, changes had to be made. The NetBurst Xeon was released under the name Foster, and it had a transistor shrink of 180nm found in the Cascades Pentium III Xeon. It was based on the Willamette core.

But in the Intel tick-tock cycle, the Prestonia was released with a shrink of fabrication down to 130nm at about the same time as the Northwood core was released. The bus speed using QDR (Quad Data Rate) was boosted to 533MHz as opposed to the 400MHz seen previously. The Prestonia also featured 512KB of L2 cache and Hyper-Threading technology. The next line was the Gallatin core, which had an improved cache memory and L3 sizes of up to 4MB.

• Intel NetBurst Architecture
• 1.4GHz - 2.0GHz
• 42 million transistors
• 0.18um - 130nm
• 512KB - 1MB L2 cache or 2MB- 4MB L3 cache
• 32-bit
• Codenames: Foster, Prestonia and Gallatin
• 400-533MHz
• Front Side Bus

2001: Intel Celeron (based on Pentium 4)

Just a couple of months before the Northwood core was released in the Pentium 4 market, the Willamette Celeron was released.

The Willamette Celeron was based on the 256KB Pentium 4, meaning it would have only 128KB of Level 2 cache (given that the Celeron was given half the cache of the mainstream CPU due to defects). The Willamette Celeron was the first Celeron in the NetBurst architecture. Speeds on this early NetBurst CPU ranged from 1.7GHz to 1.8GHz, giving little variety to the CPU.

Although the Pentium 4 Willamette did not outperform the Pentium III, the Celeron outperformed the Tualatin Pentium III Celerons by a very small margin. Once the Northwood core was released, Intel released a Celeron based on this core. The Northwood core did not outperform the Willamette as far as Intel thought it would. The new core also had heat dissipation problems because it was beginning to require more and more to keep it cool. However, unlike the Pentium 4, Intel decided to rebrand the new Celerons when released.

• Intel NetBurst Architecture
• 1.5GHz - 2.8GHz
• 42 million transistors
• 0.18um - 130nm
• 128KB - 256KB cache
• 32-bit or 64-bit
• Codenames: Willamette and Northwood
• 400-533MHz Front Side Bus

2002: Intel Mobile Pentium 4 M

This CPU had a strangely repetitive name - Mobile Pentium 4 M. Heat dissipation had always been a problem in Intel Pentium 4 CPUs. Still, the mobile editions were hit the hardest for some strange reason. Desktop CPUs consumed more power than any other CPU and produced around 80 degrees of heat. The mobile editions were suffering worse, with up to 100 degrees.

This mobile edition of the Pentium 4 had a considerable thermal design power of 35W, considering it operated only at 2GHz. This was due to a halved voltage. As power is proportional to the voltage, so does the power (P = CV2f) if the voltage drops. The Pentium 4 M was built on 130nm technology, so the processor was generally quite cool, and the 100-degree margin was rarely met, save in higher-end versions.

• Intel NetBurst Architecture
• 1.6GHz - 3.4GHz
• 42 million transistors
• 130nm
• 512KB - 1MB cache
• 32-bit
• Codenames: Northwood and Prescott
• 400-533MHz Front Side Bus

2003: Intel Pentium 4 with Hyper-Threading

The Pentium 4 did have problems. The early Willamette core was a reasonably cool and normal-temperature processor with a thermal design power of 60W. When its successor, Northwood, came along, heat dissipation was a problem. However, Northwood brought out the Mobile Pentium 4 and, in turn, the Mobile Pentium 4-M. The naming conventions used by Intel made it complex to figure out precisely what you were buying, and Intel lost money.

Intel later broke the 3GHz barrier with the Pentium 4 Northwood. The thermal design power levels went up to 90W for a 3.06GHz edition of the Northwood core. Serious heat dissipation issues also occurred at up to 80 degrees for the 3.06GHz core.

Intel had nothing to be proud of when it increased its pipelining to increase clock speed. This is considered the king example of the megahertz myth.

However, the main improvement with the Northwood core was Hyper-Threading, i.e., the ability to process two threads simultaneously per clock cycle. Hyper-threading does this by adding two arithmetical processing cores (ALUs) to the CPU and running processes across both at the same time (simultaneously). This was a significant step towards multicore computing.

