AMD or Intel. Which processor is better?

Today we will try to answer the question: “Which is better AMD or Intel?” This question is asked by users who want to buy new computer. Recall that the most powerful computer required, as a rule, for games. It is modern games that require large computer resources.

New AMD processors 2012

Let's start with the new AMD processors. In 2012, AMD managed to release some very attractive new products, among which the sensational family of AMD Trinity hybrid processors, primarily aimed at portable computers - laptops and tablets, stands out in a special way. The new family of processors has a rather extensive range of processors, covering almost all the needs of a modern user.

AMD Trinity processors are hybrid chips that combine CPU and GPU, which have become a logical continuation of the previous generation of processors called Liano. Alas, this is just an improved continuation of the old line of processors and the name change, in our opinion, is just a marketing ploy. AMD Trinity chips are manufactured on the same 32nm process using a slightly improved Bulldozer architecture, now known as x86-Piledriver.

Processor comparison Intel Core i5-2410M and AMD A10-4600M

Of the major cardinal updates of the new family of processors, one can single out only the transition to socket FM2 and a new embedded Radeon graphics card HD 7000 with excellent performance.

As already mentioned above, video accelerators of the Radeon HD 7000 family will act as an integrated GPU in the Trinity processors. This integrated video card is manufactured based on the VLIW4 architecture, which is based on the Northern Islands core, operating at a frequency of 424 - 800 MHz. The GPU used supports OpenGL 4.2 and DirectX 11 technologies, and also has from 128 to 384 stream processors.

AMD Trinity processors are available in dual-core or quad-core versions, operating at clock speeds from 2.7 GHz to 3.8 GHz. Chips support RAM DDR3 standard, equipped with a second level cache up to 4 MB and Turbo Core 3.0 technology, which allows you to increase the base frequency of the processor in case of increased computing load. In addition, the new line of AMD Trinity processors received full support for a hardware video decoder and DisplayPort 1.2 and HDMI outputs.
As already mentioned above, video accelerators of the Radeon HD 7000 family will act as an integrated GPU in the Trinity processors. This integrated video card is manufactured based on the VLIW4 architecture, which is based on the Northern Islands core, operating at a frequency of 424 - 800 MHz. The GPU used supports OpenGL 4.2 and DirectX 11 technologies, and also has from 128 to 384 stream processors.

In addition to Trinity processors, AMD introduced several different models of its processors in 2012, among which it is worth highlighting two E-Series processors designed for budget laptops. We are talking about chips E1-1200 and E2-1800. Both processors are dual-core and based on the Brazos 2.0 architecture (Bobcat core). New items support DDR3 RAM standard, are equipped with 1 MB L2 cache and have very high energy efficiency. The clock speed of the E1-1200 is 1.4 GHz, while the E2-1800 chip runs at 1.7 GHz. Both processors are equipped with integrated graphics of the Radeon HD 7300 family.

New Intel 2012 processors

In 2012, Intel also did not waste time and responded to AMD's new products with its own update of the Sandy Bridge processor family, which in the new version became known as Ivy Bridge. And if in the case of AMD processors there is only an increase in the performance of the old platform, then Intel has gone much further, moving the new generation of processors to the next technological level. Ivy Bridge chips are already being manufactured using the 22nm process technology.

Ivy Bridge are also hybrid processors, but at the same time they became the world's first three-dimensional processors based on 3D transistors (Tri-Gate), produced using Fin Field Effect Transistor technology. The use of this technology has made it possible to significantly increase the performance of new processors and, at the same time, reduce their power consumption by almost half. The Ivy Bridge processor family will feature dual and quad-core chips for both desktop and desktop applications. mobile computers. At the same time, the spread of the clock frequency of the new line of processors varies from 1.6 GHz to 3.5 GHz with the possibility of overclocking using Turbo Boost technology.

Like competitors from AMD, the new Intel chips have a built-in graphics video accelerator. Junior Ivy Bridge models are equipped with Intel HD Graphics 2500, and the older ones got a GPU at their disposal Intel HD Graphics 4000. Both options have full support for OpenGL 3.1, OpenCL 1.1 and DirectX 11 technologies, and are also equipped with an Intel Quick Sync 2.0 module for faster video stream encoding. The frequency of operation of graphics chips varies from 350 to 650 MHz, but in a special Turbo mode it can increase to values ​​of 1050 - 1300 MHz.

It's time to make a comparative analysis of the main novelties from both manufacturers and find out which of them has worked most successfully this year. So, both companies presented a wide range of models covering all types of modern computer technology, starting with server stations and ending tablet computers. No one has an advantage in this component, which, in fact, is not surprising.

Now let's take a look at the microarchitecture of the new processor lines. This is where a slight advantage of chips from Intel begins to be traced. If AMD Trinity retained the previous 32nm process technology, then Intel Ivy Bridge chips stepped up a step, having mastered the 22nm process technology and switching to 3D transistors. The transition to a new process technology allowed Intel to release much more efficient processors in terms of energy saving. So for desktop computers, AMD offers processors with a minimum power consumption of 65 W, while Intel in the same segment introduced several chips with a consumption of only 45 W and even 35 W.
The functionality of the Intel Ivy Bridge architecture is also much higher than that of the AMD Trinity. If AMD processors only support video output via HDMI or DisplayPort, then competitors from Intel also know how to work with Thunderbolt outputs. The work of Ivy Bridge processors and memory is much more efficient. For the first time, Intel's new products use a ring bus (Ring Interconnect), which allows computing units to exchange data directly through a common L3 cache and works much faster than the AMD processor bus.

In conclusion, let's look at the graphics component of the new families of hybrid processors from AMD and Intel. Almost equality is observed here with a slight advantage in the direction of AMD. The built-in GPUs of Trinity processors have a higher base frequency and support for Dual Graphics technology, which allows you to connect the capabilities of a discrete video card to the integrated video accelerator, if it is, of course, available in the system. In turn, the integrated graphics of Intel Ivy Bridge processors boast the ability to work in Turbo mode and support for Intel Quick Sync technology, which provides faster video stream encoding in Full HD quality.

