Intel has come a very long way from a small chip manufacturer to a world leader in processor manufacturing. During this time, many technologies for the production of processors have been developed, the technological process and device characteristics have been greatly optimized.

Many performance indicators of processors depend on the location of transistors on a silicon chip. The transistor arrangement technology is called microarchitecture or simply architecture. In this article, we will look at which Intel processor architectures have been used throughout the development of the company and how they differ from each other. Let's start with the most ancient microarchitectures and look all the way to new processors and plans for the future.

As I said, in this article we will not consider the capacity of processors. By the word architecture, we will understand the microarchitecture of the microcircuit, the arrangement of transistors on printed circuit board, their size, distance, technological process, all this is covered by this concept. We will not touch the RISC and CISC instruction sets either.

The second thing to pay attention to is the generations of the Intel processor. Probably, you have already heard many times - this fifth generation processor, that fourth, and this seventh. Many people think that this is indicated by i3, i5, i7. But in fact, there is no i3, and so on - these are processor brands. And the generation depends on the architecture used.

With each new generation, the architecture improved, processors became faster, more economical and smaller, they generated less heat, but at the same time they cost more. There are few articles on the Internet that would describe all this in full. Now let's look at how it all began.

Intel processor architectures

I say right away that you should not expect technical details from the article, we will consider only the basic differences that will be of interest to ordinary users.

First processors

First, let's briefly dive into history to understand how it all began. Let's not go too far and start with 32-bit processors. The first was Intel 80386, it appeared in 1986 and could operate at a frequency of up to 40 MHz. Old processors also had a generational countdown. This processor belongs to the third generation, and the 1500 nm process technology was used here.

The next, fourth generation was 80486. The architecture used in it was called 486. The processor ran at a frequency of 50 MHz and could execute 40 million instructions per second. The processor had 8 KB of the first level cache, and the 1000 nm manufacturing process was used for manufacturing.

The next architecture was P5 or Pentium. These processors appeared in 1993, here the cache was increased to 32 kb, the frequency was up to 60 MHz, and the technical process was reduced to 800 nm. In the sixth generation P6, the cache size was 32 KB, and the frequency reached 450 MHz. The process was reduced to 180 nm.

Then the company began to produce processors based on the NetBurst architecture. Here we used 16 KB of the first level cache per core, and up to 2 MB of the second level cache. The frequency increased to 3 GHz, while the technical process remained at the same level - 180 nm. 64-bit processors appeared already here, which supported addressing more memory. Many command extensions were also made, and Hyper-Threading technology was added, which allowed two threads to be created from a single core, which improved performance.

Naturally, each architecture improved over time, the frequency increased and the process technology decreased. There were also intermediate architectures, but here everything has been simplified a bit, since this is not our main topic.

Intel Core

NetBurst was replaced in 2006 by the Intel Core architecture. One of the reasons for the development of this architecture was the impossibility of increasing the frequency in NetBrust, as well as its very large heat dissipation. This architecture was designed for the development of multi-core processors, the size of the first level cache was increased to 64 KB. The frequency remained at the level of 3 GHz, but the power consumption was greatly reduced, as well as the process technology, to 60 nm.

Core architecture processors supported Intel-VT hardware virtualization as well as some command extensions, but did not support Hyper-Threading, as they were designed based on the P6 architecture, where this capability was not yet available.

First generation - Nehalem

Then the numbering of generations was started from the beginning, because all the following architectures are improved Intel versions core. The Nehalem architecture replaced the Core, which had some limitations, such as the inability to increase the clock speed. She appeared in 2007. It uses a 45 nm process and has added support for Hyper-Therading technology.

Nehalem processors have 64 KB L1 cache, 4 MB L2 cache and 12 MB L3 cache. The cache is available to all processor cores. It also became possible to integrate a graphics accelerator into the processor. The frequency has not changed, but the performance and the size of the printed circuit board have increased.

Second generation - Sandy Bridge

Sandy Bridge appeared in 2011 to replace Nehalem. The 32 nm process technology is already used here, the same amount of first-level cache, 256 MB of second-level cache and 8 MB of third-level cache are used here. Experimental models used up to 15 MB of shared cache.

Also, now all devices are available with a built-in graphics accelerator. The maximum frequency has been increased, as well as the overall performance.

Third generation - Ivy Bridge

Ivy Bridge processors are faster than Sandy Bridge processors and use the 22nm process technology. They consume 50% less energy than previous models, and also provide 25-60% higher performance. The processors also support Intel Quick Sync technology, which allows you to encode video several times faster.

Fourth generation - Haswell

The Intel Haswell processor generation was developed in 2012. The same manufacturing process was used here - 22 nm, the cache design was changed, power consumption mechanisms were improved and performance was slightly improved. But on the other hand, the processor supports many new connectors: LGA 1150, BGA 1364, LGA 2011-3, DDR4 technologies, and so on. The main advantage of Haswell is that it can be used in portable devices due to its very low power consumption.

Fifth generation - Broadwell

This is an improved version of the Haswell architecture that uses the 14nm process technology. In addition, several improvements were made to the architecture, which resulted in an average performance increase of 5%.

Sixth generation - Skylake

The next architecture of intel core processors - the sixth generation of Skylake was released in 2015. This is one of the most significant updates to the Core architecture. To install the processor on the motherboard, an LGA 1151 socket is used, DDR4 memory is now supported, but DDR3 support has been preserved. Thunderbolt 3.0 is supported, as well as the DMI 3.0 bus, which gives twice the speed. And already by tradition there was increased productivity, as well as reduced power consumption.

Seventh generation - Kaby Lake

New, seventh generation Core - Kaby Lake came out this year, the first processors appeared in mid-January. There haven't been many changes here. The 14 nm process technology has been retained, as well as the same LGA 1151 socket. Supports DDR3L SDRAM and DDR4 SDRAM memory strips, tires PCI Express 3.0, USB 3.1. In addition, the frequency was slightly increased, and the density of the transistors was also reduced. The maximum frequency is 4.2 GHz.

findings

In this article, we looked at the Intel processor architectures that were used in the past, as well as those that are used now. Further, the company plans to switch to the 10 nm process technology and this generation of Intel processors will be called CanonLake. But so far, Intel is not ready for this.

Therefore, in 2017 it is planned to release an improved version of SkyLake under the code name Coffe Lake. It is also possible that there will be other microarchitectures of the Intel processor until the company fully masters the new process technology. But we will learn about all this in time. I hope this information was useful to you.

about the author

Founder and administrator of the site site, I am fond of open source software and the Linux operating system. I currently use Ubuntu as my main OS. In addition to Linux, I am interested in everything related to information technology and modern science.

Marking, positioning, use cases

This summer, Intel launched a new, fourth-generation Intel Core architecture, code-named Haswell (processor markings start with the number "4" and look like 4xxx). The main direction of development of Intel processors now sees the increase in energy efficiency. So recent generations Intel Cores show not such a strong increase in performance, but their overall energy consumption is constantly decreasing - due to the architecture, the technical process, and effective management of component consumption. The only exception is integrated graphics, whose performance has been growing noticeably from generation to generation, albeit at the expense of deteriorating power consumption.

This strategy predictably brings to the fore those devices in which energy efficiency is important - laptops and ultrabooks, as well as the only emerging (because in its previous form it could be attributed exclusively to the undead) class of Windows tablets, the main role in the development of which should be played by new processors with reduced energy consumption.

As a reminder, we recently released brief overviews of the Haswell architecture, which are quite applicable to both desktop and mobile solutions:

In addition, the performance of quad-core Core i7 processors was explored in the article comparing desktop and mobile processors. The performance of the Core i7-4500U was also separately examined. Finally, there are reviews of Haswell laptops, including performance testing: MSI GX70 on the most powerful Core i7-4930MX processor, HP Envy 17-j005er.

This article will focus on the Haswell mobile line as a whole. AT first part we will consider the division of Haswell mobile processors into series and lines, the principles of creating indexes for mobile processors, their positioning and the approximate level of performance of different series within the entire line. In second part- let's take a closer look at the specifications of each series and line and their main features, and also move on to the conclusions.

For those who are not familiar with the Intel Turbo Boost algorithm, at the end of the article we have placed short description this technology. Recommended with him before reading the rest of the material.

New letter indexes

Traditionally, all Intel Core processors are divided into three lines:

• Intel Core i3
• Intel Core i5
• Intel Core i7

The official position of Intel (which company representatives usually voice when answering the question why there are both dual-core and quad-core models among the Core i7) is that the processor is assigned to one or another line based on its overall performance level. However, in most cases, there are architectural differences between processors of different lines.

But already in Sandy Bridge, another division of processors has appeared, and in Ivy Bridge, another division of processors has become complete - into mobile and ultra-mobile solutions, depending on the level of energy efficiency. Moreover, today it is this classification that is basic: both the mobile and ultra-mobile lines have their own Core i3 / i5 / i7 with very different levels of performance. In Haswell, on the one hand, the division deepened, and on the other hand, they tried to make the line more slender, not so misleading by duplicating indices. In addition, another class has finally taken shape - ultra-mobile processors with the Y index. Ultra-mobile and mobile solutions are still marked with the letters U and M.