However, with Hyper-Threading comes double power consumption and a ~25%-40% increase in heat dissipation. However, Intel felt that it had to keep going, and when Prescott was released with Hyper-Threading, the heat dissipation had passed the 100-degree mark. Power consumption was around 115W, the highest seen in a mainstream processor. Prescott was later released with the new LGA775 socket, later used in the Core series. The aim was to reduce the power and heat dissipation, but it, in fact, increased both. However, the heat dissipation was solved on the plus side due to the LGA775's better heat sink collection and alignments. The Prescott core featured 90nm fabrication and was the last official Pentium 4 for the mainstream market.

Later, Intel released the Cedar Mill Pentium 4, which could clock well past 8GHz but would need extreme cooling. Cedar Mill was created on the 65nm fabrication process. Most gamers, however, did not like the Pentium 4 CPU and did not feel it had a competitive edge over its competitors (mainly AMD's Athlon 64 range). As well as the shift towards the LGA775 socket, the Pentium 4 started the first of the 90nm fabrication, which started Intel's fabrication technology shrink to lead to higher energy efficiency. Intel also developed 64-bit in the Pentium 4, another move that meant the Pentium 4 was well respected. However, the Pentium 4 reached 3.8GHz before giving in to the Pentium D.

• Intel NetBurst Architecture
• 2.6GHz - 3.4GHz
• 42 million transistors
• 0.18um - 90nm
• 512KB - 2MB cache
• 32-bit or 64-bit
• Codenames: Northwood, Prescott, Prescott 2M and Cedar Mill
• 800MHz Front Side Bus

2003: Intel Pentium 4 Extreme Edition

By this point, Intel knew that heat dissipation was a problem with the NetBurst architecture, and their processors were not selling as well as they should have because AMD's counterpart was outselling and outperforming it.

Gamers were not interested in the mainstream Pentium 4, even with Hyper-Threading. This is where Intel's Extreme series was supposed to come in.

The Intel Pentium 4 Extreme Edition was a Hyper-Threading-enabled processor. It was based on much of Intel's Gallatin Xeon range. The processor had a 2MB Level 3 cache and 1MB Level 2 Cache. However, this was a major problem with heat dissipation in that it increased it several-fold. Therefore, the Pentium 4 Extreme Edition had the highest thermal design power of up to 135W.

The processor was a large failure for Intel, as they were retailing it at approximately $1000. This, however, was the start of a new era of Intel's Extreme Edition series CPUs. The Pentium 4 Extreme Edition Prescott 2M XE also had 2MB of L2 cache, a boost for most applications. Still, it had too high latency and eventually slowed the system down. The Pentium 4 Extreme Edition was initially designed as a high-end video editing and gaming edition. Still, its ability to cope with all of these applications was outdated due to higher demand from both, such as high-specification games (as the Pentium 4 only supported as far as PC2-5300 RAM) and high-definition video, which meant that a push for new technology was required.

That technology would be multi-core computing.

• Intel NetBurst Architecture 2.6GHz - 3.8GHz
• 55 million transistors 130nm - 90nm
• 512KB - 2MB L2 cache and 2MB L3 cache
• 32-bit or 64-bit
• Codenames: Gallatin XE and Prescott 2M XE
• 800MHz - 1066MHz Front Side Bus

2004: Intel Celeron D

The early Celeron D processors were based on the Prescott core using the Socket 478 package. The Celeron D was based on Prescott, which was already experiencing significant problems with heat and power consumption. The Celeron was pretty much the same in this sense. However, due to the halved Level 2 cache, the total power consumption during usage was approximately half that of the Prescott Pentium 4.

Some people believed the Celeron D was a mistake - adding an additional CPU to Intel's already crowded lineup.

Eventually, the Celeron D changed to LGA775, as the Pentium 4 did, but this did not improve the CPU's power usage. The Celeron D came later in a Cedar Mill package, produced on the 65nm fabrication. Like the Pentium 4 Cedar Mill, the Cedar Mill Celeron D was very overclockable, even as far as 8GHz, with extreme cooling. The Cedar Mill edition of the Celeron D also had the double cache memory of its predecessor.

• Intel NetBurst Architecture
• 1.5GHz - 3.6GHz
• 62 million transistors
• 0.18um - 130nm
• 256KB - 512KB cache
• 32-bit or 64-bit
• Codenames: Prescott and Cedar Mill
• 533MHz Front Side Bus

2004: Intel Xeon (90nm)

The Xeon based on the Prescott core was released in 2004. They were designed using a 90nm fabrication process. These were the first Intel 64 (IA-64) Xeon CPUs released.

