Pentium 4

The Pentium 4 is a seventh-generation x86 architecture microprocessor produced by Intel and is their first all-new CPU design since the Pentium Pro of 1995. The original Pentium 4, codenamed "Willamette", ran at 1.3, 1.4 and 1.5 GHz and was released in November 2000. Unlike the Pentium II, Pentium III, and various Celerons, the architecture owed little to the Pentium Pro design, and was new from the ground up. Notable with the introduction of the Pentium 4, was the very fast 400MHz FSB; it was actually a 100MHz Quad-Pumped bus, but the theoretical bandwidth was 4x that of a normal bus, and so it was considered to run at 400MHz- the fastest competition was running at 133MHz.

To the surprise of most industry observers, the Pentium 4 did not improve on the old P6 design in either of the normal two key performance measures: integer processing speed or floating-point performance. Instead, it sacrificed per-cycle performance in order to gain two things: very high clockspeeds, and SSE performance. As is traditional with Intel's flagship chips, the Pentium 4 also comes in a low-end Celeron version (often referred to as Celeron 4) and a high-end Xeon version intended for SMP configurations.

The Pentium 4 performs much less work per cycle than other CPUs (such as the various Athlon or older Pentium III architectures) but the original design objective - to sacrifice instructions per clock cycle in order to achieve a greater number of cycles per second (i.e. greater frequency or clockspeed) - has been fulfilled until the platform ran into unsolvable thermal problems before reaching 4GHz, far short of the original scalability claim of 10GHz. As of mid-2005, Intel has abandoned further work on Pentium 4 to focus on cooler running Pentium M derived solutions for the desktop PC and small server market. This essentially means Intel has returned to the Pentium III CPU core and only the system bus of the Pentium 4 will live on.

Contents

1 Willamette 2 Northwood 3 Extreme Edition 4 Prescott 5 Dual Core 6 Technical highlights 7 Successor

Willamette

Willamette, the first Pentium 4, suffered long delays in the design process. Most industry experts regarded the initial 1.3, 1.4 and 1.5GHz P4 release as a stopgap product, introduced before it was truly ready. According to these experts, the Willamette was released because the competing AMD Athlon Thunderbird was at that time outperforming the elderly Pentium III, and further improvements to the Intel P-III were not yet possible. The cores were produced using a 0.18 micrometre (180 nm) process and initially utilized socket 423 on motherboards, later revisions moving to socket 478.

On the test-bench, the Willamette was somewhat disappointing to analysts in that not only was it unable to outperform the Athlon and the highest-clocked P-IIIs in all testing situations, it was not clearly superior to even the low-end AMD Duron. Although introduced at a price of 819 US Dollars (in 1000 unit wholesale quantities), it sold at a modest but respectable rate, handicapped somewhat by the requirement of relatively expensive Rambus Dynamic RAM (RDRAM). The Pentium III remained Intel's top selling chip, and the Athlon also sold slightly better than the Pentium 4.

In January 2001, a still slower 1.3GHz model was added to the range, but over the next twelve months, Intel gradually started pegging back AMD's lead. April 2001 brought the 1.7GHz P4, the first one to provide performance clearly superior to the old Pentium III. July saw 1.6 and 1.8GHz models and in August 2001, Intel released 1.9 and 2.0GHz Pentium 4s. In the same month, they released a new chipset that supported much cheaper PC133 SDRAM. While use of this RAM was much slower than RDRAM, the fact that PC133 was much cheaper caused the Pentium 4's sales to massively improve, displacing the Pentium III virtually overnight to become the top-selling processor on the market.

The 2.0 GHz was the first P4 to provide a serious challenge to the rival Athlon Thunderbird, which until then had been unquestionably the fastest x86 CPU on the market. Many observers concluded that the Thunderbird was still faster overall, but the performance gap was sufficiently narrow that it was not unreasonable for partisans of either camp to claim superiority. For Intel, this was a very significant achievement. The firm had held the x86 CPU performance crown for nearly 16 years straight, with only two brief exceptions prior to the release of the AMD Athlon.