So, in order not to be confused, first we will analyze which letter indices are used in the modern line of fourth-generation Intel Core mobile processors:

• M - mobile processor (TDP 37-57 W);
• U - ultra mobile processor (TDP 15-28 W);
• Y - processor with extremely low consumption (TDP 11.5 W);
• Q - quad-core processor;
• X - extreme processor (top solution);
• H - processor for BGA1364 packaging.

Since TDP (thermal package) has already been mentioned, let's dwell on it in a little more detail. Please note that TDP is modern processors Intel is not “maximum”, but “nominal”, that is, it is calculated based on the load in real tasks when operating at the standard frequency, and when Turbo Boost is turned on and the frequency is increased, the heat dissipation goes beyond the declared nominal heat pack - there is a separate TDP for this. The TDP is also determined when operating at the minimum frequency. Thus, there are as many as three TDPs. This article uses nominal TDP in tables.

• The standard nominal TDP for mobile quad-core Core i7 processors is 47W, for dual-core processors - 37W;
• The letter X in the name raises the thermal package from 47 to 57 W (now there is only one such processor on the market - 4930MX);
• Standard TDP for U-series ultra mobile processors is 15 W;
• Standard TDP for Y-series processors - 11.5 W;

Digital indices

The indexes of fourth-generation Intel Core processors with Haswell architecture begin with the number 4, which just indicates that they belong to this generation (for Ivy Bridge, the indices began with 3, for Sandy Bridge - with 2). The second digit indicates belonging to the line of processors: 0 and 1 - i3, 2 and 3 - i5, 5–9 - i7.

Now let's analyze the last digits in the name of the processors.

The number 8 at the end means that this processor model has an increased TDP (from 15 to 28 W) and a significantly higher nominal frequency. Another distinguishing feature of these processors is the Iris 5100 graphics. They are focused on professional mobile systems that require stable high performance in all conditions for constant work with resource-intensive tasks. They also have overclocking with Turbo Boost, but due to the strongly raised nominal frequency, the difference between the nominal and maximum is not too great.

The number 2 at the end of the name indicates a TDP reduced from 47 to 37 W for a processor from the i7 line. But you have to pay for lower TDP with lower frequencies - minus 200 MHz to the base and boost frequencies.

If the second digit from the end in the name is 5, then the processor has a GT3 - HD 5xxx graphics core. Thus, if the last two digits in the processor name are 50, then the GT3 HD 5000 graphics core is installed in it, if 58 - then Iris 5100, and if 50H - then Iris Pro 5200, because Iris Pro 5200 is only available for processors BGA1364.

For example, let's analyze the processor with the 4950HQ index. The name of the processor contains H - means BGA1364 package; contains 5 - means GT3 HD 5xxx graphics core; combination of 50 and H gives Iris Pro 5200; Q - quad-core. And since quad-core processors are only in the Core i7 line, this is the mobile Core i7 series. This is confirmed by the second digit of the name - 9. We get: 4950HQ is a mobile quad-core eight-thread processor of the Core i7 line with a TDP of 47 W with GT3e Iris Pro 5200 graphics in BGA design.

Now that we have dealt with the names, we can talk about the division of processors into lines and series, or, more simply, about market segments.

4th generation Intel Core series and lines

So, all modern Intel mobile processors are divided into three large groups depending on power consumption: mobile (M), ultra-mobile (U) and "ultra-mobile" (Y), as well as three lines (Core i3, i5, i7) depending on performance. As a result, we can make a matrix that will allow the user to choose the processor that best suits his tasks. Let's try to bring all the data into a single table.

Series/line	Options	Core i3	Core i5	Core i7
Mobile (M)	Segment	laptops	laptops	laptops
	cores/threads	2/4	2/4	2/4, 4/8
	Max. frequencies	2.5 GHz	2.8/3.5 GHz	3/3.9 GHz
	turbo boost	No	there is	there is
	TDP	tall	tall	maximum
	Performance	above average	high	maximum
	autonomy	below the average	below the average	low
Ultramobile (U)	Segment	laptops / ultrabooks	laptops / ultrabooks	laptops / ultrabooks
	cores/threads	2/4	2/4	2/4
	Max. frequencies	2 GHz	2.6/3.1 GHz	2.8/3.3 GHz
	turbo boost	No	there is	there is
	TDP	average	average	average
	Performance	below the average	above average	high
	autonomy	above average	above average	above average
Ultra-ultramobile (Y)	Segment	ultrabooks / tablets	ultrabooks / tablets	ultrabooks / tablets
	cores/threads	2/4	2/4	2/4
	Max. frequencies	1.3 GHz	1.4/1.9 GHz	1.7/2.9 GHz
	turbo boost	No	there is	there is
	TDP	short	short	short
	Performance	low	low	low
	autonomy	high	high	high

For example: a customer needs a laptop with high processor performance and moderate cost. Since a laptop, and even a productive one, requires an M-series processor, and the requirement for moderate cost forces one to stop at the Core i5 line. We emphasize once again that, first of all, you should pay attention not to the line (Core i3, i5, i7), but to the series, because each series may have its own Core i5, but the performance level of Core i5 from two different series will be significantly differ. For example, the Y-series is very economical, but has low operating frequencies, and the Y-series Core i5 processor will be less powerful than the U-series Core i3 processor. And the mobile Core i5 processor may well be more productive than the ultra-mobile Core i7.

Approximate performance level depending on the line

Let's try to go one step further and compile a theoretical rating that would clearly demonstrate the difference between processors of different lines. For 100 points, we will take the weakest processor presented - a dual-core four-thread i3-4010Y with a clock speed of 1300 MHz and a 3 MB L3 cache. For comparison, we take the highest frequency processor (at the time of this writing) from each line. We decided to calculate the main rating by the overclocking frequency (for those processors that have Turbo Boost), in parentheses - the rating for the nominal frequency. Thus, a dual-core, four-threaded processor with a maximum frequency of 2600 MHz will receive 200 conditional points. Increasing the third-level cache from 3 to 4 MB will bring it a 2-5% (data obtained from real tests and research) increase in conditional points, and an increase in the number of cores from 2 to 4 will double the number of points, which is also achievable in reality with a good multi-threaded optimization.

Once again, we strongly draw your attention to the fact that the rating is theoretical and is based mostly on the technical parameters of the processors. In reality, a large number of factors are combined, so the performance gain over the weakest model in the line will almost certainly not be as big as in theory. Thus, one should not directly transfer the obtained ratio to real life - final conclusions can only be drawn from the results of testing in real applications. Nevertheless, this estimate allows us to roughly estimate the place of the processor in the lineup and its positioning.

So, some preliminary notes:

• Core i7 U-series processors will be about 10% ahead of Core i5 due to slightly higher clock speeds and more L3 cache.
• The difference between the Core i5 and Core i3 U-series processors with a TDP of 28W without Turbo Boost is about 30%, i.e. ideally, performance will also differ by 30%. If we take into account the capabilities of Turbo Boost, then the difference in frequencies will be about 55%. If we compare the Core i5 and Core i3 U-series processors with a TDP of 15 W, then with stable operation at the maximum frequency, the Core i5 will have a frequency of 60% higher. However, its nominal frequency is slightly lower, i.e. when operating at the nominal frequency, it can even be slightly inferior to the Core i3.
• In the M-series, the presence of 4 cores and 8 threads in the Core i7 plays a big role, but here we must remember that this advantage is manifested only in optimized software (usually professional). Core i7 processors with two cores will have slightly better performance due to higher overclocking frequencies and a slightly larger L3 cache.
• In the Y series, the Core i5 processor has a base frequency of 7.7% and an overclocking frequency of 50% higher than the Core i3. But in this case, there are additional considerations - the same energy efficiency, the noise of the cooling system, etc.
• If we compare the processors of the U and Y series, then only the frequency gap between the U- and Y-core processors i3 is 54%, while Core i5 processors are 63% at maximum overclocking frequency.

So, let's calculate the score for each line. Recall that the main score is calculated according to the maximum overclocking frequencies, the score in brackets - according to the nominal ones (that is, without overclocking using Turbo Boost). We also calculated the performance factor per watt.

¹ max. - at maximum overclocking, nom. - at rated frequency
² coefficient - conventional performance divided by TDP and multiplied by 100
³ Overclocking TDP data for these processors is unknown

From the table below, the following observations can be made:

• The U and M-series dual-core Core i7 processors are only marginally faster than the equivalent Core i5 processors. This applies to comparisons for both base and overclocking frequencies.
• The Core i5 processors of the U and M series, even at the base frequency, should be noticeably faster than the Core i3 of similar series, and in the Boost mode they will go far ahead.
• In the Y series, the difference between processors at minimum frequencies is small, but with Turbo Boost overclocking, the Core i5 and Core i7 should go far ahead. Another thing is that the magnitude and, most importantly, the stability of overclocking are very dependent on the cooling efficiency. And with this, given the orientation of these processors to tablets (especially fanless ones), there may be problems.
• The Core i7 of the U-series is almost on par with the performance of the Core i5 of the M-series. There are other factors (it's harder to achieve stability due to less efficient cooling, and it's more expensive), but overall it's not a bad result.