Shortly before this release, Intel had just released the Itanium and the Itanium 2 processor in a joint partnership with Hewlett Packard. The result was the IA-64 architecture, an extension to the x86 architecture, which resulted in the 64-bit computer industry known as x86-64. The later Xeon processor based on Prescott's NetBurst was a slightly higher performer than earlier but had a higher thermal design power. These later CPUs dropped direct support for the legacy PCI-X bus in favour of PCI-E and dropped legacy IDE for the more modern SATA and SAS connectors. These Xeon processors were also the first to feature 1.8V DDR2 SDRAM.

• Intel NetBurst Architecture
• 2.6GHz - 3.8GHz
• 80 million transistors
• 90nm 1MB - 2MB L2 cache
•  64 bit
• Codenames: Nocona and Irwindale
• 667MHz - 800MHz
• Front Side Bus

2005: Intel Pentium Extreme Edition

The next processor developed by Intel continues on the same NetBurst architecture but introduces multiple cores for the first time. It is practically two Pentium 4 Extreme Edition CPUs on one chip manufactured under 90nm and later 65nm. The processor brought dual-core technology to the market but also kept the Hyper-Threading technology.

The Pentium Extreme Edition is based on the original Pentium 4 and uses 90nm and 65nm technology. It had a 130W thermal design power, the standard for an 'Extreme Edition' CPU, and is generally considered acceptable for such a processor. The early Smithfield XE (two Prescott cores on the one chip in a non-monolithic design) processors were 90nm, but the later Presler XE (two Cedar Mill cores on the one chip) chips were produced with 65nm fabrication technologies. The Pentium Extreme Edition had its core frequency unlocked as well. This powerful CPU looked to many operating systems as though it had four cores or the system had four processors, which was caused by the Hyper-Threading processing capabilities, meaning it had two physical cores within each of those two logic cores (provided by Hyper-Threading). However, the main issue found with this processor was its lack of mainstream ability. It may have been suitable for gamers, but pricing meant it was too expensive for the mainstream. Intel had to develop a dual-core mainstream CPU before its competitors did.

• Intel NetBurst Architecture
• 3.2GHz - 3.8GHz
• 125 million transistors
• 90nm - 65nm
• 512KB - 4MB cache
• 32-bit or 64-bit
• Codenames: Gallatin XE and Prescott 2M XE
• 800MHz - 1066MHz Front Side Bus

2005: Intel Xeon (90nm - 65nm)

The NetBurst Xeon's final edition had the highest power consumption, requiring 130W but running two cores. The idea was that it would run two cores (almost identical to the Pentium 4 chip) on the same chip. Paxville received the worst reputation for any Intel CPU due to its high power consumption yet low performance. The performance per watt ratio was exceptionally low on this CPU.

Dempsey was the final NetBurst Xeon to be released. It was based on NetBurst, but it was a 65nm version that allowed for a 1066MHz FSB clock push. It was again 64-bit and had a 4MB L2 cache. The dual-core idea worked well in Intel CPUs' Xeon and desktop lines.

• Intel NetBurst Architecture
• 2.8GHz - 3.8GHz
• 130 million transistors
• 90nm - 65nm
• 2MB - 4MB L2 cache
• 32-bit or 64-bit
• Codenames: Paxville and Dempsey
• 800MHz - 1066MHz Front Side Bus

2005: Intel Pentium D

Intel's NetBurst architecture was slowly beginning to disappear. It was obvious that NetBurst was ready for replacement. Despite this, NetBurst was moving towards a better future in its last moments. Intel had achieved high clock frequencies of 3.8GHz, the Pentium Extreme Edition was now a dual-core Hyper-Threaded CPU, and the Celeron D had covered the low-end market. But, the Pentium D was still to come.

The Pentium D is, effectively, two Pentium 4 processors on one chip, similar to the Pentium Extreme Edition. It is, once again, based on the NetBurst microarchitecture. While dual-core, its Hyper-Threading capabilities have been turned off, leaving that for the Pentium Extreme Edition and the unlocked core multiplier. The Pentium D was a big hit on its debut. Its 70 million transistors over the Pentium 4 increase from the transistor shrink to 65nm. The Pentium D had a TDP of 95W or 130W, depending on its model and frequency. Pentium D was the first mainstream dual-core CPU to be released by Intel. Its heat dissipation issues were considered somewhat decent for a mainstream CPU with good cooling, achieving around 78 degrees Celsius. Its multimedia capacity made it an excellent choice for late Windows Media Center and Windows Vista Capable systems.