AMD or Intel?

As can be seen from our comparative analysis, today in 2012 the clear leader is Intel with its line of Ivy Bridge processors. In addition to the already announced advantages of Intel processors, it is also worth noting a number of other significant advantages that consolidate the leadership of Ivy Bridge chips. Firstly, the Ivy Bridge chips have retained socket compatibility with the previous Sandy Bridge line, so it will be much cheaper for users to switch to new processors. Secondly, Ivy Bridge processors have a larger cache memory. And finally, thirdly, the new line of chips from Intel has a wider model range, which allows you to choose a processor for the needs of any user, depending on his financial capabilities.

Back in 2002, I wrote about the expected problems of processor manufacturers when switching to thinner ones. processes. Some of them were solved completely unnoticed by the general computer community, some (for example, problems with gate leakage current) were widely covered in the press.
Movement to thin tech. processes is not only a way to search for new technical ideas, but also a way of financial costs and technological compromises, which in turn limit the success of processor manufacturers

History of processor development

This is a constant desire to increase its performance, and to do this, make the main cell (key) - a complimentary pair of transistors as fast as possible or as small as possible. This was achieved by the use of increasingly fine technological processes.
Moreover, the performance of the cell is the higher the thinner those. process.

The development of processors can be divided into two stages.

The first one is around 2005.

At the first stage, the main goal of the designers was to make the size of the CMOS key as small as possible in order to obtain ever higher clock frequencies of the processor and, accordingly, increase its performance. And only then, due to the small size, increase the number of keys in order to obtain a more complex structure, the optimization of which also gives some performance gain.

And only then there was a decrease in the power consumed by the key and, accordingly, by the processor, and other advantages of thin technologies.

Moreover, the main increase in processor performance was provided precisely by the increase in clock frequencies.

The second one began in 2005, from the moment the processor clock speed stopped growing.

At the second stage, the race to reduce the size of the CMOS key continues. Its goal was to place as many keys as possible on a chip, in order to be able to complicate the structure of the processor (including an increase in the number of cores and the size of the cache "she"), which allows to increase performance. The second reason for the movement is to reduce the power consumed by the key and, accordingly, the processor. Growth of clock frequency stopped.

It was in this way that the performance of server processors was increased earlier, when even thinner tech. the process did not provide an opportunity to increase productivity.

From that moment on, manufacturers switched to the so-called rating - the equivalent performance of the processor.

On the way to fine technological processes, many problems arose and were solved. Some of them were solved completely unnoticed by the general computer community, some (for example, problems with gate leakage current) were widely covered in the press. This path is not only the path of searching for new technical ideas, but also financial costs and, most importantly, the path of compromises that impose certain restrictions on the development of technology.

At the beginning of May there was information

about the Intel solution, literally - “launch a program to phase out their processors Core i7 940”, both retail models and OEM products.

The Intel-style formula, "run-on-withdraw" is a nasty action, looks pretty positive! Not at all like "out of production".

Note that since the release of the first chips of the Core family i7 has already passed almost half a year, and this is a rather short period for the processor industry ... and here is the solution!

The Core i7 940 was followed by the Core i7 965!

What does this mean?

Some people think that against the backdrop of the crisis, the principle of Intel's work does not work - "everyone will take what we offer, with an appropriate advertising company."

There is an opinion that this is an attempt to sell stocks of 4-series system logic sets, the demand for which has fallen due to the global economic crisis. But "realize by giving up" is a no-win formula. All the same, losses here or there.

Another opinion is that the cost of manufacturing Intel Core i7 940 turned out to be high and it does not have the demand that allows it to have cost-effective production.

Another opinion, against the background of the crisis at Intel exacerbated internal problems.

So far, one can only guess why the life of the Core i7 940 and Core The i7 965 turned out to be so short, but usually the reasons for the cessation of production should be quite weighty, because the money was spent, and the crisis is in the yard. Moreover, new Core i7 975 and 950 models are planned for release - not very different in performance.

But most likely this is the whole set of the above, which was superimposed on the problems of mastering more subtle technological processes.

Processor trends

Each step in the development of fine technological processes means a decrease in the linear dimensions of the transistor by about 1.4 times and its area by about 2 times.

The last two points significantly affect the economic feasibility of releasing new models for sale, and their price.

The most characteristic points in the history of the development of new technological processes are shown in Table 1.

Table 1.

*Intel considers the year of development of those. process, the year of presentation to consumers of a chip sample made according to this tech. process. Previously, it took several weeks from the introduction of a processor chip to its release for sale. Starting from 45 nm tech. The process, after the presentation of the memory chip (the technology is now being developed on them), takes up to six months before the presentation of the first processor, and the deployment of mass production (many models) takes up to six months. Therefore, here is the date of serial production of the first model of processors for this tech. process. Therefore, this line may have values ​​different from those accepted by Intel.

** in Intel plans for 2009.

According to Intel Corporation officials, in 2012, microchip manufacturers will switch to a 10-nm process technology. Vice President of Intel Digital Enterprise Group (Intel division responsible for the design and manufacture of discrete chips) Pat Gelsinger believes Intel factories will be able to make transistors as small as 10 nanometers or less.

Such a statement devoid of logic can only be considered as another publicity stunt, since such a range of technology changes previously required from 4 to 7 years. Since each step is associated with the introduction of new technologies, equipment and their debugging.

But in the history of not only Intel, but in general, there were no such abrupt transitions to new tech. processes. Therefore, realistically, in my opinion, we can expect the step taken by Intel. processes, which will give a series of 32, 22, 16, 11 nm.

Not even considering the unknown that awaits the developer.

See Table 1.

Table 2.

1. With the transition to more subtle technologies,

the length of the channel of transistors that make up the discrete structures of the processor is shortened, and this, in turn, causes an increase in their speed.