Northwood

In October 2001 the Athlon XP regained a clear lead for AMD, but in January 2002 Intel released Pentium 4s with their new Northwood core at 2.0 and 2.2 GHz. Northwood combined an increase in the secondary cache size from 256k to 512k (increasing the transistor count to 55 million, up from 42 million) with a transition to a new 0.13 micrometer (130 nm) fabrication process. By making the chip out of smaller transistors, chips can run at the same speed and produce less heat, or run at higher clocks. Unfortunately for many consumers, the new core also made upgrades impossible due to the requirement of a new socket (socket 478), though some of the later Willamettes were produced with this socket.

With Northwood, the P4 came of age. The battle for performance leadership remained competitive (as AMD introduced faster versions of the Athlon XP) but most observers agreed that the fastest Northwood P4 was usually a fraction ahead of its rival. This was particularly so in the summer of 2002, when AMD's changeover to a 0.13 micrometre (130 nm) production process was delayed and the P4s in the 2.4 to 2.8 GHz range were clearly the fastest chips on the market.

A 2.4 GHz P4 was released in April 2002, a 2.53 GHz part in May (at which point the bus speed was increased from the original 400MHz to 533MHz), 2.6 and 2.8 GHz parts in August, and a 3.06 GHz Pentium 4 arrived in November.

The 3.06 GHz processor supported Hyper-threading (first appeared in Xeon), enabling multiple threads to be run together by duplicating some parts of the processor in order to let the operating system believe that there are two logical processors. HyperThreading was present in all Northwood CPUs, but was disabled in the core in all but the 3.06GHz model.

In April 2003, Intel launched new variants, ranging from 2.4 to 3.0GHz. The key difference on these new versions was that they all supported Hyper-Threading, and ran their system buses at 800 MHz. This was supposedly to compete better with AMD's Hammer line of processors. However, only Opteron was launched, and AMD initially refused to provide an AGP controller, thus preventing the Opteron from encroaching on the Pentium 4's territory. AMD did boost the Athlon XP's bus speed from 333 MHz to 400 MHz, but it wasn't enough to hold off the new 3.0 GHz P4– and the FSB wasn't the problem; the 333MHz to 400MHz transition yielded little to no performance increase. A 3.2 GHz variant was launched in June and a final 3.4GHz version was launched in early 2004.

Overclocking Northwood cores yielded a startling phenomenon. When VCore was pumped past 1.7v, the processor literally died. This occurrence has since been called "NDS," or Northwood Death Syndrome.

Extreme Edition

In September 2003, at the Intel Developer Forum, the Pentium 4 Extreme Edition (P4EE) was announced, just over a week before the launch of Athlon 64, and Athlon 64 FX (AMD64 FX). The design was mostly identical to Pentium 4 (to the extent that it would run in the same motherboards), but differed by an added 2 MB of Level 3 cache. It shared the same Gallatin core as the Xeon MP, though in a Socket 478 form factor (as opposed to Socket 603 for the Xeon MP) and with an 800MHz bus, twice as fast as that of the Xeon MP. An LGA775 version is also available.

While Intel maintained that the Extreme Edition was aimed at gamers, some viewed it as an attempt to steal the Athlon 64's launch thunder, nicknaming it the "Emergency Edition". Curiously, many condemned Intel for cannibalizing the Xeon line, but no such complaints were aimed at AMD, who did the same with their Athlon 64 FX (which differed less from Opteron than the Extreme Edition did from Xeon MP).

The effect of the added cache was somewhat variable. In office applications, the Extreme Edition was generally a bit slower than the Northwood, owing to higher latency added by the L3 cache. Some games benefited from the added cache, particularly those based on the Quake III and Unreal engines. However, the area which improved the most was multimedia encoding, which was not only faster than the Pentium 4, but also both Athlon 64s.

A slight performance increase was achieved in late 2004 by increasing the bus speed from 800MHz to 1066MHz. Only one Gallatin-based chip at 3.46GHz was released before the Extreme Edition was migrated to the Prescott core. The new 3.73GHz Extreme Edition had the same features as a 6x0-sequence Prescott, but with a 1066MHz bus. In practice however, the 3.73GHz Extreme Editon almost always proved to be slower than the 3.46GHz version.

