Publication date:

15.06.2009

Over the past six months, laptop sales have increased significantly, and netbooks have played the most significant role in this. Interestingly, sales of expensive laptops fell. This is understandable: buyers have learned to value money and invest it wisely.

Against the general backdrop of the global crisis, such giants as ASUS, Acer and Dell announced high percentages of profits thanks to the sale of netbooks.

Where do the legs of netbooks come from?

The concept of netbooks appeared in 2008 at the Intel Developer Forum in Shanghai. According to Intel, the main vector for the development of mobile devices is the creation of low-cost mobile Internet devices (MIDs). Such devices provide the main thing - access to networks and information at any time and anywhere for a long time. These devices must be compact and truly portable. At IDF, Intel presented the corresponding Intel Centrino Atom platform and thereby announced the appearance of devices built on the Atom architecture and called netbooks by Intel.

Netbooks (netbook) are a family of laptops designed to work on the Internet and for nothing else (net - network, book - short for notebook).

Netbooks belong to a class of laptops called subnotebooks, which are small, portable laptops with ultra-low power consumption. Such laptops have a low cost (from 200 to 600 USD), a weight of about 1 kg, and a small display (from 7 to 10 inches). As you know, networking does not require high performance, so you should not expect high performance from netbooks.

Intel Centrino Atom processor technology, formerly known as Menlow, includes the first Intel Atom processor (formerly known as Silverthorne) and Intel System Controller Hub ( Poulsbo). These components were developed from the very beginning for the MID segment.

All mobile systems are rated based on performance per watt of power consumption, showing that there is always a trade-off between performance and power consumption. Well, as you know, energy-intensive devices require larger power supplies. Consequently, by reducing energy consumption, developers automatically reduce the size of devices.

Intel Atom Architecture

The new microarchitecture is based on a 45nm manufacturing process using new metal gate transistors with high-k dielectric. Surprisingly, Atom is fully compatible with the command set Intel Core 2 Duo, supports Hyper-Threading and SSE3 multimedia command set extension. Even Intel VT virtualization is supported. True, it is not needed for mobile tasks, but apparently the developers want to use these processors as an ideology for the development of architecture in all directions, creating a kind of universal process, and then refining it in one direction or another. We can say that, taking into account the inherent capabilities, the Intel Atom microarchitecture is the basis for future processors.

The Intel Atom microarchitecture includes revolutionary power management features such as Intel Deep Power Down (C6), Enhanced Intel SpeedStep, Active Clock Gating, CMOS mode, and Split I/O. All these innovations allow you to optimize energy consumption and heat dissipation both in general and in standby, operating and peak load modes.

The Intel Atom processor is Intel's smallest processor today. It is even smaller than the chipset chips! At the same time, it is the fastest processor in the world, consuming less than 3 W of electricity. One chip with an area of ​​less than 25 mm2 contains more than 47 million transistors (significantly less than desktop processors).


The thermal power of the new processors is 0.65-2.4 W, average power consumption does not exceed 160-220 mW , and in the standby state these devices consume only 80-100 mW.

The idle power consumption of the Intel Atom processor was measured as the power consumption in the Intel Deep Power Down state (state C6). Intel Deep Power Down Technology (C6) puts the processor into a state with minimal power consumption by turning off the main system bus clock, the PLL=Phase-locked loop (PLL=Phase-locked loop), and L1 and L2 caches.

From a circuit design point of view motherboard The PLL controls the dynamic reduction of the system bus frequency and its auto-tuning. If you optimally configure the system so that the bus frequency quickly dynamically decreases when there is no load, then this can save more than half the energy supplied to generate pulses on the bus.

The cache memory must be disabled for obvious reasons: it contains the bulk of the processor's transistors: by disabling them, we will save the second largest share of the source energy.

Intel Centrino Atom processor technology-based kit, including Intel System Controller Hub and Intel Atom processor at 800 MHz, 1.10, 1.33, 1.60 or 1.86 GHz , costs 45, 45, 65, 95 and 160 US dollars respectively (for orders of 1000 pieces). As we can see, such solutions are not expensive and allow you to create systems within 200-400 USD.