Those skilled in the art are aware of dependencies that relate the length of a MOS transistor channel (process size) and its speed. Look at the graph describing this dependence in Fig.3.

Picture 1.

The concept of speed, up to 90 nm tech. process, was clearly associated with clock frequency processor. The speed of the transistor grows - the clock frequency of the processor core also grows.

Now, the speed does not mean - the clock frequency of the core.

Restrictions.

In existing technologies for manufacturing system (motherboard) boards, switching times equal to the switching times of core transistors cannot be used for external buses.

Because, with the growth of performance, the requirements for the accuracy of the time of arrival (synchronism) of signals on parallel buses for information transmission and synchronization increase.

This is not a critical limitation, it can be bypassed by applying the transfer of information over serial channels.

2. The areas occupied by the transistor in the chip are reduced,

approximately two times for each step of reducing technological standards, as a result, its internal capacities should decrease .

But the use of a high-k dielectric for gate insulation of transistors made at 45 nm tech. process, reduces the gate capacitance slightly. This reduces the specific (by 1 key) power consumption less than before when switching from one tech. process to another.

Not taking into account this factor (or maybe just to test the technology on users - a structure designed for 22 nm process at 32 nm technical process), Intel released Intel Core i7 940 chips for sale with TDP equal to 130 W. And so, at the beginning of May, there was information about their removal from production (although their appearance on the 32 nm technical process is possible).

In fact, heat dissipation powers of more than 100 W require a special approach to the problem of cooling the processor and system block. The slightest inaccuracy in this matter leads to the appearance of temperature gradients on the chip, which does not contribute to its durability.

According to my data, the use of a high-k dielectric as an insulator led to the preservation of capacitance (100-70%) of the transistor when going from 65 to 45 nm. those. process.

As a result of an increase in the number of transistors, the power consumed by the processor and a slight decrease in the gate capacitance of the transistor increases. An example of this is the Intel Core i7 940.

3. Despite the increase in performance

the clock frequencies of the processor core stopped growing before reaching 4 GHz.

Rice. 1 (my data).	Rice. 2 (data http://ru.wikibooks.org/wiki/ search word Processor)
On fig. 1 and 2 are graphs of the processor clock frequency. They are not synchronized along the horizontal axis, since Fig. 2 used from another source. And fig. 1 shows only the characteristic region. But its task is to show the change in the clock frequency over time or with a decrease in those standards. processes they perform quite clearly.

I'm not talking here about:

• ability to overclock the processor, since the overclocked mode is your experiment - your risk, in which stable operation of the processor is not guaranteed.
• equivalent processor performance, which is determined not only by the core clock speed, but by a complex of processor characteristics.

Here we are talking only about the core clock frequency determined by the manufacturer.

Of course, one can dispute the drop in the core clock frequency by 45 nm tech. process, but already, no one disputes the lack of its growth. And a comparison of the increase in clock frequency during the transition from 250 nm to 180 nm tech. process, clearly not in favor of similar situations after 90 nm.

And the statements of some "craftsmen" about a high clock frequency are highly controversial. Since overclocking mine sample (as I said, not everyonesampleCPU can be overclocked Intel processor up to just over 4 GHz, they were never able to translate their "know-how" into the category of a standard solution for a wide range of at least "craftsmen", and, as far as I understand, they themselves do not use record modes constantly.

Otherwise, by analogy with the headings "Processor XXXXXXX - exceeded 4.2 GHz!" the headlines "Processor XXXXXXX - 3 years at 4.2 GHz!"

It is believed that there is another reason for the limitation of the processor clock frequency - this is the limitation of the bus bandwidth for communication between PC devices.

4. Reducing the area occupied by the transistor

allows a larger number of transistors to be placed on a substrate of the same size, complicating the structure of the processor. This to some extent has a positive effect on the overall speed of calculations.

This is what processor developers use. The number of transistors on a chip is constantly growing.

Figure 2.

The increase in the number of transistors is due to the complexity of the processor structure and the placement of a larger number of cores on the processor chip, larger caches (which, by the way, tend to increase), memory controllers, .....

It should be noted that the heaviest for the chip, in terms of heat dissipation, are the cores that operate at high clock frequencies.

In order not to heat the chip, there is an idea, in multi-core processors, to use additional cores sharpened to perform some kind of narrow (specialized) tasks. This will allow you to turn them off when there are no tasks and thereby reduce the power consumed by the processor and heat dissipation.

On the other side - Increasing the total number of transistors in a chip - as an aspiration Intel fit into the Procrustean bed of Moore's Law.

This requires an increase in the number of nodes on a chip and, as a result, an increase in the number of transistors. But not doubling every year or two.

If the main thing is not the optimal operation of the processor, but Moore with his law, there is an easy way to comply with this law, just increase the cache. After all, it is known that each bit of the cache requires 6 transistors to store a bit of information, and together with controllers - interfaces, strapping (practice shows) there are already more than 50 transistors per 1 bit of a level 3 cache. This is a significant contribution to the triumph of Moore's Law.

Although there is a refutation of Moore's Law, this is a processor:

Intel Atom Z515 - 1.20 GHz (512 KB L2, 400 MHz FSB, 1.4 W TDP) - introduced April 8, 2009, Silverthorne- 45 nm process technology and having a crystal of 47 mil. transistors. He positioned as a microprocessor for ultra-mobile systems / Netbook and Nettop class systems.

There is a drop in the number of transistors!

Another significant contribution to the increase in the number of transistors is the use of architectures with multiple cores.

But the number of nodes - cores and the size of the cache cannot be infinite, starting from a certain level, managing them will take so many resources that the increase in processor performance will stop.

Therefore, talk about the use of 100 and 1000 core processors in a PC is still premature.

The result of this is an increase in the number of external connections (lines) of the processor and an increase in the number of contacts of its connector - Socket "a.

5. The number of processor cores is growing

the number of which in forecasts is approaching a hundred. The increase in their number is caused by the desire to increase system performance.