Prescott

On February 1, 2004, Intel introduced a new core codenamed "Prescott." The core uses a 90nm process for the first time, and "[it] is also a major reworking of the Pentium 4's microarchitecture - major enough that some analysts are surprised Intel didn't opt to call this processor the Pentium 5" ([1] (http://techreport.com/reviews/2004q1/p4-prescott/index.x?pg=1)). Although a Prescott clocked at the same rate as a Northwood, benchmarks show that a Northwood runs slightly faster than a Prescott. The Prescott's architecture allows it to be easily set at higher clock-rates. (See Overclocking.) Currently, 3.8 GHz is the fastest Prescott-based processor.

Upon release, the Prescott turned out to generate approximately 60% more heat clock-for-clock than the Northwood, and almost every review of it was negative. A shift in socket type (from Socket 478 to LGA775) was expected to reduce the heat to more acceptable levels, but in fact proved to have the opposite effect, with power requirements increasing by a further 10%. However, the LGA775 reference cooler and mounting system were somewhat better designs, so average temperatures were slightly lowered. Subsequent revisions to the processor by Intel engineers were expected to reduce average temperatures, but this never happened outside of the lowest speed grades.

Finally, the thermal problems were so severe, Intel decided to abandon the Prescott architecture altogether, and attempts to roll out a 4 GHz part were abandoned, as a waste of internal resources. Also of concern was the fact that reviews showed in extreme cases it took a 5.2 GHz Prescott core to match the performance of an Athlon FX-55 [2] (http://www.hardwareanalysis.com/content/article/1787/). Considering Intel boasted at launch the Pentium 4 architecture was designed for 10 GHz operation, this can be seen as one of the most significant, certainly most public, engineering shortfalls in Intel’s history.

The Pentium M has now reportedly become the internal reference layout for Intel’s design teams, and P4 development has been essentially abandoned. To this extent the little-funded Israeli design team that produced the Pentium M core, has now taken over the much larger desktop development project.

Why the Prescott ended up in such a disaster can be attributed to internal politics at Intel. The marketing department wanted ever higher clock speeds, to differentiate their products from AMD. Processor design was dictated by marketing needs, rather than architectural requirements. Careers were then built on the concept of higher clock speeds, and the termination of the P4 project when it finally came, had consequences for many members of the management team at the well-funded desktop division.

LGA775 Prescotts use a rating system, labeling them as the 5xx series (Celerons are the 3xx series, while Pentium Ms are the 7xx series). The fastest, the 570J is clocked at 3.8GHz. Plans for 4GHz processors were recently axed by Intel in favor of dual core processors. [3] (http://www.geek.com/news/geeknews/2004Oct/gee20041015027427.htm)

The 570J processor also introduced the XD Bit (eXecute Disable) or Execute Disabled Bit [4] (http://www.intel.com/business/bss/infrastructure/security/xdbit.htm) to Intel's line of processors. This technology, first introduced to the x86 line by AMD and called NX (No eXecute), can help prevent certain types of malicious code from exploiting a buffer overflow to get executed.

Intel recently (Q'1 05) released a new Prescott core with 6x0 numbering. It features new 64-bit technology, execution disable, speed step technology, and 2 MB of L2 cache. However, any advantage introduced by the added cache is mostly negated due to higher cache latency, and the double word size if using EM64T mode. Rather than being a targeted speed boost the double size cache is intended to provide the same space and hence performance for 64 bit mode operations.

Dual Core

Intel has planned three mainstream dual-core variants of the Pentium 4. Reputed to have a 60-80% increase in performance per clock-speed, this would be marketed as the Pentium D. An extreme edition is also planned. These chips have already started production and would be launched by 2Q 2005.

Dubbed the Smithfield core, the first Pentium-D processors will basically be two connected Prescott cores. Power consumption is estimated to be above 130 watts. Intel developers have managed to keep power consumption of the processor incrementally larger than the Prescott's power usage (115w) due to the fact that each core is clocked significantly lower. The EE edition will run at 3.2 GHz and the mainstream models will be clocked at 3.2, 3.0, and 2.8, respectively. The major difference between the EE edition of P-D and the mainstream models is the 1066 MHz bus (the mainstream models will have a P4 standard 800 MHz FSB) and the retention of HyperThreading Technology for each core.

This will be followed in Q1'06 by the Presler, which has a dedicated bus interface bridging the two cores together. Presler will also be a 65nm architecture.