Intel SCH Family was developed from the very beginning as a high-performance, energy-saving solution for single-chip devices with high degree integration. The Intel SCH controller includes integrated graphics with hardware-accelerated video decoding, supporting HD 720p and 1080i modes. All standard desktop and handheld I/O interfaces are supported, including PCI Express, SDIO and USB.
Intel introduced three versions of SCH supporting modules DDR memory 2 400/533 MHz with a capacity of 512 MB/1 GB, video as in standard definition, and high definition, Intel High technology Definition Audio, DX9L and OpenGL.
At the driver level there is support for various operating systems.

Mobile Internet devices based on Intel Atom are going to be produced by Aigo, Asus, BenQ, Clarion, Fujitsu, Gigabyte, Hanbit, KJS, Lenovo, LG-E, NEC, Panasonic, Samsung, Sharp, Sophia Systems, Tabletkoisk, Toshiba, USI, WiBrain and Yuk Yung.
As you can see, most of these companies represent the segment of mobile devices, communicators, handheld computers, and a few represent the subnotebook segment.

Application in embedded systems

Embedded solutions are industry and industrial solutions (primarily automation controllers, medical and military systems, measuring instruments), characterized by high reliability and low power consumption. Such systems are small in size, low profile and passively cooled. For a long time, this segment was coexistent with the Intel Celeron M with the i945GME Express chipset and the less “gluttonous” VIA C7. The time has come to displace these apologists of constancy - the change in architecture has reached the embedded systems segment.
This was to be expected: all trends were towards reducing die sizes and crossing the performance of desktop chips, optimization from the server segment and mobile chips with low and ultra-low power consumption. And the result of the combination was Intel Atom.

Intel Atom Processor and Intel SCH Controller It was decided to promote it in the embedded systems segment. In this segment, the company offers two processor models: Atom Z530 with a frequency of 1.6 GHz and Z510 with a frequency of 1.1 GHz. They are designed for a 7-year life cycle. Naturally, Intel provided developers with all the means to implement new CPUs in embedded systems.

The new architecture on 2 chips (single-chip chipset) will reduce the size of devices by more than 80% compared to the previous solution, which included three chips (Celeron M ULV and 945GME Express).

Atom processors in the bottom line

So that's it Intel crystals Atom is made using a 45-nm process technology using metal gates and Hi-k dielectrics and can be divided into CPUs for netbooks and nettops and CPUs for mobile Internet devices.
These crystals partially inherited a lot from the Centrino 2 architecture, but were optimized and cut down in some places.

CPU for netbooks and nettops

All these crystals have 1 core, except model 330 : it received 2 cores and 2 L2 caches with a capacity of 512K per core (total volume - 1MB). All other chips have a 512 KB L2 cache.

Processors with the letter Z in the marking have the lowest power consumption - from 0.65 W (Z500) to 2.4 W (Z550). Models Z500, Z510, Z515 operate at a bus frequency of 400 MHz (to reduce power consumption).
Z520, Z530, Z540, Z550 more energy-intensive, as they are clocked at a bus frequency of 533 MHz.

All these models appeared in the 1st quarter of 2009.

Previously, there was only one model N270. It is designed for a heat dissipation (TDP) of 2.5 W (temperature up to 90 degrees, versus 85 for the Z530 model with the same frequency). It differs only in that its core supply voltage varies within 0.9V-1.1625V, while for the Z530 it varies from 0.8 V. That is why the N270 consumes 2.5 W and not 2.4 W. In fact, the Z530 can be considered an optimized model of the N270.

The N270 crystal has dimensions of 26 mm2 (22x22 mm), contains 47 million transistors and is housed in a new PBGA437 package. This means that it cannot be installed in existing systems Centrino 2.

All netbook manufacturers that introduced their solutions in 2008 based them on the N270.

The hottest Intel Atom crystals - models 230 and 330. In fact, these are the same processors. The difference is that the 330 model contains 2 identical cores and, accordingly, a cache with 2 times the capacity.
Well, as a result, the TDP of the 330 increased from 4 W to 8 W.
By the way, only these crystals of all Atoms are 64-bit!

CPU for mobile internet devices

In fact, these are the same processors with the same specifications, but in a slightly different circuit design.
Instead of a standard chipset, they are supposed to be used in conjunction with the system's controller-hub crystals Intel UL11L, US15L, US15W.