It is clear that such an increase cannot continue indefinitely. After all, synchronization and control of parallel computing also requires computing resources. An end to multiplying the number of cores where their further increase does not give a performance increase.

But, we must not forget that an increase in the number of cores, as well as the size of caches, also requires resources.

Information flashed that Intel plans to sharpen individual cores of a multi-core processor for separate specialized tasks, which will increase their performance and turn them off if there are no tasks for them. The latter will reduce power consumption. For example, one of the cores can be sharpened to perform graphics operations.

But it seems that the extreme situation of such development is the chip where all the main processor nodes are located, leaving outside it only those nodes that do not have a significant impact on the speed of the PC.

It is clear that the development, improvement of processors is aimed at increasing the speed of its work and the speed of the system. To do this, its architecture is optimized, including the number of cores, the size of caches of all levels, memory controllers are transferred to the processor.

This requires an increase in the number of nodes on a chip and, as a result, an increase in the number of transistors.

6. Approximation of TDP to the limit value,

which in modern processor designs in conditions of optimized cases has a value of the order of 130 - 150 watts.

This limitation is imposed not by the presence of effective coolers, but by the design features of the processor itself, the size of the crystal, and the inhomogeneity of heat release on its surface.

You probably noticed that lately processors with a TDP of about 130 W have sometimes appeared. Most often, these are processors designed for release on a thinner tech. process. For example processor Intel Corei7 940 architecture Nehalem made on 45 nm tech. process has a TDP of about 130 W, being made at 32 nm. TP it will have a TDP of 65 to 95 watts, depending on the clock speed.

Usually TDP does not exceed 100W.

130 W is the maximum power that can be output from a semiconductor device with such dimensions of heat-conducting surfaces and a similar design.

But it imposes high requirements on the level of semiconductor device cooling technology.

These are the connections of a crystal with a heat distribution plate on a processor with a thermal resistance of about 0.01 ° C / W, effective heat-conducting materials (pastes), coolers with a thermal resistance of less than 0.1 ° C / W, and cases with effective ventilation.

When TDP approaches 150 W, it threatens with local overheating on the chip, reducing its noise immunity, sensitivity to external cooling devices and, accordingly, overall reliability.

Restrictions imposed -

TDP limits the number of transistors on a chip and the processor clock speed.

7. Increase in the number of processor socket pins

(connector) - Socket" a.

3 factors influencing the increase in the number of contacts:

1. The complexity of the processor structure,
2. The increase in current consumption,
3. Increasing the interference frequency.

1. The complication of the processor structure and the increase in its external connections creates a need to increase the number of contacts on the Socket "e of the processor. But the increase is not only in the number of external connections. The transmission goes through pairs of conductors, so the number of contacts increases by twice the number of external connections of the processor.

This is logical and understandable.

Figure 4

2. As we know, the number of contacts on the Socket "e, to supply power to the processor, exceeds 150 pairs. This is required by high currents supplied to power the processor. Moreover, a decrease in the supply voltage leads to an increase in the current supplied to the processor. This occurs even while maintaining the consumed power processor, because the value of the supply voltage is reduced (so far, up to 1 V).

And with a current limit of 0.5 A (0.5 A - limit) for one pair of contacts, you can estimate how many contacts are required for this. (The current safety factor requires approximately 0.3A per pin) But the number of contacts insocket" e allocated for these purposes is always more. The increase in the number of contacts determined by the maximum current, especially with a decrease in the supply voltage, is not a trend, but a technical necessity. (With a supply voltage of 1.1 V and a power consumption of 130 W, more than 230 pins are required for this.)

3. Parallel connection of power lines requires not only power supply, but also the removal of broadband interference generated by the processor during operation outside the chip and Socket. many contacts.

This is especially important at 0.45 nm or less TC, because the upper frequency limit of interference exceeds 50 GHz.

But with the transition to thinner tech. processes, the RF boundary of the generated interference increases and a decrease in the inductance of the power supply lines to the processor is required and, as a result, an increase in the number of Socket contacts.

Therefore - at the complexity of the processor structure, the increase in current consumption, and the need to remove the noise generated by it from the processor - all this requires an increase in the number of contacts by socket"e.

Increasing the number of contacts increases the size socket" and, accordingly, the inductance of the connections on it. At a certain size socket"

But this process is not unlimited.

Increasing the number of contacts increases the size socket" and, accordingly, the inductance of the connections on it. At a certain size socket" and an increase in the number of contacts does not give the necessary reduction in their inductance.

8. New solutions

Appear periodically in print. They mainly refer to new, faster transistors.

For instance:

• The so-called vertical structure transistors,
• Double gate transistors.
• New semiconductor materials, ...... The list is constantly updated.

Of course, transistors of new structures with operating frequencies (cutoff frequency) of 20.50 GHz are an interesting thing, and not only from the point of view of application in digital (discrete) technology.

But don't forget:

The nature of the operation of transistors in the switching mode is the same, and all structures with high switching speeds always have concomitant negative phenomena that limit their capabilities.

Yes, and CMOS structures made at 45 nm. - have a switching time of the order of 10 ps and an operating frequency (cutoff frequency of the slope - characterizing its amplifying properties in the linear mode) of about 16 GHz. This means that the transistors modern processors made at 45 nm. those. process is theoretically capable of running at a processor speed of 16 GHz. But, those very negative phenomena do not allow this.

After some refinement of the design and structure of the processor, the operation of MOS structures is possible at frequencies approaching the cutoff frequency of the slope. So the processor is made at 45 nm. those. The process is capable of running at core clock speeds of 7-10 GHz.

The increase in the number of processor cores continues 2, 4, 8 and in the future 60, 80, 100. Although the latter is doubtful for widespread use.

I would like to say a few words about new semiconductor materials that have a significant impact on the performance and operating temperature of the processor.

Now appeared new semiconductor materials, transistors made on which work for more high frequency, at higher temperatures.

Table 3

According to [L.1]

Field-effect transistors based GaN already on sale.