As for the ratio of power consumption and performance rating, we can draw the following conclusions:

• Despite the increase in TDP when the processor switches to Boost mode, energy efficiency increases. This is because the relative increase in frequency is greater than the relative increase in TDP;
• Processors of different series (M, U, Y) are ranked not only by decreasing TDP, but also by increasing energy efficiency - for example, Y-series processors show greater energy efficiency than U-series processors;
• It is worth noting that with an increase in the number of cores, and hence the number of threads, energy efficiency also increases. This can be explained by the fact that only the processor cores themselves are doubled, but not the accompanying DMI, PCI Express and ICP controllers.

From the latter, an interesting conclusion can be drawn: if the application is well parallelized, then a quad-core processor will be more energy efficient than a dual-core one: it will finish calculations faster and return to idle mode. As a result, multi-core could be the next step in the fight for energy efficiency. In principle, this trend can also be noted in the ARM camp.

So, although the rating is purely theoretical, and it's not a fact that it accurately reflects the real alignment of forces, even it allows us to draw certain conclusions regarding the distribution of processors in the line, their energy efficiency and the ratio of these parameters to each other.

Haswell vs. Ivy Bridge

Although Haswell processors have been on the market for a long time, the presence of Ivy Bridge processors in ready-made solutions even now remains quite high. From the point of view of the consumer, there were no special revolutions during the transition to Haswell (although the increase in energy efficiency for some segments looks impressive), which raises questions: is it worth it to choose the fourth generation or can you get by with the third?

It is difficult to directly compare the fourth generation Core processors with the third, because the manufacturer has changed the TDP limits:

• the M series of the third generation Core has a TDP of 35W, while the fourth has a TDP of 37W;
• the U series of the third generation Core has a TDP of 17W, while the fourth has a TDP of 15W;
• the Y-series of the third generation Core has a TDP of 13W, while the fourth has a TDP of 11.5W.

And if for the ultra-mobile lines the TDP has dropped, then for the more productive M series it has even grown. However, let's try to make an approximate comparison:

• The top quad-core processor Core i7 of the third generation had a frequency of 3 (3.9) GHz, the fourth generation had the same 3 (3.9) GHz, that is, the difference in performance can only be due to architectural improvements - no more than 10%. Although, it is worth noting that with heavy use of FMA3, the fourth generation will outrun the third by 30-70%.
• The top dual-core Core i7 processors of the third generation of the M-series and U-series had frequencies of 2.9 (3.6) GHz and 2 (3.2) GHz, respectively, and the fourth - 2.9 (3.6) GHz and 2, 1(3.3) GHz. As you can see, the frequencies, if they have grown, are insignificant, so the performance level can grow only minimally, due to the optimization of the architecture. Again, if the software knows about FMA3 and knows how to actively use this extension, then the fourth generation will have a solid advantage.
• The top dual-core Core i5 processors of the third generation of the M-series and U-series had frequencies of 2.8 (3.5) GHz and 1.8 (2.8) GHz, respectively, and the fourth - 2.8 (3.5) GHz and 1.9(2.9) GHz. The situation is similar to the previous one.
• The top dual-core Core i3 processors of the third generation of the M-series and U-series had frequencies of 2.5 GHz and 1.8 GHz, respectively, and the fourth - 2.6 GHz and 2 GHz. The situation is repeating itself.
• The top dual-core Core i3, i5 and i7 processors of the third generation of the Y-series had frequencies of 1.4 GHz, 1.5 (2.3) GHz and 1.5 (2.6) GHz, respectively, and the fourth - 1.3 GHz, 1.4(1.9) GHz and 1.7(2.9) GHz.

In general, the clock speeds in the new generation have practically not increased, so a slight performance gain is obtained only by optimizing the architecture. The fourth generation Core will get a noticeable advantage when using software optimized for FMA3. Well, do not forget about a faster graphics core - optimization can bring a significant increase there.

As for the relative performance difference within the lines, the third and fourth generation Intel Core generations are close in this indicator.

Thus, we can conclude that in the new generation, Intel decided to lower TDP instead of increasing operating frequencies. As a result, the increase in the speed of work is lower than it could be, but it was possible to achieve an increase in energy efficiency.

Suitable Tasks for Different 4th Generation Intel Core Processors

Now that we have figured out the performance, we can roughly estimate what tasks this or that fourth-generation Core line is best suited for. Let's put the data in a table.

Series/line	Core i3	Core i5	Core i7
Mobile M	surfing the web office environment old and casual games	All of the above plus: professional environment at the edge of comfort	All of the above plus: professional environment (3D modeling, CAD, professional photo and video processing, etc.)
Ultramobile U	surfing the web office environment old and casual games	All of the above plus: corporate environment (e.g. accounting systems) undemanding PC games with discrete graphics professional environment on the verge of comfort (it is unlikely that you will be able to work comfortably in the same 3ds max)
Ultra-Mobile Y	surfing the web simple office environment old and casual games		office environment old and casual games

This table also clearly shows that, first of all, you should pay attention to the processor series (M, U, Y), and only then to the line (Core i3, i5, i7), since the line determines the ratio of processor performance only within the series, and performance varies markedly between series. This is clearly seen in the comparison of i3 U-series and i5 Y-series: the first in this case will be more productive than the second.

So what conclusions can be drawn from this table? Core i3 processors of any series, as we have already noted, are interesting primarily for their price. Therefore, it is worth paying attention to them if you are constrained by funds and are ready to put up with a loss in both performance and energy efficiency.

The mobile Core i7 stands apart due to architectural differences: four cores, eight threads and noticeably more L3 cache. As a result, it is able to work with resource-intensive professional applications and show an extremely high level of performance for a mobile system. But for this, the software must be optimized for the use a large number cores - it will not reveal its advantages in single-threaded software. And secondly, these processors require a bulky cooling system, i.e. they are installed only in large laptops with a large thickness, and they do not have much autonomy.

Core i5 mobile series provide a good level of performance, sufficient to perform not only home-office, but also some semi-professional tasks. For example, for photo and video processing. In all respects (energy consumption, heat generation, autonomy), these processors occupy an intermediate position between the Core i7 M-series and the ultra-mobile line. In general, this is a balanced solution, suitable for those who value performance more than a thin and light body.

The dual-core mobile Core i7 is about the same as the M-series Core i5, only slightly more powerful and usually noticeably more expensive.

Ultra-mobile Core i7 have about the same level of performance as mobile Core i5, but with caveats: if the cooling system can withstand prolonged operation at increased frequency. Yes, and they get pretty hot under load, which often leads to strong heating of the entire laptop case. Apparently, they are quite expensive, so their installation is justified only for top models. But they can be put in thin laptops and ultrabooks, providing a high level of performance with a thin body and good autonomy. This makes them an excellent choice for frequent travel professional users who value energy efficiency and light weight, but often require high performance.

Ultra-mobile Core i5 show lower performance compared to the "big brother" of the series, but they can cope with any office load, while they have good energy efficiency and are much more affordable. In general, this is a universal solution for users who do not work in resource-intensive applications, but are limited to office programs and the Internet, while still wanting to have a laptop/ultrabook that is suitable for traveling, i.e. light, light weight and long battery life.

Finally, the Y-series also stands apart. In terms of performance, its Core i7, with luck, will reach the ultra-mobile Core i5, but, by and large, no one expects this from it. For the Y series, the main thing is high energy efficiency and low heat generation, which makes it possible to create fanless systems as well. As for performance, the minimum acceptable level is sufficient, which does not cause irritation.

Briefly about Turbo Boost

In case some of our readers have forgotten how Turbo Boost overclocking technology works, we offer you a brief description of its work.

Roughly speaking, the Turbo Boost system can dynamically increase the processor frequency in excess of the set one due to the fact that it constantly monitors whether the processor is out of the normal operating modes.

The processor can only work in a certain temperature range, i.e. its performance depends on heating, and heating depends on the ability of the cooling system to effectively remove heat from it. But since it is not known in advance which cooling system the processor will work with in the user's system, two parameters are indicated for each processor model: the operating frequency and the amount of heat that must be removed from the processor at maximum load at this frequency. Since these parameters depend on the efficiency and proper operation of the cooling system, as well as external conditions (primarily ambient temperature), the manufacturer had to lower the frequency of the processor so that even under the most unfavorable operating conditions it would not lose stability. Turbo Boost technology monitors the internal parameters of the processor and allows it, if external conditions are favorable, to work at a higher frequency.

Intel originally explained that Turbo Boost technology uses "thermal inertia effect". Most of the time in modern systems, the processor is idle, but from time to time for a short period of time it is required to perform at its maximum. If at this moment we strongly increase the frequency of the processor, then it will cope with the task faster and return to the idle state earlier. At the same time, the processor temperature does not rise immediately, but gradually, so during short-term operation at a very high frequency, the processor will not have time to heat up so as to go beyond the safe limits.

In reality, it quickly became clear that with a good cooling system, the processor is able to work under load even at an increased frequency indefinitely. Thus, for a long time, the maximum overclocking frequency was absolutely working, and the processor returned to the nominal value only in extreme cases or if the manufacturer made a low-quality cooling system for a particular laptop.

In order to prevent overheating and failure of the processor, the Turbo Boost system in the modern implementation constantly monitors the following parameters of its operation:

• chip temperature;
• consumed current;
• power consumption;
• the number of loaded components.