It was a powerful CPU at the time, eliminating many of the issues the Pentium 4 had.

However, NetBurst's legacy ends here, and a new monolithic multi-core architecture lies across the horizon for Intel.

• Intel NetBurst Architecture
• 2.8GHz - 3.8GHz
• 125 million transistors
• 90nm - 65nm
• 512KB - 4MB cache
• 32-bit or 64-bit
• Codenames: Gallatin and Prescott
• 533MHz - 800MHz Front Side Bus

Intel P6 Derived

2003: Intel Pentium M

The first Pentium M-based CPU was the Banias core. This core was officially released in 2003 but did not hit the market until 2005. The Pentium M drops all of the previous NetBurst architecture changes and starts the whole process from the Pentium II-based systems and, more likely, the Pentium III-based.

The discovery happened in Israel, where Intel has a large manufacturing and research centre. This centre discovered that the P6 architecture could be improved and outperform the Pentium 4-based NetBurst, yet being far cooler, quieter and more energy-efficient. By this point, notebook PCs were running on either the really old Pentium III-M processors or Pentium Mobile 4-M processors. The Pentium 4-M did not perform as well as the Pentium III-M, which meant that Intel had to put some research funds into researching why. They did this in their Israel plant, which quickly discovered that the large pipeline installed into the NetBurst architecture meant it was not performing as well as it could. Also, the Pentium III had more room for cache memory, which could improve performance more than increased CPU frequency.

This led to the dropping of the Pentium 4 for mobile applications and the resurrection of the Pentium III, in one way. Still, the P6 architecture was changed to make it perform better.

The main changes in the architecture can be described very easily. Basically, Intel adapted it to give it a 2 to 5-stage longer pipeline, a faster front-side bus, better branch prediction (as a longer pipeline will need this), a faster base speed (giving clock speeds of up to 2.26GHz on a 16-stage pipeline when combined with a 133MHz front side bus) and a larger Level 2 cache of up to 2MB. The NetBurst architecture was mainly outperformed here due to the shorter pipeline, better branch prediction, and a larger Level 2 cache. These architectural changes are still apparent on modern CPUs compared with the NetBurst architecture, which has faded away entirely.

Intel also released the Centrino brand, which was the platform that included an Intel Pentium-M (never a Celeron-M, but later Centrino was able to be different CPUs, just to a Celeron), an Intel Wi-Fi card and an Intel chipset.

The first Pentium M, named Banias, only featured a 1MB Level 2 cache, which was still the same as the Pentium 4 at the time (Prescott) but was of lower latency than the Pentium 4's cache memory. The Pentium M may be the future, even in the desktop market. The Dothan CPU was released with a higher front-side bus of 533MHz and up to 2MB of L2 cache. The Dothan core was also produced using the 90nm fabrication instead of the 130nm used by the Banias core.

• Intel P6 Derived Architecture
• 900MHz - 2.26GHz
• 77 million transistors
• 130nm - 90nm
• 1MB - 2MB cache
• 32-bit
• Codenames: Banias and Dothan
• 400MHz - 533MHz Front Side Bus

2003: Intel Celeron M (Pentium M-based)

As the Pentium M was released, a budget version was also released, known as the Celeron M. This was the first mobile-only Celeron to be created, and it was the final un-unified Celeron collection. It was based entirely on the 32-bit Pentium M; to that extent, it was just a Pentium M with half the Level 2 cache. As always, it had half the cache because it was only to be made room for if during the production of a Pentium M there is a mistake, Intel could claim it to be a Celeron, a process known as binning. Little, however, did anyone know that this architecture would be even further developed.

• Intel P6 Derived Architecture
• 800MHz - 2.00GHz
• 77 million transistors
• 130nm - 90nm
• 512KB - 1MB cache
• 32-bit
• Codenames: Banias and Dothan
• 400MHz - 533MHz Front Side Bus

Intel Enhanced Pentium M Microarchitecture

2006: Intel Core Duo

The Core Duo comprised two cores on the same die, as opposed to any NetBurst CPU, which composed two CPUs on the same chip. This meant that the Core Duo is a single-die, dual-core CPU (2,1) instead of the NetBurst dual cores, which are dual-die, dual-core (2,2). This improves energy efficiency and speed. The Core Architecture means a shared Level 2 cache on the CPU, not two Level 2 caches. This means that both cores work together and share data, but it also means that data is halved equally. The reason is that both cores work together, rather than against each other for data, improving latency and, more importantly, reducing bottlenecks. The cores also share the same front-side bus, meaning each core sends simultaneously.