Intel conducts research on the possibility of using semiconductors III groups (which includes GaN).

There are several design options for high-performance processors.

Now there are new technologies on the horizon

What will they be?

It is difficult to guess, but obviously not purely optical, which is still at an early stage.

But optical technologies are already being explored by Intel and others.

So far, not optical processors, but only high-speed optical I/O interfaces for chip-to-chip interconnects.

According to Intel -
“Currently used 15-20 Gb/s copper-based interconnect technologies are marginal due to the inevitable signal degradation, power dissipation, and increased negative impact of electromagnetic interference at ultra-high clock frequencies.”

Figure 5

Intel is already working on optical data transmission systems [L.4].

And not only creates technologies that allow embedding optical data transmission systems into processor chips, but also already has prototypes of such transceivers.

Such transceivers ( electronic devices interfacing, used, in particular, to connect computers to a network) on CMOS transistors will be able to operate at clock frequencies of the order of 14 GHz, which is quite enough to provide a data transfer rate of 20 Gbit / s. [L.2]

The latest models are capable of 40 Gb/s data exchange, and an 8-channel transceiver with a throughput of up to 1 Gb/s is expected to appear in the near future.

And computer models with similar transceivers (optical communication channels) used instead of external processor buses are already being tested in Intel laboratories.

“The Moscow inventor, Alexander Verbovetsky, was able to change the microcircuit of this board in such a way that it was possible to increase performance, noise immunity, reliability and survivability personal computers using optoelectronic motherboards.

This result was achieved through the use of optical methods of input / output and signaling, which can dramatically increase the data transfer rate, as well as through the use of a group bus architecture.

Additional processors, processor interface units, optical communication nodes of each circuit block with each other (processor with system bus, cache memory with system bus, system control unit with system bus, etc.) control unit.

Such a combination of blocks and connections between them made it possible to obtain a device with more than 100 times greater performance, noise immunity and reliability than conventional modern motherboards for personal computers." (end quote)

These two solutions are the practical creation of a single high-speed optical bus on which all its nodes can be planted, providing internal and external PC communications.

There are solutions

Which allow you to increase the clock frequency without increasing TDP, other solutions allow you to increase the TDP of the processor at least twice, which allows you to increase the processor clock frequency at least 2 times without changing anything in modern design approaches. By changing the organization of the internal structure of the processors and applying some design solutions 2 more times.

In total, this makes it possible to have a processor clock frequency of more than 10 GHz. This is where synchronization issues come into play.

Conclusion

Of course, these are not all trends and problems in the development of processors.

There is no end to the depth of the issue, dozens of deeply scientific works can be written on it, but still time will pass and new problems will arise that need to be addressed.

I wanted to tell here that the history of the development of processors is a constant compromise, the result of which is often not at all what the leaders of the industry are planning. And the number of compromises and, accordingly, the restrictions becomes greater when approaching the physical limits of the main element of the CMOS transistor processor. And then it remains only to fulfill Moore's "Law" at the expense of huge caches.

An example of such a compromise is limiting the clock frequency of the processor.

The accumulation of these compromises eventually becomes insurmountable, and this is the dead end of this technology.

According to information Fujitsu , made on 45 microns of those. process, eight nuclear processor SPARC64 VIIIfx ( Venus ) has a computation speed of 128 GFLOPs, - 2.5 times faster than the best Intel, dual-core Itanium 2, however, even with built into Venus memory manager consumes only 33% of Itanium 2 , therefore about 35W.

One of the "specialists" calculated the clock speed of this processor as 16 GHz.

This is not true already because, with a similar structure of transistors, modern tech. processes and a TDP of 35 W, its clock frequency cannot exceed 4 GHz.

But soon there will be processors new generation, where instead of buses for communication with external devices, optical data transmission systems built into the processor will be used. These are buses for information exchange with memory, external devices (PCI-E, ...), and even communication buses with HDD, SSD, ....

And the processor, like the computer, will appear in a new form and possibly quality.

P.S.

The article was written in 2009, and now it’s mid-2013 (a year from the forecast), and now, after some silence, there were reports from observers like “10 nm process technology is a reality of 2015” and practitioners attribute the development of the processor to 2018. In the meantime, since the beginning of 2012, only processors based on the 22 nm process technology have been mass-produced.

With a further decrease in technological standards (already from 45 nm process technology), the technological difficulties of their development progressively increase.

An example of this is the 22 nm process technology, which was demonstrated in 2008, and mastered in the production of processors only 4 years later (2012).

Therefore, even if (not at all a 10 nm process technology) a 14 -18 nm process is demonstrated in 2015-2018, it can be mastered no earlier than 2020-2025.

I'll be happy to be wrong.

August 2013

Literature.

1. “The GaN transistor is the toughest nut yet” V. Danilin, T. Zhukova, Yu. 4/3

2. Intel Introduces Prototype High-Speed ​​Optical I/O Interface for Chip-to-Chip Interconnects by Ian Young, http://www.intel.com/corporate/europe/emea/rus/country/update/ contents/it04041.htm

3. A Russian specialist has developed a new-generation optoelectronic mainboard that surpasses even modern analogues from IBM. http://www.sciteclibrary.ru/rus/catalog/pages/5833.html

4. Optical future news from [email protected] Daw, Chip, September 2009

A. Sorokin

New laptops based on the Intel Ivy Bridge platform have already appeared on sale, and many are wondering what this mobile platform is. By the way, it is available both for mobile computers and for regular desktop computers. Traditionally, it is believed that the desktop version is more productive, but now Intel promises us that mobile processors will be very powerful.

In this article we will talk about the new processors from Intel in 2012, but in the next one we will touch on the integrated graphics of the Intel HD 4000. We will deliberately not go into technical details. Still, our site is more aimed not at professionals, but at those who simply want to better understand the design of laptops. Well, if this is not enough for you, there are a lot on the network. detailed reviews Intel Ivy Bridge.