Modern systems based on Ivy Bridge are capable of operating at an increased frequency in almost all modes, except for the simultaneous serious load on the central processor and graphics. As for Intel Haswell, we do not yet have sufficient statistics on the behavior of this platform under overclocking.

Note. Author: It is worth noting that the temperature of the chip indirectly affects the power consumption - this effect becomes apparent upon closer examination physical device the crystal itself, since the electrical resistance of semiconductor materials increases with increasing temperature, and this in turn leads to an increase in power consumption. Thus, the processor at a temperature of 90 degrees will consume more electricity than at a temperature of 40 degrees. And since the processor "warms up" both the PCB of the motherboard with tracks and the surrounding components, their loss of electricity to overcome higher resistance also affects power consumption. This conclusion is easily confirmed by overclocking both "in the air" and extreme. All overclockers know that a more productive cooler allows you to get additional megahertz, and the effect of superconductivity of conductors at a temperature close to absolute zero, when the electrical resistance tends to zero, is familiar to everyone from school physics. That is why, when overclocked with liquid nitrogen cooling, it is possible to achieve such high frequencies. Returning to the dependence of electrical resistance on temperature, we can also say that to some extent the processor also heats itself up: when the temperature rises, when the cooling system cannot cope, the electrical resistance also increases, which in turn increases power consumption. And this leads to an increase in heat dissipation, which leads to an increase in temperature ... In addition, do not forget that high temperatures shorten the life of the processor. Although manufacturers claim relatively high maximum temperatures for chips, it is still worth keeping the temperature as low as possible.

By the way, it is likely that "turning" the fan at higher speeds, when due to it the system's power consumption increases, is more profitable in terms of power consumption than having a processor with a high temperature, which will lead to power losses due to increased resistance.

As you can see, temperature may not be a direct limiting factor for Turbo Boost, that is, the processor will have a completely acceptable temperature and not go into throttling, but it indirectly affects another limiting factor - power consumption. Therefore, you should not forget about the temperature.

Summing up, Turbo Boost technology allows, under favorable external operating conditions, to increase the processor frequency beyond the guaranteed nominal value and thus provide a much higher level of performance. This feature is especially valuable in mobile applications where it allows for a good balance between performance and heat.

But it should be remembered that the other side of the coin is the inability to estimate (predict) the net performance of the processor, because it will depend on external factors. This is probably one of the reasons for the appearance of processors with "8" at the end of the model name - with "raised" nominal operating frequencies and increased TDP because of this. They are intended for those products for which stable high performance under load is more important than energy efficiency.

The second part of the article provides detailed description all current series and lines of Intel Haswell processors, including the technical specifications of all available processors. And also conclusions are drawn about the applicability of certain models.

You shouldn't expect any significant performance gains from mainstream Haswell quad-cores (unless, of course, the software is adapted to new sets of processor instructions), their strong point is reduced power consumption and price-performance ratio. However, when it comes to top-end hardware, the "win at all costs" approach is still relevant.

Mainstream Intel CPUs became dual-core in 2006 with the advent of the Core 2 Quad. Quad-cores "went to the people" in 2008, with the transition to Nehalem and LGA1156, and in the near future the number of cores will not change - at least until 2014, when the release of Broadwell chips is planned, which will be produced according to the 14 nm process technology. This decision is quite justified, given that the advantages provided by additional cores are still not claimed by most programs - the effect of a more powerful GPU or additional cache memory will be more significant. However, with processors of the highest price range, everything was not so simple, because. Workstation and server software is highly optimized for multi-core processors, and both increasing the number of cores and increasing the output of each core can bring results.

Now, thanks to our sources at the IDF, we can shed some light on the situation. As our readers already know, by the middle of next year, the top processor for server systems, the 10-core 2.4 GHz Xeon E7 4800 of the “Westmere EX” family, will be replaced by the representative of the “Ivy Bridge EX” architecture Xeon E7 4800 v2, which has 15 cores and operating at frequencies from 2.2 GHz, which will be installed in the LGA2011 socket, but with a different pinout. In mid-2014, it can be replaced with a 16-20 core Xeon E7 4800 / 8800 v3 (Haswell EX architecture), and the socket will remain the same. It will be followed by the Xeon E7 4800 / 8800 v4 (Broadwell EX architecture), which will be released as early as 2015. The last three models have a common feature

In the form of a QPI bus with three lanes - Westmere has four - which will negatively affect the ability to interact with Xeon Phi coprocessors or the ability to access system memory at full speed, which can be useful when connecting an FPGA.

The most interesting case is the dual-processor configuration, because it has a lot in common (at least the socket and chipset) with hardware positioned as home high-end. At the moment the situation looks like this:

The current 8-core Xeon E5 2600 / 4600 (Sandy Bridge EP) will be replaced in the middle of next year by a 10-core Xeon E5 2600 / 4600 v2 (Ivy Bridge EP) that will use the same socket. The next upgrade is planned for 2014 - Xeon E5 2600 / 4600 v3 (Haswell EP) will have as many as 14 cores and a 14-channel DDR4-2133 controller, replacing the DDR3 used in Ivy Bridge EP systems and dual QPI channels with bandwidth about 9.6 GT/s, slightly more than

Now, which will be installed in a socket similar to 2011 in size but with a different pinout. But why increase the number of cores even further if the EX-series components are already the benchmark for performance?

Two main reasons come to mind. Firstly, the increase in specific performance per core provided by Haswell is not so great - about 10% compared to Ivy Bridge, unless the software is adapted to use new processor instructions that may not be used in all algorithms. Which is not surprising, since the main focus of Haswell's design was on reducing power consumption (ultrabooks!). So where do you get productivity gains to spur sales?

On the other hand, lower power consumption allows you to fit more cores on a single chip at the same TDP. Thus, a 14-core processor fits into the limit of 145 W (for servers) and 160 (for workstations), while the volume of the L3 cache per core remains the same - 2.5 MB. Whether such an extensive growth strategy is justified is debatable. Within the same TDP, I'd rather see a processor with fewer cores but more cache per core and more

High clock speeds, and a significant number of owners of high-end machines

We would agree with me, because the ability of the software to use a larger number of threads has increased slightly over the 5 generations of Intel processors that have changed. One way or another, even with 14 cores, the new models should have at least the same clock speeds as their Ivy Bridge predecessors within the same TDP, which means at least 3.2 GHz for top-end workstation models.

At these frequencies, the theoretical peak performance per socket will be 3/4 teraflops double precision, so one dual processor work station a sample of the middle of 2014 will produce 1.5 teraflops “on the mountain”. Add to that an 8-channel DDR4 memory controller and Nvidia Maxwell has a serious competitor. After all, if the CPU has sufficient

With power and for it, you don’t need to rewrite software almost from scratch, why not use it? In any case, optimizing applications for GPGPU with their huge number of threads will also lead to the fact that not a single core in multi-core processors will be idle. Also, do not forget that Intel is not the only company on the market, and its competitor has experience in the development of combined computing units, which, in light of the convergence of CPU and GPU, may be AMD's chance to catch up. Looking forward to the Opteron APU?

If AMD has been engaged in guerrilla warfare on the processor front for a long time, then as far as the video adapter market is concerned, it has so far only had to fight with its “sworn friend” Nvidia. But the situation may soon change.

The next generation of Intel's architecture, codenamed Haswell, is not just another "tick" in the measured cycle of technology improvement by Intel, it is a new stage in its activities. The stage at which it becomes a serious threat to both AMD and Nvidia. For the first time, Intel is poised to challenge both of them in the mainstream graphics market while undermining Nvidia's position in the GPGPU business. At the same time, low-power and energy-efficient solutions (ULV versions of the processors have a TDP of 10 W) will be serious competitors for SoCs based on the second generation Brazos platform from AMD (codenamed Kabini), as well as any laptops based on ARM processors based on Windows 8 which companies like Qualcomm can bring to market.

Let's take a closer look at this architecture, starting with the CPU.

Wider, bigger, faster.

Haswell is a logical continuation of the microarchitectural improvements first introduced by Intel in Sandy Bridge. The new chip received support for the second generation of the Advanced Vector Extensions (AVX2) processor instruction set, which doubles the peak FPU core bandwidth. The L1 and L2 cache bandwidth has been doubled to keep the execution units busy, and the integer and FPU register files have been enlarged. The branch prediction performance has also been improved. Haswell's single-thread performance in real-life tasks on unoptimized code is expected to improve by 10-15%. If there is an optimization for AVX2, the gap will be much larger - AVX2 algorithms include support for vectorization of integer values, which is not in the first version.

The FPU power upgrades and additional AVX2 functionality will go a long way in boosting floating point performance. The processor is capable of performing up to 32 standard precision floating point operations on a single core and 16 double precision floating point operations. That is twice as much as Sandy Bridge; theoretically, an eight-core processor based on Haswell architecture with a clock speed of 3.8 GHz will produce 972.8 gigaflops at the standard level of accuracy and 486.4 gigaflops at doubled accuracy. And although the current generation of GPUs show even better results, Intel has a trump card in its sleeve - x86 compatibility. Intel trashed the history of RISC supercomputer vendors in the 1990s and early 2000s simply because their processors were "good enough," and the same thing is now threatening Nvidia and its GPGPU concept. L1/L2 cache throughput has increased drastically, throughput bus L1 is also doubled. All the extra bandwidth is meant to keep the AVX2 units from being idle; It is expected that Haswell will show a fairly close correspondence between theoretical performance values ​​and the speed of real-world tasks.