Technical highlights

Following is a list of the current revisions of the Intel Pentium 4 processor, and their differentiating characteristics.

Pentium 4 Processors, Designations, and Characteristics
Public Designation	Core (Intel Codename)	CPU Frequency	Frontside Bus Frequency / Theoretical Bandwidth	Cache	Additional Features
(original release revision)	Willamette	1.3 GHz - 2.0 GHz	100 MHz / 3.2 GB/s	8 KiB L1 data + 12 KiB L1 instruction / 256 KiB L2	N/A
P4A	Northwood	1.6 GHz - 3.0 GHz	100 MHz / 3.2 GB/s	8 KiB L1 data + 12 KiB L1 instruction / 512 KiB L2	Improved branch prediction and other microcodes tweaks; these are carried over into subsequent revisions
P4B	Northwood	2.0 GHz - 3.06 GHz	133 MHz / 4.2 GB/s	8 KiB L1 data + 12 KiB L1 instruction / 512 KiB L2	Faster FSB, Hyperthreading enabled in the 3.06 GHz model
P4C	Northwood	2.4 GHz - 3.4+ GHz	200 MHz / 6.4 GB/s	8 KiB L1 data + 12 KiB L1 instruction / 512 KiB L2	Hyperthreading
P4E/5x0 series	Prescott	2.8 GHz - 3.8 GHz	200 MHz / 6.4 GB/s	16 KiB L1 data + 12 KiB L1 instruction / 1 MiB L2	Hyperthreading, longer pipeline, SSE3 instructions
P4A*	Prescott	2.4GHz - 2.93 GHz	133 MHz / 4.2 GB/s	16 KiB L1 data + 12 KiB L1 instruction / 1 MiB L2	No Hyperthreading, longer pipeline, SSE3 instructions
Extreme Edition	Gallatin	3.2 GHz - 3.4 GHz	200 MHz / 6.4 GB/s	8 KiB L1 data + 12 KiB L1 instruction / 512 KiB L2 / 2 MiB L3	Hyperthreading, addition of on-die L3 cache
P4F/5x1 series	Prescott	3.2 GHz - 3.8 GHz	200 MHz / 6.4 GB/s	16 KiB L1 data + 12 KiB L1 instruction / 1 MiB L2	Support of EM64T and eXecute Disable bit (equivalent of AMD's No eXecute bit)
6x0 series	Prescott 2MB**	2.8 GHz - 3.8 GHz	200 MHz / 6.4 GB/s	16 KiB L1 data + 12 KiB L1 instruction / 2 MiB L2	Larger L2 cache, support of EM64T, Speedstep and Thermal Monitoring 2.
Extreme Edition	Prescott 2MB**	3.73 GHz	266 MHz / 8.5 GB/s	16 KiB L1 data + 12 KiB L1 instruction / 2 MiB L2	Faster frontside bus
Pentium D#	Smithfield	2.8 GHz - 3.2 GHz	200 MHz / 6.4 GB/s	16 KiB L1 data + 12 KiB L1 instruction Per Core / Shared 2 MiB L2	Support of EM64T and eXecute Disable bit (equivalent of AMD's No eXecute bit) Dual Core Processor
Notes: Pentium 4 processors use a frontside bus that transfers data four times per cycle * - in the case of the budget Prescott processor line, Intel duplicated the "P4A" designation which retailers are supposed to use to identify the processor to buyers; no reason was given for this decision ** - the official core name for the 600-series, though the core is sometimes called by its Xeon equivalent, Irwindale in order to distinguish between it and the initial Prescott. # - the Pentium D consists of 2 Prescott cores on a single die.

Successor

The successor Tejas and Jayhawk were in development but were canceled in May 2004 due to, according to analysts, a heat issue.

A 65nm version of the 6x0 series called Cedar Mill will be released in 2006. This is not expected to be different from the 90nm version in any significant way. Cedar Mill will probably be the final revision of the Pentium 4, with a replacement called Merom based on a totally new core scheduled for introduction later that year.

af:Pentium 4

de:Pentium 4 es:Intel Pentium 4 eo:Pentium 4 id:Intel Pentium 4 nl:Pentium IV (processor) ja:Pentium 4 pl:Pentium 4 pt:Pentium 4 zh:Pentium 4