So, Intel Ivy Bridge is a logical continuation of another popular one that made a lot of noise at one time. Only if the latter was made according to the 32nm process technology, then Ivy Bridge is a step forward, because these processors are made according to the 22nm process technology. This made it possible to reduce their size, while increasing performance and reducing power consumption.

Recall that Intel releases its processors using its own tick-tock strategy. “Tick” is the miniaturization of the process technology (for example, from 32 to 22 nm), “so” is already a completely new architecture. Thus, Sandy Bridge processors are "so", and the new Ivy Bridges are "tic". In 2012, a new “so” is expected - Haswell processors.

Ivy Bridge processors belong to the Intel Core family already known to us, this is their third generation. The manufacturer promises us higher performance and improved support compared to older versions. wireless networks. Well, plus the built-in USB 3.0, which will make it possible to exchange data with external devices much faster. Reduced power consumption was added to the increased performance - from 17 to 55 watts, depending on the processor.

Processors in the new line - both quad-core and dual-core. The most productive of them is the 3920XM, which runs at a stock clock speed of 2.9 GHz, but at the same time can be overclocked using Turbo Boost technology up to 3.8 GHz. New processors will transcode video much faster - almost four times faster. In addition, they heat up less.

The processor also has a built-in random number generator plus protection operating system from hacker attacks aimed at "elevation of privileges". The memory controller has also been improved - it now supports faster and less power-hungry DDR3 RAM.

Built-in support for the Thunderbolt interface, which is now used by Apple in their laptops. It provides very fast data transfer with external devices. Well, in general, up to three displays can be connected to a laptop on this platform at once, and if you want to play games in this way or simply expand your desktop, you are welcome.

Here is a list of all 2012-2013 Intel Ivy Bridge mobile processors:(as of 05/06/2013)

Meaning of indexes:

M- Mobile processors
XM- extreme 4-core processors with unlocked multiplier
QM- 4-core processors
U- processors with reduced TDP (power consumption)
Y- ultra-low TDP processors

And there are more, most of which are designed for installation in ultrabooks and tablets:

(image enlarges by clicking)

Meaning of indexes:

E- embedded processors
QE- 4-core embedded processors
ME- embedded mobile
LE- performance optimized
UE- energy optimized

Intel's main competitor is AMD at the last Financial Analyst Day exhibition announced its plans to launch its desktop products on the market. As a result, it can be argued that many interesting new products await us this and next year.

The release of the Radeon HD 7900 was just the beginning of a whole family of Southern Islands graphics accelerators.: this year we will see desktop accelerators of medium and productive levels, and, in addition, powerful mobile video cards will appear. In 2013, AMD is planning to update its graphics cards with the release of the new Sea Islands line. These chips will be manufactured on the same 28nm process as Southern Islands, but will offer a new architecture with increased performance in the field of heterogeneous computing.

Talking about the development of the traditional model range FX processors from AMD, we can say that we are waiting for the appearance on the market of a more advanced Bulldozer architecture, codenamed Piledriver. The company promises to release Vishera chips with this architecture this year. For 2013, the company does not seem to be planning to update the FX line, although work on the Steamroller architecture is proceeding at an active pace.

This year, the strongest update is expected in the field of mid-range APUs. In this area, AMD is preparing Trinity chips that will replace Llano processors. They will be dual- and quad-core with Piledriver architecture and next-generation graphics, making them a great choice for low-cost desktops and laptops. And in the sector mobile devices Trinity APUs will appear with a power consumption of only 17 W - this indicates that AMD does not intend to give away the niche of ultrabooks to competitors represented by Intel. The market will also be present and mobile APU Trinity with "standard" power consumption, which does not exceed 35 watts.

Hybrid mid-range processors in 2013 will be improved even more seriously. The 32nm Trinity chips will be replaced by 28nm Kaveri APUs. These systems will be based on a chip with new CPU cores of the Steamroller architecture (two or four cores, depending on the model) and graphics of the GCN architecture with support for more advanced HSA (Heterogeneous Systems Architecture) heterogeneous computing.

The entry-level APU segment in 2012 will not be replenished with any special novelties. The Brazos platform (Series C and E) will be updated with the 40nm "Brazos 2.0" platform, which will bring support for USB 3.0 and Turbo Core. In this regard, 2013 is much more interesting: we are promised a 28-nm Kabini platform. These systems will have a chip with new Jaguar cores (an evolution of Bobcat) and a new graphics component of the GCN and HSA architecture, which will replace the VLIW5. The SoC approach and thinner manufacturing process should be expected to significantly increase performance while reducing power consumption.

Among other things, AMD has officially announced the launch in 2012 of new energy-efficient APUs for tablet PCs, codenamed Hondo. These APUs will be manufactured using the same 40nm process, but will offer some optimizations relative to the current Z-01 (Desna) chip.

How many processor cores do you really need? Two? Four? Six? In many ways, the answer depends on what you use your PC for. It has been found that most games will run well on machines with at least three cores. It is also known that many resource-intensive applications, such as video editing, will use as much "horsepower" as you give them. And there are still applications that do not use multithreading at all.

In fact, the key to optimality is . When you see how the global giant of the microprocessor and graphics market - AMD - is preparing a new line of solutions for desktop computers with a completely new architecture, you immediately imagine incredible powerful system at the same incredibly high price.

But when marketing slides in detail present the company's flagship desktop processor as a solution for relatively cheap PCs, we can only expect a processor designed specifically for this market segment. Of course, it's hard to believe that this is anything more than an advertising spoiler, but enthusiasts who were hoping to see AMD's Bulldozer with an architecture that will destroy Sandy Bridge and fight Sandy Bridge-E will have to adjust their expectations. Instead, the company is clearly not going to spend as much money on hardware as they have done in the past. Anyway, for now.

It's weird, isn't it? On the other hand, Intel's experience with Sandy Bridge showed that society experienced users You don't need $1,000 processors to get incredible performance. AMD's unlocked $200 chips are capable of 4.5 GHz and also perform better in a number of benchmarks (including gaming tests). If AMD can offer even better value for money in this market, then, as they say, go ahead and do not hesitate.