And while the green-flag team is likely to retain its net performance advantage, a Haswell quad-core that reaches 4 GHz in turbo mode will deliver 256 gigaflops for double precision operations (512 gigaflops for standard precision). This level of standard-precision performance is very close to the Nvidia GT 640. And since double-precision performance on Nvidia's consumer cards has always been lackluster, Haswell's quad-core processors could well outperform Nvidia's GTX 680 and possibly match the GTX 580 in double precision operations.

Nvidia may win the battle for high-end users, but at the cost of losing in other areas if Intel decides to compete seriously. Even worse, don't forget the fact that every PC equipped with an Nvidia graphics card comes with an Intel accelerator by default. Undoubtedly, Intel is going to play on a potential connection with Xeon Phi, given that the company's three IDF workshops addressed the vectorization issue and touched on both Haswell and Xeon Phi.

Haswell GPU increases pressure on Nvidia, AMD.

The Haswell GPU is essentially a modified version of the cores currently used in Ivy Bridge. The main changes are in the shader array - Intel will offer Haswell in versions with a block that includes 10, 20 or 40 shaders (GT1, GT2, GT3 respectively). The chip will also be offered in variants that include up to 128MB of onboard memory, which gives each GPU a small dedicated amount of memory. Intel hasn't been very vocal about the changes it's made to the GPU, but the company has said that the performance gain shown by the new GT3 configuration compared to the performance graphics core HD 4000 built into Ivy Bridge is up to 200%.

Even if we take this information with a healthy dose of skepticism, it still does not bode well for AMD and Nvidia. According to Anandtech, the Trinity GPU is on average 18% faster than Liano in games. Compared to Sandy Bridge, Trinity is almost 80% faster. If you compare it with Ivy Bridge, the advantage is reduced to 20%. Given what we already know about the Haswell GPU and its predicted performance, it shouldn't be too hard for Intel to deliver a 30-50% performance boost in real-world games. If this happens, Trinity will lose its status as the fastest integrated GPU on the market, moving into the middle class, and AMD will lose its trump card in the video card market, which it has played since the launch of the AMD 780G chipset four years ago.

Thus, Sunnyvale has little to no wiggle room. The 28nm Kaveri APU, equipped with the next-generation Radeon HD 7000-based graphics core, and new processors based on the Steamroller architecture have yet to receive an announcement date. This means we may not see them until the end of 2013, and that's if production goes smoothly. AMD is likely to offer an update - something like Trinity 2.0 - to keep Haswell under pressure, but slightly higher clocks are unlikely to save the day for AMD.

AMD's last bastions are markets that Intel is generally not interested in. This is a precarious position for any company that dreams of challenging the market leader; AMD simply cannot afford to spend enough on R&D to catch up with its longtime rival. And it's hardly worth resting on Nvidia's laurels. Intel's plans make it clear that the company is absolutely committed to minimizing the value of individual GPUs by using integrated solutions where possible and supporting the transition to ever smaller form factors where this is (yet) not possible.

Thus, if Haswell is not a complete failure, it is he, and not Kaveri, who will become the new reference point for enthusiasts. This 10W chip won't be able to compete directly with potential rivals - Tegra 4 based tablets are a Bay Trail challenge, a 22nm SoC based on Atom.

No, Haswell won't bankrupt AMD or scare Nvidia into leaving Tesla - but if Intel's plan doesn't fail, both companies will be squeezed into niche product markets. AMD takes this move for a living - it is squeezed into the markets of low-end products that are of no value to Intel. Nvidia will now have to work very hard to convince OEMs to find a place for a separate GPU in their computers, although Intel's marketing policy and customer preferences pull in the other direction. Enthusiast preferences, historically weak Intel driver support, and Nvidia's brand strength will help, but the dustbin of IT industry history is full of companies that thought their brand would keep users even if their product specifications were worse than those of the competition. Enthusiasts are only interested in performance, not what company is behind it.

However, so far we have only talked about solutions for enthusiasts and desktop solutions, which is a little illogical, given the growing market share of laptops and ultrabooks by leaps and bounds. Many improvements to the Haswell architecture were aimed specifically at optimizing for them. What exactly? Let's figure it out.

Integration

Haswell for ultrabooks will have a TDP of 15W, almost like the Sandy Bridge that ultrabooks are based on today. The big news here is that Intel will be moving the PCH (Platform Controller Hub) to the same substrate as the processor, so that the Ultrabook version of Haswell will contain all of the platform's components on a single chip. Sandy Bridge consisted of two components supplied by Intel - a processor and a PCH, while Haswell would be a single MCP (multi-chip package). This means that two computational chips will be placed on one substrate, which is often a prerequisite for combining the chips themselves (perhaps after switching to 14 nm process technology?). A single MCP will take up less footprint than the CPU + PCH bundle currently in use today, allowing for less dense motherboard layouts (or smaller boards), and possibly even bigger batteries in ultrabooks. This is a significant step and shows that the line between tablet and ultrabook hardware is starting to blur.

It's worth noting that Ultrabook Haswell can have a maximum of two cores, although laptop and desktop versions can have more.

Energy Efficient Memory and New Socket

The list of supported memory has also been adjusted to optimize power consumption. All three versions of Haswell will support DDR3L, although the desktop version can optionally use regular DDR3 and the ultrabook version can use LPDDR3. All three variants are equipped with two memory channels.

It's important to note that despite Haswell's focus on energy efficiency, the architecture scales just as well as Sandy Bridge (desktop components with 95W TDPs will be available, though a direct comparison of thermals may not be entirely accurate). Which is logical, since a single efficient architecture can usually cover a wide range of TDPs without sacrificing efficiency.

Other Haswell features include built-in voltage regulators (which should simplify motherboard layouts), support for the AVX 2.0 instruction set and, of course, AES-NI and Hyper-Threading. The exit of Haswell will also entail a socket change: to desktop computers prescribe LGA-1150.

Conclusion

In fact, there is little surprise here. Everyone knew integrated graphics cores would keep getting faster, though it's still unclear exactly how powerful the GT3 variant will be. The real test of its capabilities will be the decision of manufacturing companies whether to continue to install discrete video adapters in their products (Apple's example in relation to, say, Macbook Pro). As far as we know, Intel's plans to strengthen its position in the integrated graphics segment were met with full approval in Cupertino.

Continued integration of new features into a single chip is a significant step forward for high-end x86 CPUs, and all indications are that the difference between tablets and laptops will continue to blur in 2013.

The types of intel processors are numerous. Haswell is the name of the fourth generation of equipment that used an innovative architecture.

Especially for them, a family of new chipsets of the eighth series has been developed. Work with SSD is optimized. The release of the architecture took place in early June 2013.

Haswell Review

Since 2013, many processor models have been developed. The standalone processor was positioned by developers for use in laptops, ultrabooks and tablets due to its low power consumption. Performance will improve, allowing developers to imagine Haswell as best processors intel for mobile devices this moment. Core i3 haswell dual-core processors are available in three varieties:

1. i3-4340;
2. i3-4330;
3. i3-4130.

They differ in clock frequency, which for the three models is respectively 3.6, 3.5, 3.4 GHz. The new graphics core for the first two models is represented by HD Graphics 4600, and for the third - HD Graphics 4400. The frequency of these cores is 1150 MHz for all. L3 - cache 4, 4 and 3 MB, respectively. The price differs slightly - for the first option - $160, for the second - $150 and for the third - $130.

The haswell quad-core i5 is equipped with an HD Graphics 4600 graphics core. The clock frequency is 3.2 GHz, with a turbo boost of 3.6. 6 MB cache. Heat dissipation is low, so that active use no additional cooler required.

But i7 processor is superior to i3 or i5. Represented by i7-4770K, i7-4770, i7-4770S, i7-4770T and i7-4765T. The first two run on a quad-core processor with 8 threads, while the rest have four.

The clock frequency is the lowest latest model and is equal to 2 GHz, the highest of the first is 3.5 GHz. Cache 8 MB

Haswell Features

Haswell is the name of the new processor architecture, processors based on it are also called. The computing core of the device has undergone changes compared to the previous version. The preprocessor is almost unchanged. The core decoder is four-channel, and since the average command length is 4 bytes, it can simultaneously process up to 16 bytes. It consists of four simple decoders and one complex one. Instructions are decoded using Macro-Fusion and Micro-Fusion technologies.

The 8-channel decoded op cache stores 1500 micro-ops in 4 bytes. Each of the 8 banks has 32 cache lines, which include 6 micro-operations each. The point of such a bank is not to re-decode, but to pull the already decoded operation directly from the cache.

Changed execution blocks in the kernel. The number of ports has been increased to 8. Now up to 8 micro-ops are performed in one cycle. A new set of instructions has been introduced.

Device performance tests were conducted on the basis of Windows and Android. Testing of the intel core i7 - 4770 was carried out with basic processes and applications, and the time taken to complete a given operation was taken as an indicator. As a result of the test on non-gaming applications, the indices of Intel Haswell processors turned out to be higher than in previous models.