The Core Duo was a more powerful version of the Pentium M; it was the next step in 65nm technology and only consumed 25W. It was also a 12-stage pipeline system. The Core Duo was not 64-bit.

The Core Duo replaced the Pentium M and was launched under the name 'Centrino Duo' as part of the Centrino platform.

The Core Duo was the first Intel processor to feature in an Apple product and marked the change from the PowerPC architecture developed by IBM for Apple to Intel.

• Intel Enhanced Pentium M Microarchitecture
• 1.5GHz - 2.5GHz
• 151 million transistors 65nm
• 2MB cache
• 32-bit
• Codenames: Yonah
• 533MHz - 667MHz Front Side Bus

2006: Intel Core Solo

The Core Solo was a defect in manufacturing, like the Celeron. It usually would be priced at around half of that for a dual-core equivalent. Like the Duo, the Core Solo is also manufactured on the 65nm technology. When a Core Duo is being manufactured, and a core fails to work, it is binned and claimed to be a Core Solo, which is why all Core Solos have two cores (but with one disabled).

The Core Solo is also an ultra-low-power processor that only requires around 15W. However, compared with the Pentium D and Pentium M, it is significantly outperformed. However, the series was seen only in the mobile computing environment, but it eventually made it to the net top line. It was the direct successor of the Pentium M. The Core Architecture significantly improved performance per watt compared with the NetBurst architecture by lowering the pipeline and increasing the core count. Still, at the same time, it could not cope with high-end applications that NetBurst could perform, such as high-definition video decoding. This is why the Core Architecture was replaced by its younger brother, the Core 2 Architecture.

• Intel Enhanced Pentium M Microarchitecture
• 1.5GHz - 2.5GHz
• 151 million transistors 65nm
• 2MB cache
• 32-bit
• Codenames: Yonah
• 533MHz - 667MHz Front Side Bus

Intel Core Microarchitecture

2006: Intel Core 2 Extreme

The early CPUs in the Core 2 series were the Core 2 Extreme line. Although the Core Microarchitecture is not directly considered a new architecture, in most white papers, it is referred to as a new architecture due to the drastic changes compared with the Enhanced Pentium M Microarchitecture. The Core 2 Extreme was exceptionally early for an Extreme processor from Intel. The first Core 2 Extreme processors had two cores. The performance level of the Core 2 Extreme gave it the boost that no other CPU had, as its front-side bus and clock frequencies were far higher than any other CPU by Intel or AMD. In addition, the Core 2 Extreme had an effective 64-bit address bus, improving performance.

Intel denoted an Extreme CPU with an XE at the end, and the Core 2 Extreme followed this convention.

Another move by Intel to improve performance was the move to quad-core CPUs. Intel released another enthusiast product known as the Core 2 Extreme Kentsfield XE. The quad-core edition of this had an unlocked multiplier and a high frequency. The Core 2 Extreme quad-core had a thermal design power of 130W. This made this equally as power-demanding as the Pentium Extreme Edition on the NetBurst architecture. The Core 2 Extreme was released first as a QX6700. The Q stands for the quad-core, and the X stands for the Extreme Edition, which was clocked at 2.67GHz.

Initially, the Extremes had a 1066MHz front-side bus, but later editions of this Extreme Edition CPU featured a 1333MHz front-side bus and core speeds of up to 3.0GHz.

After the Core 2 Quad was released, Intel dropped the enthusiast market for quite a while and needed to bring back the Extreme Edition with the new features found in the later Wolfdale and Yorkfield cores.

The Core 2 Extreme Edition with 45nm fabrication was in stock from early 2008, with up to 12MB of Level 2 cache. The 45nm edition was based on the Core 2 Quad's model, known as the 9000 series of Yorkfield cores. This CPU again had an unlocked multiplier and a new logo. In addition, Intel included VT Technology and SSE4.1 instructions in the CPU.