At least on paper, AMD's processor lineup looks comprehensive and competitive. The AMD processor line has seven models in the FX family, ranging from the FX-8150 to the FX-4100. They are all based on AMD's Zambezi design, manufactured by Globalfoundries on a 32nm process, and consist of approximately 1200,000,000 transistors (AMD recently revised its transistor estimates upwards). 315mm² heads are smaller than Thuban processors (346mm²) but larger than Deneb processors (258mm²). For comparison, Sandy Bridge processors are 216 mm².

AMD Processors FX have eight, six and four cores depending on the model (respectively four, three, and two Bulldozer modules). The numbers at the beginning of the designation of the model range will help you determine the number of cores: FX-8xxx - eight, FX-4xxx - four. The three numbers that follow the main ones indicate the level of performance. They are not consistent with clock speed, TDP, or L2 cache. You just have to remember that in the FX-8xxx segment, the 8150 is better than the 8120, which is better than the 8100.

All processors in the FX line come with an unlocked multiplier, which can make them a tidbit for overclockers, depending on how aggressively AMD is tuned to the speed potential of their processors. Sometime in 2008, one enthusiast jumped over the well-known 4 GHz with the Core i7-920, because the chip itself allowed it. It remains to be seen whether it is possible to achieve the same success with a 32nm Globalfoundries chip.

One thing is clear - for a reasonable price (compared to the offers of a competitor represented by Intel), AMD offers not a computing monster, but a quite serious solution for a home computer capable of solving absolutely any task in terms of complexity.

Recently, AMD's position in terms of desktop processors in the market has been a little shaken. First, due to the lack of progressive architectures, the company had to reduce the prices of its products. After some time, AMD completely left the position of high-performance CPUs of the top price policy. And all this was energized by another failure associated with the release of processors based on the Bulldozer architecture, on which high hopes were pinned. Bulldozer was supposed to compete with older Intel processors (LGA 2011 and LGA 1155). However, the new architecture greatly disappointed with its slowness and high power consumption. As a result, Bulldozer can only compete with mid-range Intel processors due to twice the number of cores.

But, fortunately, a series of failures did not deprive the company's engineers of enthusiasm, and a year after the release of Bulldozer, a new improved version of this microarchitecture called Piledriver was launched. The results of testing the AMD FX-8350 processors showed that the time spent on improving the line was not wasted. Senior member of the Vishera line of desktop processors delivers significant performance improvements AMD platforms. From testing, it became clear that the performance increased by about 15%, so to consolidate its successful development and enhance the effect, AMD set a very affordable price. All these events made people talk about Vishera processors in a positive way.

The flagship FX-8350 is even faster than previous generation AMD processors. Given the democratic pricing policy of the company, FX-8350 can be advised to install on an inexpensive desktop system that involves solving resource-intensive tasks in the form of creating and processing high-resolution content or final rendering. However, before making a final decision, it is worth considering its shortcomings. In the foreground - high level energy consumption. The next thing to note is the imperfect load distribution across the eight cores.

It is also worth paying attention to the FX-8320. This model is practically in no way inferior to the FX-8350, but it is an order of magnitude lower in terms of price. In professional applications, the speed of the FX-8320 is at its best. And given the fact that modern AMD FX processors are endowed with unfixed multipliers, it will not be difficult for the FX-8320 to be overclocked to the flagship level and even higher.

The six-core modification of Vishera, at first glance, does not stand out from all the presented models. Due to the deactivation of one of the 2-core modules in the FX-6300, its peak performance is quite low compared to the 4-core Intel ones. However, AMD far-sightedly used pricing tactics for the FX-6300, which in this respect competes not with the Core i5, but with the Core i3. This approach opens up great prospects for the six-core Vishera, especially since, unlike the developments in the Core i3 line, the FX-6300 can be overclocked.

In the AMD FX-4300 processors, the company's engineers disabled half of the cores and cut down half of the L3 cache, and therefore this category does not differ in high performance, but is endowed with high efficiency rates.

Intel's main competitor is AMD at the last Financial Analyst Day exhibition announced its plans to launch its desktop products on the market. As a result, it can be argued that many interesting new products await us this and next year.

The release of the Radeon HD 7900 was just the beginning of a whole family of Southern Islands graphics accelerators.: this year we will see desktop accelerators of medium and productive levels, and, in addition, powerful mobile video cards will appear. In 2013, AMD is planning to update its graphics cards with the release of the new Sea Islands line. These chips will be manufactured on the same 28nm process as Southern Islands, but will offer a new architecture with increased performance in the field of heterogeneous computing.

Speaking about the development of the traditional AMD FX processor lineup, we can say that we are expecting a more advanced Bulldozer architecture, codenamed Piledriver, to appear on the market. The company promises to release Vishera chips with this architecture this year. For 2013, the company does not seem to be planning to update the FX line, although work on the Steamroller architecture is proceeding at an active pace.

This year, the strongest update is expected in the field of mid-range APUs. In this area, AMD is preparing Trinity chips that will replace Llano processors. They will be dual- and quad-core with Piledriver architecture and next-generation graphics, making them a great choice for low-cost desktops and laptops. And in the mobile device sector, Trinity APUs with a power consumption of only 17 W will appear - this indicates that AMD does not intend to give away the niche of ultrabooks to competitors represented by Intel. The market will also be present and mobile APU Trinity with "standard" power consumption, which does not exceed 35 watts.

Hybrid mid-range processors in 2013 will be improved even more seriously. The 32nm Trinity chips will be replaced by 28nm Kaveri APUs. These systems will be based on a chip with new CPU cores of the Steamroller architecture (two or four cores, depending on the model) and graphics of the GCN architecture with support for more advanced HSA (Heterogeneous Systems Architecture) heterogeneous computing.