The largest increase in the indicator in Photoshop, Adobe Premier Pro, etc.

With the help of 3DMark Professional, a test was carried out for the operation of gaming applications. The results show that there is progress in the work of the graphics subsystem. Game without discrete graphics card impossible. Integrated graphics processor is not good.

Benefits of the Haswell Processor

Haswell is a generation of Intel Core that has quite a few opponents. They find flaws in it, such as being overpriced or having to update the platform too often. However, this equipment has a number of advantages. These are high efficiency and productivity, and a functional platform, etc.

• The main advantage that the processor has is the integrated graphics core. It has become competitive. Now you can support multiple monitors and a significant increase in performance;
• The device has increased energy efficiency. Compared to previous versions, it was possible to reduce it by 5 watts in idle mode. It's not such a big difference for a desktop PC, but a significant one if you're opting for a laptop or ultrabook. Power consumption under load is low;
• Productivity has increased by 5 - 10% compared to previous generations. It differs depending on the test conditions. In some cases, it may be higher or lower. The difference is not significant enough to upgrade an existing previous generation system, but significant if you choose a haswell processor over a much outdated one;
• The system for overclocking the processor through the base frequency has become more flexible. Thus, the developers responded to the claims of users previous versions devices.

Processors intel pentium haswell are designed primarily for laptop applications. Powerful hardware for desktop PCs is not yet available, while laptops cannot achieve the highest clock speeds, huge caches and the use of full-fledged 8 cores. Thus, fans of stationary PCs will have to wait for other developments.

Intel can be reproached for anything - from overpricing and the need for frequent platform changes to blocking overclocking tools in its younger models. But one thing cannot be taken away from the semiconductor giant: for many years now, the release of new products has been strictly following the so-called "Tick-Tock" strategy, where for every "Tick" there is a transition to a new, more subtle technological process of production, and an update falls on "Tick" microarchitecture. Last year, Intel announced 22nm Ivy Bridge semiconductor dies, which replaced their predecessors - 32nm Sandy Bridge. The differences between the representatives of the two generations were in the modernization of the graphics subsystem, while the computing cores have undergone minimal changes. At the same time, the transition to a thin technological process was by no means painless, as a result of which the overclocking potential of the 22-nm Ivy Bridge was not as impressive as that of its predecessors. Needless to say, enthusiasts and advanced users were looking forward to the official announcement of the carriers of the new microarchitecture, codenamed Haswell. Even before the announcement, a variety of hypotheses circulated on the Internet, attributing unprecedented overclocking potential combined with the highest performance to the latest Intel CPUs. And now we can finally pull back the veil of secrecy and present detailed overview fourth-generation Intel Core CPU - Core i7-4770K.

The new family includes a variety of products, from energy-efficient models for ultra-thin laptops and All-in-One systems, to classic processors with an optimal ratio of performance and power consumption, as well as modifications with unlocked multipliers designed for advanced users and overclockers.

Haswell microarchitecture features

The manufacturer sensibly reasoned that in most home use scenarios, and in many areas of professional use, four computing cores are more than enough, therefore, the Core i5 and Core i7 processors are based on Haswell quad-core semiconductor crystals. The use of a thin 22-nm lithographic process technology made it possible to fit 1400 million semiconductor devices in an area of ​​177 square meters. mm. The transistors themselves have a three-dimensional design (Tri-Gate), which ensures their small physical dimensions and minimizes leakage currents. This design was first used in the Ivy Bridge processors, which pioneered the development of the 22-nm process technology. In addition to reducing manufacturing costs, these measures have reduced the supply voltage by up to 20% compared to 32nm Sandy Bridge.

The semiconductor chip of the Haswell processor includes four processing cores, a graphics accelerator, a third-level cache array, and a “system agent”, which includes a dual-channel DDR3 RAM controller, DMI and PCI Express bus controllers, and digital image transmitters. The processor cores and the integrated video card use a shared shared cache memory, and the high-speed ring data bus, which first appeared in the Intel Sandy Bridge processors, is used to communicate between the internal blocks.

The Haswell computing cores themselves have undergone a minimum of changes compared to Ivy Bridge, in any case, the design of the computing pipeline has remained the same, and all the improvements are in the nature of optimizations. For example, the selection and branch prediction mechanisms have been improved, the throughput of the task manager has been increased by adding two additional ports, the size of the TLB buffer (translation lookaside buffer) in the L2 cache has been optimized, and latency has been reduced during the operation of virtualization technologies. The operation of blocks processing vector instructions has undergone minor changes, which received support for new AVX2 instructions that speed up cryptography, hashing, and multimedia processing. Also, compared to Ivy Bridge, the depth of data fetching from L1 and L2 caches per clock has doubled, which means that in optimized tasks, new Haswell processors can be noticeably faster than their predecessors.

As for the graphics component of Haswell processors, most Core i5 and Core i7 desktop modifications will use the Intel HD Graphics 4600 video core, which contains 20 unified shader processors, two rasterization units and four texture modules. The graphics accelerator is compatible with DirectX 11, and support for the OpenCL API and DirectCompute 5.0 gives a boost in non-graphics computing. The video core also includes a hardware decoding unit Quick Sync, the use of which provides an increase in the speed of processing video content, and as a nice addition, we note support for simultaneous display of images on three monitors. The distinguishing feature of Intel HD Graphics 4 Series is their modular design, which makes it easy to scale the number of functional blocks, creating on their basis both entry-level solutions and fairly powerful video accelerators.

Controller random access memory Haswell processors inherited from Ivy Bridge almost unchanged. It supports two channels of DDR3 RAM with frequencies of 1333 MHz and 1600 MHz, including low-voltage DDR3L. However, no one interferes with the operation of high-frequency modules, for this the controller supports a large set of multipliers that are multiples of the effective 200 and 266 MHz. To communicate with the chipset, the DMI 2.0 bus is used, the bandwidth of which reaches 20 Gb / s. The connection of discrete graphics accelerators is provided by the PCI Express 3.0 bus controller, 16 lines of which can be flexibly configured to organize systems from several video cards.

But the most unexpected of the innovations in the architecture of Intel Haswell was the placement of an integrated voltage regulator on a semiconductor chip! According to the developers, this is the only way to achieve the most flexible power management, which is the key to high energy efficiency. It is not yet clear how this will affect the overclocking potential, but it is already quite clear that the VRM of the motherboard now requires only two voltages: Vddq, which is necessary to power the RAM modules, and Vccin, from which the integrated regulator generates all the voltages necessary for operation. internal blocks of the central processor.

The nominal value of Vccin is about 1.8 V, but, if necessary, for example, during acceleration using liquid nitrogen, it can be increased to 3 V. The integral regulator provides two voltage control modes: static, in which the user specifies the desired value in explicit form, and dynamic, when the increase to the standard value is set. Obviously, the first method will be in demand among overclockers, while the second will provide the necessary voltage regardless of the operating mode. Obviously, such a radical change in the power subsystem required a transition to the new Socket LGA1150 processor socket, which is part of the new Intel platform - Lynx Point.

Lynx Point Platform

The Lynx Point platform is based on Intel 8-series chipsets. The updated system logic has retained a single-chip layout, while the functionality has expanded somewhat compared to its predecessors. For convenience, the comparative characteristics of the Intel 7th and 8th series chipsets are shown in the following illustration.

The total number of SATA connectors has not changed, there are still six, but they are all compatible with the high-speed SATA 6Gb / s interface. The number of USB 3.0 ports has increased from four to six, while the total number is the same 14 pieces. The 8-series chipsets have completed the transition to the xHCI (eXtended Host Controller Intarface) controller, which provides enhanced control over data transfer between system board and the periphery. Also, the Lynx Point platform does not support the PCI bus, which was found in modifications B and Q of the Intel 7-series system logic.

One of the key differences of the Lynx Point platform from its predecessors is the change in the approach to the formation of clock frequencies for individual functional blocks of the processor and motherboard. In the Intel 8-series system logic, two such signals are generated: a fixed frequency of 100 MHz, from which the chipset controllers are synchronized, and a controlled BCLK, from which the entire ensemble of frequencies necessary for the operation of the internal blocks of the central processor is formed through a system of multipliers.

As you remember, the main complaint about the LGA1155 platform from overclockers was the lack of a margin for increasing BCLK due to the instability of the DMI and PCI Express bus controllers at higher frequencies. AT Intel chipsets The 8th series uses x1.00, x1.25 and x1.67 multipliers to form the base frequency of the processor and its blocks. A similar solution can be found in the LGA2011 platform. Now BCLK can easily be raised to 125/167 MHz (± 5%) without affecting the sensitive components of the system.

As we already mentioned, Haswell processors received a new Socket LGA1150, which looks almost indistinguishable from the usual LGA1155. The location and dimensions of the holes for mounting the cooling system are identical, so coolers compatible with the LGA1155 and LGA1156 platforms are suitable for motherboards for Intel Haswell.