• Intel Core Microarchitecture
• 2.93GHz - 3.3GHz
• 291 million transistors (dual) - 582 million transistors (quad)
• 65nm - 45nm
• 4MB cache (dual) - 12MB cache (quad)
• 32-bit or 64-bit
• Codenames: Conroe XE (dual), Kentsfield XE (quad) and Yorkfield XE (quad)
• 1066MHz - 1600MHz Front Side Bus

2006: Intel Core 2 Duo

Although released after the enthusiast edition, the Core 2 Duo succeeded the Core Duo in the mainstream market. The Core 2 Duo was the most accessible multicore CPU on the market at its time and held its reputation for years to come. It was not a success at its initial launch, but in 2007, its actual adoption into the market came as no big shock to anyone.

The Core 2 Duo was initially clocked at 1.86GHz with an 800MHz front-side bus and had up to 4MB of cache. However, if a manufacturing defect occurred, such as a section of the cache being broken during manufacture, Intel marketed it as a 2MB L2 cache version known as the E6300 and E6400 (through the process of binning).

Intel was directly ahead of the competition with the Duo due to the reduced power consumption over competitors. The Duo also outperformed the AMD Turion X2 in the mobile market and the Athlon X2 in the mainstream desktop market. However, the Core 2 Duo was about to meet its match. AMD was working on an FX edition of the Athlon with two cores. This would eventually outperform and end up cheaper than the Core 2 Duo.

Many considered the Core 2 Duo the most significant Intel CPU released after the Pentium.

Intel also released the Core 2 Duo and the Centrino 2 platform, including a Core 2 Duo CPU, a specific Wi-Fi card, and a specific chipset.

Intel announced a 45nm iteration of the Core Microarchitecture to keep it competitive, known by its codename 'Wolfdale', which led to the renumbering of the E7000 and E8000 series. The Core 2 Duo also got a redesigned logo with the later releases. Finally, this CPU update brought two of the most important improvements—SSE4.1 and Intel VT technology, or Virtualization Technology.

Wolfdale is also designed with an improved front-side bus of 1333MHz and a 3MB or 6MB L2 cache, giving the CPU one and two MB of bonus cache, respectively, compared with Conroe and Allendale.

• Intel Core Microarchitecture
• 1.2GHz - 3.0GHz
• 291 million transistors
• 65nm - 45nm
• 2MB - 6MB cache
• 32-bit or 64-bit
• Codenames: Conroe, Allendale and Wolfdale
• 800MHz - 1333MHz Front Side Bus

2006: Intel Pentium Dual-Core

The first Pentium Dual-Core was manufactured on 65 nm. The processors were low-end Core 2 Duos based on the Allendale core. They were designed with 1MB of L2 cache instead of the 2MB found on Core 2 Duos under Allendale.

Intel designed it as a lower power consumption edition for notebook PCs and the standard 65W equivalent to a Core 2 Duo for desktops. If the cache fails again, Intel could still sell the CPU as a different product.

Due to its price and well-established Pentium name, the Pentium Dual-Core was one of the most successful processors in the dual-core market.

Featuring the 45nm fabrication process, this CPU was able to accommodate a 2MB Level 2 cache system, which boosted performance. This 45nm edition also meant that the CPU was highly overclockable, and Intel released an E6500K edition, which had an unlocked multiplier like the Core 2 Extreme Edition, making it more overclockable.

• Intel Core Microarchitecture
• 2.00GHz - 3.0GHz
• 291 million transistors
• 65nm - 45nm
• 1MB cache
• 32-bit or 64-bit
• Codenames: Allendale, Yonah and Merom
• 533MHz - 1066MHz Front Side Bus

2006: Intel Celeron Dual-Core

The first dual-core Celeron was the dual-core version. The Celeron Dual-Core is based on the Allendale Pentium Dual-Core, with just 512KB of Level 2 cache. The CPU is 65W, like the Pentium Dual Core and Core 2 Duo.

Intel's marketing strategy for this product was that it was the low end, while the Pentium Dual-Core was the mainstream processor, the Core 2 Duo was the entry-level enthusiast processor, the Core 2 Quad was the mainstream enthusiast processor, and the Core 2 Extreme was the complete package.

Being completely based on the Core 2 architecture reduced the life of the architecture and made it look very cheap. Intel later released a Wolfdale version of this core, which gave it VT and SSE4.1, and it was manufactured using a 45nm fabrication process.