The entry-level APU segment in 2012 will not be replenished with any special novelties. The Brazos platform (Series C and E) will be updated with the 40nm "Brazos 2.0" platform, which will bring support for USB 3.0 and Turbo Core. In this regard, 2013 is much more interesting: we are promised a 28-nm Kabini platform. These systems will have a chip with new Jaguar cores (an evolution of Bobcat) and a new graphics component of the GCN and HSA architecture, which will replace the VLIW5. The SoC approach and thinner manufacturing process should be expected to significantly increase performance while reducing power consumption.

Among other things, AMD has officially announced the launch in 2012 of new energy-efficient APUs for tablet PCs, codenamed Hondo. These APUs will be manufactured using the same 40nm process, but will offer some optimizations relative to the current Z-01 (Desna) chip.

How many processor cores do you really need? Two? Four? Six? In many ways, the answer depends on what you use your PC for. It has been found that most games will run well on machines with at least three cores. It is also known that many resource-intensive applications, such as video editing, will use as much "horsepower" as you give them. And there are still applications that do not use multithreading at all.

In fact, the key to optimality is . When you see the world's microprocessor and graphics giant, AMD, preparing a new line of desktop solutions with a completely new architecture, you immediately imagine an incredibly powerful system at the same incredibly high price.

But when the marketing slides detail the company's flagship desktop processor as a solution for comparatively low-cost PCs, we're left with a processor designed specifically for that market segment. Of course, it's hard to believe that this is anything more than an advertising spoiler, but enthusiasts who were hoping to see AMD's Bulldozer with an architecture that will destroy Sandy Bridge and fight Sandy Bridge-E will have to adjust their expectations. Instead, the company is clearly not going to spend as much money on hardware as they have done in the past. Anyway, for now.

It's weird, isn't it? On the other hand, Intel's experience with Sandy Bridge showed that the power user community doesn't need $1,000 processors to get incredible performance. AMD's unlocked $200 chips are capable of 4.5 GHz and also perform better in a number of benchmarks (including gaming tests). If AMD can offer even better value for money in this market, then, as they say, go ahead and do not hesitate.

At least on paper, AMD's processor lineup looks comprehensive and competitive. The AMD processor line has seven models in the FX family, ranging from the FX-8150 to the FX-4100. They are all based on AMD's Zambezi design, manufactured by Globalfoundries on a 32nm process, and consist of approximately 1200,000,000 transistors (AMD recently revised its transistor estimates upwards). 315mm² heads are smaller than Thuban processors (346mm²) but larger than Deneb processors (258mm²). For comparison, Sandy Bridge processors are 216 mm².

AMD Processors FX have eight, six and four cores depending on the model (respectively four, three, and two Bulldozer modules). The numbers at the beginning of the designation of the model range will help you determine the number of cores: FX-8xxx - eight, FX-4xxx - four. The three numbers that follow the main ones indicate the level of performance. They are not consistent with clock speed, TDP, or L2 cache. You just have to remember that in the FX-8xxx segment, the 8150 is better than the 8120, which is better than the 8100.

All processors in the FX line come with an unlocked multiplier, which can make them a tidbit for overclockers, depending on how aggressively AMD is tuned to the speed potential of their processors. Sometime in 2008, one enthusiast jumped over the well-known 4 GHz with the Core i7-920, because the chip itself allowed it. It remains to be seen whether it is possible to achieve the same success with a 32nm Globalfoundries chip.

One thing is clear - for a reasonable price (compared to the offers of a competitor represented by Intel), AMD offers not a computing monster, but a quite serious solution for a home computer capable of solving absolutely any task in terms of complexity.

Recently, AMD's position in terms of desktop processors in the market has been a little shaken. First, due to the lack of progressive architectures, the company had to reduce the prices of its products. After some time, AMD completely left the position of high-performance CPUs of the top price policy. And all this was energized by another failure associated with the release of processors based on the Bulldozer architecture, on which high hopes were pinned. Bulldozer was supposed to compete with older Intel processors (LGA 2011 and LGA 1155). However, the new architecture greatly disappointed with its slowness and high power consumption. As a result, Bulldozer can only compete with mid-range Intel processors due to twice the number of cores.

But, fortunately, a series of failures did not deprive the company's engineers of enthusiasm, and a year after the release of Bulldozer, a new improved version of this microarchitecture called Piledriver was launched. The results of testing the AMD FX-8350 processors showed that the time spent on improving the line was not wasted. As the senior member of the Vishera line of desktop processors, AMD's platform performance has been dramatically improved. From testing, it became clear that the performance increased by about 15%, so to consolidate its successful development and enhance the effect, AMD set a very affordable price. All these events made people talk about Vishera processors in a positive way.

The flagship FX-8350 is even faster than previous generation AMD processors. Given the democratic pricing policy of the company, FX-8350 can be advised to install on an inexpensive desktop system that involves solving resource-intensive tasks in the form of creating and processing high-resolution content or final rendering. However, before making a final decision, it is worth considering its shortcomings. First of all, there is a high level of energy consumption. The next thing to note is the imperfect load distribution across the eight cores.

It is also worth paying attention to the FX-8320. This model is practically in no way inferior to the FX-8350, but it is an order of magnitude lower in terms of price. In professional applications, the speed of the FX-8320 is at its best. And given the fact that modern AMD FX processors are endowed with unfixed multipliers, it will not be difficult for the FX-8320 to be overclocked to the flagship level and even higher.

The six-core modification of Vishera, at first glance, does not stand out from all the presented models. Due to the deactivation of one of the 2-core modules in the FX-6300, its peak performance is quite low compared to the 4-core Intel ones. However, AMD far-sightedly used pricing tactics for the FX-6300, which in this respect competes not with the Core i5, but with the Core i3. This approach opens up great prospects for the six-core Vishera, especially since, unlike the developments in the Core i3 line, the FX-6300 can be overclocked.

In the AMD FX-4300 processors, the company's engineers disabled half of the cores and cut down half of the L3 cache, and therefore this category does not differ in high performance, but is endowed with high efficiency rates.