But, of course, it will not be possible to install processors of previous generations in the new socket due to a different arrangement of mechanical keys and a different number of contact pads. Processor Intel Core i7-4770K

At the time of its announcement, the LGA1150 CPU product line will consist of quad-core Core i5 and Core i7, which are distinguished by support for Hyper Threading technology, which allows two computational threads to be executed on one logical core. As usual, by varying clock frequencies and TDP values, the manufacturer has created a whole range of models based on a single crystal:

	Intel Core i7-4770/ i7-4770K*	Intel Core i7-4770S	Intel Core i7-4770T	Intel Core i7-4765T	Intel Core i5-4670/ i7-4670K*	Intel Core i5-4670S	Intel Core i5-4670T	Intel Core i5-4570	Intel Core i5-4570S	Intel Core i5-4570T
Family	Haswell	Haswell	Haswell	Haswell	Haswell	Haswell	Haswell	Haswell	Haswell	Haswell
connector	LGA1150	LGA1150	LGA1150	LGA1150	LGA1150	LGA1150	LGA1150	LGA1150	LGA1150	LGA1150
Process technology CPU, nm	22	22	22	22	22	22	22	22	22	22
Number of cores	4 (8 threads)	4 (8 threads)	4 (8 threads)	2 (4 threads)	4 (4 threads)	4 (4 threads)	4 (4 threads)	4 (4 threads)	4 (4 threads)	2 (4 threads)
Rated frequency, GHz	3,4/3,5*	3,1	2,5	2,0	3,4	3,1	2,3	3,2	2,9	2,9
Turbo Boost frequency, GHz	3,9	3,9	3,7	3,0	3,8	3,8	3,3	3,6	3,6	3,6
Volume of L3 cache, MB	8	8	8	8	6	6	6	6	6	4
Graphics core	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600	GMA HD 4600
	1200/1250*	1200	1200	1200	1200	1200	1200	1150	1150	1150
memory channels	2	2	2	2	2	2	2	2	2	2
Supported memory type	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600	DDR3-1333 DDR3-1600
Hyper Threading	+	+	+	+	-	-	-	-	-	+
AES-NI	+	+	+	+	+	+	+	+	+	+
Intel vPro	+	+	+	+	+	+	+	+	+	+
TDP, W	84	65	45	35	84	65	45	84	65	35
Recommended cost, $	303/339*	303	303	303	213/242*	213	213	192	192	192

* — the multiplier is unlocked for increase; for K-series models.

Among products that differ in clock speeds and TDP, users can easily find exactly those models that the best way correspond to the assigned tasks. Fans of energy-efficient and compact system units will be interested in modifications with the letter "T", which have the best efficiency. Users collecting universal system unit, most likely, they will pay attention to the “S” series models, which offer a balance between speed and moderate power consumption, but for overclockers and overclocking professionals there are “K” series processors with an unlocked multiplier. Standing apart is the energy-efficient Core i5-4570T model, in which the number of cores is reduced to two, and the L3 cache array is reduced to 4 MB. As for retail prices, they almost do not differ from the equal-frequency Ivy Bridge, however, there is no doubt that in the near future the Haswell lineup will be supplemented by junior Core i3 and Pentium models.

For comprehensive testing, the senior from Haswell, the Core i7-4770K, was delivered to our laboratory. This processor has an unlocked multiplier, which means it is best suited for overclocking experiments. Alas, the prototype came to us without any delivery kit, so we cannot judge the design of the cooling system and the design of the retail package.

The appearance of the novelty is almost indistinguishable from its predecessors, Haswell can be identified by marking and the absence of a cutout in the lower part of the metal heat-distributor cover, which covers the semiconductor crystal. On the reverse side there are metal contacts and electronic components, and it is not difficult to find a LGA1150 model among other processors, thanks to the darker shade of the substrate textolite and fewer auxiliary elements on the "belly".

Intel Core i7-2600K (left), Core i7-3770K, Core i7-4770K (right)

The nominal frequency of the computing cores of the novelty is 3500 MHz, but in this mode the processor operates at maximum computing load.

Most of the time, the processor runs at higher frequencies, which, thanks to Intel technologies Turbo Boost dynamically change depending on the load of the computing cores and the overall level of power consumption of the central processor. Thus, in applications optimized for multi-threaded computing, the Core i7-4770K, as a rule, operates at frequencies of 3600-3700 MHz.

If the program code does not work efficiently on multi-core central processors, the involved computing units are overclocked to frequencies of 3800-3900 MHz, while the heat dissipation remains within the TDP.

During idle times, the clock frequency of the computing cores decreases to 800 MHz, which should have the most positive effect on the heating level.

As far as voltage is concerned, current versions monitoring programs fix values ​​​​in the range of 1.069-1.104 V, which is very similar to the truth, in any case, 22-nm predecessors had a similar order of values. By the way, as you remember, many users scolded Intel for using a low-efficiency thermal interface between the semiconductor chip and the heat spreader cover in Ivy Bridge, as a result of which 22-nm processors showed elevated temperatures during overclocking. Has the situation changed with the release of Haswell - we will find out right now, during the check overclocking potential Intel Core i7-4770K!

Overclocking potential

Before proceeding to the assessment of the safety margin of the Core i7-4770K, let's look at the frequency shaping scheme in Haswell processors, the overclocking procedure of which slightly, but still differs from that of Ivy Bridge and Sandy Bridge. Primarily, maximum value multiplier for processor cores now equals x80, this fact will undoubtedly be appreciated by professional overclockers working with cryogenic cooling systems. Then, a separate multiplier appeared, controlling the frequency of the internal ring bus. Its value can be less than or equal to the multiplication factor of computing cores. And finally, thanks to the introduction of an additional PEG / DMI multiplier, it became possible to increase the base frequency to 125 or 167 MHz, without compromising the stability of the PCI Express and DMI buses.

Most likely, not all motherboards and processors will allow you to set the base frequency to 167 MHz, while increasing the BCLK to 125 MHz will be a real and effective way to overclock younger Haswells, whose multiplier is locked up. Our Core i7-4770K has a free multiplier, so this advantage was used in our overclocking experiments. Due to the lack of Haswell overclocking statistics, we took advantage of the experience gained while working with Ivy Bridge processors. The power on the processor cores was increased to 1.24 V, the internal Vring voltage was increased by +0.1 V. The Internal PLL Overvoltage function was set to Enable, the base frequency was fixed at 100 MHz, and the power limits of Turbo Boost technology were increased to 500 W. With these settings, the processor passed the Linpack stress test at 4500 MHz, while the ring bus frequency was 4200 MHz.

We draw your attention to the evidence temperature sensors, according to which the hottest core warmed up to 97 ° C, although we used one of the best air coolers - Thermalright Silver Arrow to remove heat. Despite the high temperature, the processor remained completely stable, but all further overclocking experiments had to be stopped, since the slightest increase in voltage led to overheating, which caused BSOD. Let's hope that we just got an unsuccessful copy of the Core i7-4770K, while for the most part Intel processors in the LGA1150 version will show much better overclocking results.

It turns out that Haswell inherited from its ancestor Ivy Bridge the same “hot temper”, which even the best of air coolers can hardly cope with overclocking. By the way, the Core i7-3770K provided for testing in overclocking to 4700 MHz at a voltage of 1.312 V showed a similar thermal regime, easily warming up to 91 C and above.

It seems that the Core i7-4770K and Core i7-3770K use the same not very efficient thermal interface between the semiconductor crystal and the heat spreader cover, which, coupled with the small area of ​​the processor core, leads to high temperatures during overclocking. test bench

For a comprehensive assessment of the performance of the Intel Core i7-4770K, we chose the older models of Ivy Bridge and Sandy Bridge processors - Core i7-3770K and Core i7-2600K as rivals. Thus, we will be able to track the increase in speed when changing generations, as well as evaluate the scalability of performance during overclocking. But first, let's get acquainted with the technical characteristics of the participants in today's testing.

	Core i7-4770K	Core i7-3770K	Core i7-2600K
Core	Haswell	Ivy Bridge	Sandy Bridge
Number of transistors, million	1400	1400	995
Crystal area, sq. mm	177	160	216
Number of cores (threads)	4 (8)	4(8)	4(8)
Process technology, nm	22	22	32
Frequency, MHz	3500	3500	3400
Maximum frequency in Turbo Boost mode, MHz	3900	3900	3800
Factor	39*	39*	38*
L1 cache, KB	4 x (32+32)	4 x (32+32)	4 x (32+32)
L2 cache, KB	4x256	4x256	4x256
L3 cache, KB	8192	8192	8192
Supported memory	DDR3-1600	DDR3-1600	DDR3-1333
Integrated graphics	Intel HD Graphics 4600	Intel HD Graphics 4000	Intel HD Graphics 3000
Graphics core frequency, MHz	1250	1150	1350
TDP, W	84	77	95

* - Turbo Boost frequency

All three processors have four processing cores, support Hyper Threading, and also have the same cache memory organization. Meanwhile, the clock speeds of the Core i7-4770K and Core i7-3770K are exactly the same, while their 32nm brother is 100 MHz behind both at stock and Turbo Boost. In a word, the characteristics of the rivals are very close, so in the nominal mode we expect to get similar performance results.

As a basis for test stand LGA1150 used system ASUS board Sabertooth Z87 (UEFI 3009 dated 05/24/2013), a detailed review of which we will publish in the very near future.