• Intel Core Microarchitecture
• 2.00GHz - 3.0GHz
• 291 million transistors
• 65nm - 45nm
• 512KB cache
• 32-bit or 64-bit
• Codenames: Allendale and Wolfdale
• 800MHz - 1066MHz Front Side Bus

2006: Intel Xeon (65nm - 45nm dual-core)

The redesign of the Xeon was based on the Core Microarchitecture. This meant that instead of two dies on one chip, there were two cores on the one die. The fabrication change also resulted in lower power consumption.

Before this Xeon was released, the Xeon family was somewhat out of place. The Core 2 series brought back the Xeon with the first Xeon based on the Woodcrest architecture. However, none of these cores would be possible for desktop processors. Despite this, things were changing; the desktop processors were just server ones with disabled features.

None of these Xeon processors featured Hyper-Threading, as seen in NetBurst. Intel came up with the idea to base their Xeon processors on the successful dual-core systems called Conroe, seen in the Core 2 Duo processors. Shockingly, they use the equivalent socket, LGA775, as the Core 2 Duo processors do. The Conroe Xeon was mainly designed for workstation-based computers rather than high-end servers.

They feature EIST and Intel VT (Virtualisation Technology) technology. They also only function in a single CPU operation and do not support multi-CPU systems.

• Intel Core Microarchitecture
• 1.60GHz - 3.0GHz
• 300 million transistors
• 65nm - 45nm
• 4MB - 6MB L2 cache
• 32-bit or 64-bit
• Codenames: Woodcrest, Allendale and Wolfdale
• 1066MHz - 1600MHz Front Side Bus

2007: Intel Core 2 Solo

The Core Solo previously made Intel's defective/binned CPU due to manufacturing issues within the Core Microarchitecture.

Before the Core 2 Solo, the Core 2 series had nowhere to fall back to as it could not claim it was a Core Solo if it was not the same architecture. For this reason, Intel designed a fallback processor known as the Core 2 Solo. If one core failed in the Core 2 Duo, Intel could claim it was a Core 2 Solo, reducing waste.

The Core 2 Solo is mainly seen in notebooks but also on nettops.

Intel's Core 2 Solo was also redesigned when the Core 2 Duo was. This gave it the 45nm fabrication and the VT technology found in the Wolfdale core.

• Intel Core Microarchitecture
• 1.0GHz - 1.4GHz
• 291 million transistors
• 65nm - 45nm
• 2MB - 3MB cache
• 32-bit or 64-bit
• Codenames: Merom-L and Penryn-L
• 533MHz Front Side Bus

2007: Intel Core 2 Quad

The Core 2 Quad became a new mainstream processor. Although there were initially only two models, the Q6600 and the Q6700, the Core 2 Quad eventually started to attract a larger mainstream market. In late 2007, the final 65nm Core 2 Quad was released at 3.0GHz with a thermal design power of 95W.

The Core 2 Quad significantly changed the design to that of the Core 2 Duo. The Core 2 Quad comprised two Intel Core 2 Duo processor dies. The result was that the processor bandwidth was lower per core. The four cores work well together - but not as well as if it were a native four-core system.

The Core 2 Quad 65nm was codenamed the Kentsfield core and comprised of two Conroe Core 2 Duo dies on the one chip featuring 2x4MB caches, resulting in a total Level 2 cache of 8MB. A serious problem for the early Quad line was that the CPU had problems sharing the cache between two CPUs. The Core 2 Quad was a decent CPU but was criticised for its concept of two processors on one chip or a multi-chip module, which Intel did back in the NetBurst Architecture. The Yorkfield core was released in mid-2008. The core featured four cores as standard, but instead of using Conroe or Allendale cores, it was based on Wolfdale cores, which gave it the functions of Wolfdale, such as VT and SSE4.1. Yorkfield was a minor revision but a significant improvement on the Kentsfield core.

This CPU marked the end of another era for Intel, the end of the desktop Core Microarchitecture, but more importantly, the end of the Socket LGA775 era, one that lasted for three years and through two microarchitectures.

• Intel Core Microarchitecture
• 2.40GHz - 3.0GHz
• 582 million transistors
• 65nm - 45nm
• 4MB - 12MB cache
• 32-bit or 64-bit
• Codenames: Kentsfield and Yorkfield
• 1066MHz - 1333MHz Front Side Bus

2007: Intel Xeon (65nm - 45nm quad and hexacore)

Intel's success with the dual-core Core 2 family meant that the quad-core chips were also successful. This led to the production of the Core 2 Quad, which in turn led to the quad-core Xeon.