For testing Ivy Bridge, we took the MSI Z77 MPOWER motherboard (UEFI Setup 17.8 dated 04/23/2013), and in experiments with the Intel Core i7-2600K processor, we used the ASRock Z77 Extreme6 (UEFI Setup 2.60 dated 01/23/2013), which has proven itself in overclocking Sandy Bridge processors.

Common to all test benches were the following components:

• cooler: Thermalright Silver Arrow (fan 140 mm, 1300 rpm);
• Memory: G.Skill TridentX F3-2400C10D-8GTX (2x4 GB, DDR3-2400, CL10-12-12-31);
• video card: ASUS HD7950-DC2T-3GD5 (Radeon HD 7950);
• storage: Intel SSD 320 Series (300 GB, SATA 3Gb/s);
• power supply: Seasonic X-650 (650 W).

The equipment was running OS operating system MS Windows 7 Enterprise 64 bit (90 day trial) which was upgraded to SP1 via Windows Service update. A driver has been installed for the video card AMD Catalyst 13.5 from 04/24/2013, and for the processor and system logic, Intel Management Engine 9.5.0.1345 from 03/27/2013 and Intel INF Update Utility 9.4.0.1017 from 03/18/2013 were used, respectively. The paging file and UAC were disabled, no other optimizations were made.

Each of the processors was tested in two modes: at the nominal frequency and at the maximum overclocking achievable using our air cooler.

	Core i7-4770K	Core i7-4770KOC	Core i7-3770K	Core i7-3770KOC	Core i7-2600K	Core i7-2600KOC
CPU frequency, MHz	3900*	4500	3900*	4700	3800*	4800
Voltage Vcore, V	1,106	1,243	1,048	1,312	1,184	1,46
RAM frequency, MHz	1600	2400	1600	2400	1600	2133
Timings	9-9-9-24-1T	10-12-13-31-1T	9-9-9-24-1T	10-12-13-31-1T	9-9-9-24-1T	10-12-13-31-1T

* - Turbo Boost frequency

The “old” Sandy Bridge showed the greatest margin of safety, the 22nm Core i7-3770K had slightly worse results, while the newcomer's achievements were the most modest.

The set of software used in the tests is as follows:

• AIDA64 2.80.2300 (Cache & Memory benchmark);
• SuperPI XS 1.5;
• wPrime Benchmark 2.06;
• Futuremark PCMark 7(v1.4.0);
• 7-Zip 9.20 x64 (built-in test);
• Adobe Photoshop CS5 (Retouch Artist Benchmark);
• Cinebench 11.5R (64bit);
• TrueCrypt 7.1a (built-in test);
• x264 HD Benchmark v5.0;
• Futuremark 3DMark 11(v1.0.3);
• Batman: Arkham City
• Hitman: Absolution
• F1 2012;
• Metro 2033.

Test results

Synthetic Applications

Compared to its predecessors, Haswell has a drop in data read speed, while the Core i7-4770K is the leader in write and copy operations, in addition, the newcomer has better latency.

In the Super Pi single-thread test, after Overclocking Core The i7-4770K keeps up with its rivals, while in the nominal mode it shows the shortest task completion time.

Testing in the wPrime benchmark shows that after boosting the performance, all three processors cope with the task equally well, while at standard frequencies the newcomer comes to the finish line first. Still, small improvements in the design of Haswell clearly benefited!

In all subtests without exception, Haswell outperformed its predecessor both in normal mode and after overclocking, although the Core i7-3770K clock speed is 200 MHz higher. As for Sandy Bridge, against the background of his descendants, he looks unconvincing.

In the overall standings in the Futuremark 3DMark 11 gaming test package, the rivals showed very close results, since the influence of the processor on the final result is minimal.

However, in the subtests related to the calculation of a realistic physical model, in the nominal mode, there is an approximate parity between Ivy Bridge and Haswell, while the 32-nm Core i7-2600K is again noticeably inferior to its competitors. After overclocking, the Core i7-3770K wins, having an advantage in frequency over the Core i7-4770K. As for Sandy Bridge, it can't keep up with new models, even despite its impressive overclocking potential.

In 3D imaging tasks in Cinebench 11.5R, the hero of today's review demonstrates a significant advantage, which is especially pronounced in real-time animation tests using the OpenGL API, where Haswell is almost 20% faster than the Intel Core i7-3770K.

Unexpectedly, in the popular graphics editor Adobe Photoshop, the newcomer is inferior to its relatives, and at standard frequencies the lag is almost 10%, while after overclocking Haswell keeps at the level of Sandy Bridge. Clearly, the Photoshop code is not too sympathetic to the innovations made in the Haswell microarchitecture.

But, in encryption operations, the Core i7-4770K is easily ahead of its predecessors, presumably due to improvements in the operation of blocks that process vector instructions.

The speed of transcoding HD video increases noticeably from generation to generation by about 5-7%, so the advantage of Haswell is quite expected.

3D gaming performance

Comparing the speed of older processors in video games is not an easy task, however, we tried to find such applications that are most demanding on the speed of the computing part. True, with the inclusion of full-screen anti-aliasing, the productivity of the video card became a "bottleneck", therefore, anti-aliasing had to be disabled.

In the shooter Batman: Arkham City, the hero of today's testing showed the best results, the advantage of Haswell in the default mode is especially noticeable, while in overclocking, rivals show identical results.

At stock frequencies, Hitman: Absolution has a slight lead from Ivy Bridge, while the Core i7-4770K shows slightly lower frame rates. After overclocking, all three processors show the same speed, despite different level clock frequencies.

In the F1 2012 racing simulator, Haswell's performance is on par with the competition, but only in normal mode. After overclocking, Ivy Bridge takes the lead, followed by the Core i7-2600K, and the performance of the newcomer is somewhat inferior to its rivals. Obviously, the lowest clock speed of the Core i7-4770K affects.

In the Metro 2033 shooter, the Ivy Bridge and Haswell processors demonstrate identical frame rates, both in overclocking and at the nominal frequency. The Core i7-2600K is slightly behind the leaders, but the lag is not so noticeable as to speak of a decrease in the comfort of the gameplay.

power usage

To evaluate the power efficiency of processors, we used the Basetech Cost Control 3000 device, which estimated the average power consumption of test benches "from the socket" during system idle time, as well as the peak power consumption during the LinX stress test.

In normal mode, the Core i7-4770K demonstrates the best energy efficiency, outperforming the 22-nm processor of the previous generation by 12% in idle and by almost 5% at maximum load, and the Core i7-2600K turned out to be the most "gluttonous". After overclocking, the situation changes, and the Core i7-3770K shows the lowest power consumption, while Haswell's energy efficiency decreases. Most likely, this is due to the peculiarities of the power subsystem of the LGA1150 platform, or a flaw in the control microcode in an early version of the motherboard firmware.

findings

Before drawing conclusions, let's try to figure out whether the appearance of the Haswell microarchitecture corresponds to the Tik-Tok strategy, because the answer to this question is not at all unambiguous. On the one hand, the computing part of the new processors was inherited from Ivy Bridge practically unchanged, which is indirectly confirmed by the test results, and this does not seem to correspond to the “So” iteration. In any case, the breakthrough that was observed with the advent of Sandy Bridge is not observed, in terms of performance. latest processors Intel Haswell turned out to be only slightly faster than the representatives of the previous generation. But on the other hand, the reorganization of the power subsystem and the transfer of the voltage regulator to a semiconductor chip is a unique solution that radically distinguishes Haswell from all Intel processors of previous generations. The reasons for this development are understandable, in terms of computing power, Intel CPUs are significantly superior to competitor products, which means that the chipmaker can focus on improving energy efficiency and optimizing production costs. This achieves the unification of the model range CPUs, which allows you to use Haswell in the most various devices: from monoblocks and classic "desktops" to tablets and thin laptops.

As for the hero of today's review - the Intel Core i7-4770K processor, against the background of the previous generation models, it demonstrated a steady increase in performance. However, the advantage is often calculated by a few percent, so users are unlikely to notice a significant difference when switching from Ivy Bridge to Haswell. Another thing, if there is a question of choosing a basis for a new PC, there is definitely a reason to think about purchasing the LGA1150 platform, as the most modern and promising one. An important advantage of Lynx Point, in addition to expanding functionality, is the improvement in overclocking capabilities, which makes it possible to overclock even the younger Haswell models. The very potential of the Core i7-4770K was not too impressive, the processor turned out to be extremely hot, which negatively affected the overclocking results, while the performance gain from increasing the frequency pleasantly surprised us. In general, the new Intel processors are definitely a success, although the choice is yours, dear readers!

Testing equipment was provided by the following companies:

• ASRock - ASRock Z77 Extreme6 Motherboard;
• ASUS - ASUS Sabertooth Z87 motherboard and HD7950-DC2T-3GD5 video card;
• G.Skill - G.Skill TridentX F3-2400C10D-8GTX Memory Kit;
• Intel - Intel Core i7-4770K, Core i7-3770 and Core i7-2600K processors, Intel SSD 320 Series 300GB;
• MSI - Motherboard MSI board Z77 MPOWER and Intel processor Core i7-3770K;
• Syntex - Seasonic X-650 power supply;
• Thermalright - Thermalright Silver Arrow cooler.