This Xeon was originally based completely on the Core 2 Quad Kentsfield and later the Yorkfield core. The clock frequencies measure from 2.13GHz to 2.66GHz for the Kentsfield core, with 2x4MB caches and a 1066MHz front-side bus. The Yorkfield cores were later based on the exact same cores as the desktop equivalent, with frequencies of 2.66GHz, 2.83GHz, and 3.0GHz. The cache sizes go as far as 12MB.

The final quad-core server Xeon in the Core 2 architecture was the Harpertown. The Harpertown was based on the LGA771 package and had two dual-core dies on the same core. It brought the front side bus speed up to 1600MHz.

The final quad-core Xeon released in this architecture was a Core 2 Extreme Edition chip known as the QX9775. This chip used socket 771 and could run a high-end gaming machine known as Intel's Skull Trail. This core in the Core Microarchitecture Xeon collection was called the Dunnington core. It contained six cores, i.e., "Core 2 Hexa"—three dual-core dies on the same chip.

• Intel Core Microarchitecture
• 2.13GHz - 3.4GHz
• 600 million transistors
• 65nm - 45nm
• 4MB - 16MB cache
• 32-bit or 64-bit
• Codenames: Kentsfield, Yorkfield and Dunnington
• 1066MHz - 1600MHz Front Side Bus

Intel Atom Microarchitecture

2008: Intel Centrino Atom

The Centrino Atom processor was not very different from its successor, the Intel Atom, but it met the Centrino requirements.

The Centrino Atom was ditched because the requirements for the Centrino platform at the time were too broad to continue to call it a member of the Centrino platform.

The Centrino Atom, however, could only be supported for netbooks rather than Mobile Internet Devices (MIDs). For this reason, the Centrino Atom was removed from the Centrino family and replaced by the Intel Atom.

Not surprisingly, it outperforms the early Intel Centrino platforms based on the Pentium M 740 Banias.

• Intel Atom Microarchitecture
• 800MHz - 2.0GHz
• 47 million transistors
• 45nm
• 512KB cache
• 32-bit or 64-bit
• Codenames: Silverthorne, Lincroft
• 533MHz Front Side Bus

2008: Intel Atom

Intel released a new mobile CPU slightly later than their dual-core Celerons. This processor would be the Atom. The Atom would not perform as well as a Celeron in terms of the MIPS and GFLOPS it could perform, but it would be a cheap and low-power mainstream processor and not a budget one.

It easily outperforms the Celeron in tasks where dual-core is essential, as the Hyper-Threading makes up for the missing second core and performs slightly better than two cores.

The processor was first made available in early 2008 as Silverthorne, with a maximum thermal design power of 8W, 3W less than the best Celeron. A new Intel Atom-based Mini-ITX board with an Intel Atom soldered on it would cost just £30 when it was first released.

The Atom was small enough that Intel could manufacture far more than any other CPU per silicon wafer (about 2500 CPUs per 300mm diameter silicon wafer). It was initially a basis of the Centrino platform, making the Centrino Atom, but this was dropped in favour of the broader Intel Atom-based name. The Silverthorne-based systems were designed specifically for the Mobile Internet Device (MID) market.

This segment had never seen a CPU as powerful as the x86 CPUs. Intel's Diamondville processor soon took the role of netbooks, which had just recently been announced and were based on AMD's Geode and the ARM processor netbook market that had been created recently. This would take over from the Centrino Atom, which currently uses Intel Core 2 Duo processors.

Due to its exceptionally low power consumption and temperatures, the Atom does not feature an Integrated Heat Spreader (IHS).

Netbooks based on the Atom architecture are slimmer than the Centrino Atom processors seen previously in netbooks. Their chipsets have a lower power consumption of just 12W (including the Atom itself). The Atom can use both DDR2 and DDR3 RAM and is also available as a dual-core version for nettops and higher-end netbooks. The dual-core Atoms are named the 300 series, and they are dual-core, dual-die modules again, like the Pentium D processor found in the far earlier NetBurst architecture.

• Intel Atom Microarchitecture
• 1.2GHz - 2.0GHz
• 47 million transistors
• 45nm, 32nm
• 512KB cache
• 32-bit or 64-bit
• Codenames: Diamondville, Pineview, Cedarview
• 533MHz Front Side Bus