Desktop CPU

In principle, Atom processors can be easily used to build inexpensive office PCs, which many OEM assemblers have taken advantage of.

This assumes the use of Atom N270, 230 and 330 processors with the i945GC Express chipset.

In general, we can summarize that Intel Atom is the most mobile and low-power processor for netbooks and mobile systems for now.

Intel Atom processors are built on a completely new architecture, they are characterized by low power consumption and are suitable for both mobile Internet devices (MIDs) and low-cost PCs. Among the advantages, we note support for x86, which allows you to run a wide range of available programs. In our article we will compare the performance of the Atom 230 platform with competing solutions from AMD, Intel and Via.

Introduction

For several months now, a new Intel processor has been rumored, designed for MID (Mobile Internet Devices, mobile Internet devices) and designed to compete with ARM processors. Originally known as "Silverthorne" and "Diamondville", the new processors were called "Atom". And they have a lot of surprises.

Interesting choice

Atom processors are amazing because they integrate modern features (EM64T, SSSE3, etc.) into the old architecture. Atom is the first x86 processor with queuing instructions since the Pentium. When developing the processor, Intel carefully monitored power consumption and manufacturing costs, even at the expense of performance. Therefore, you should not expect new competitors to Core 2 Duo from Atom. But what do Atom processors actually offer? Let's get a look.

Intel and lower power consumption

Power consumption and processor integration into a portable or embedded device have always been issues for Intel, and this isn't the first time the company has offered processors in this area. But Atom is radically different from previous attempts in that it is based on a new architecture specifically designed to minimize power consumption.

Short story

Before Pentium M

Back in the days of the 80386, Intel offered versions with reduced energy consumption, aimed at the mobile sphere. The 80386EX, for example, had some of the chipset's features integrated into the processor, and the system consumed significantly less power than the standard 386. Then came the 486, Pentium, and Pentium II (Dixon, with 256 KB of on-chip cache) versions with lower power consumption. But, in any case, they used a similar, if not identical, architecture to their desktop "brothers". In practice, the processors performed efficiently, but the differences between the standard version of the CPU and the mobile processor were small.

Pentium M

Released in 2003, the Pentium M processor was revolutionary in that it used a different architecture than the Pentium 4 and consumed significantly less power while still delivering high performance. Yes, the processor could be called a derivative of the Pentium III, with the same shortcomings, but subsequent improvements to the Pentium M, which led to the Core 2 processors, only increased power consumption. Intel tried to release low-power processors (A1x0, for example), but they were Pentium M variants with reduced frequencies.

Atom changed everything

The Atom processor is built on a different architecture, it was originally designed to minimize power consumption, so the design of the processor is completely new. This is not an adaptation of old architecture. Today, Intel can offer processors that consume very little power: high-end versions of Atom consume less power than the typically slow ULV versions of standard processor architectures.

Atom Z500 and SCH (Poulsbo)

The first generation of Atom processors, formerly known as "Silverthorne", received model numbers Z5x0. Atom Z500 processors are aimed at MIDs (the famous Mobile Internet Devices) and are paired with the new Poulsbo SCH (System Controller Hub) chipset.

Competitor to ARM processors?

Since the orientation is announced on MID, Intel's competitor is obvious - ARM processors. This is a very popular architecture (the vast majority of phones, PDAs and GPS navigators use it), supported by processors from many manufacturers (ARM licenses the instruction set), it gives good performance with very low power consumption. In the portable space, with the exception of some rare devices based on the MIPS architecture (PSP pocket game console, for example), ARM processors make up the majority. Intel, interestingly, also produced ARM processors for various devices(XScale, then the division was sold to Marvell), and today it offers products such as, for example, processors for RAID controllers (the same IOP333). In practice, switching from ARM architecture to x86 is not a problem - Linux supports both, as does Windows CE (used in many GPS navigators) and Windows Mobile(at least older versions). Also, x86 can run the latest Windows versions, and the architecture benefits from broader software (and technical) support compared to ARM processors.

Z500 processors

Before we dive into the Atom architecture, let's take a look at the Z500 line. These processors are tiny, the packaging size is only 13 x 14 mm. The processors consist of approximately 47 million transistors (more than the original Pentium 4), equipped with 56 KB L1 cache (24 KB for data and 32 KB for instructions), as well as a 512 KB L2 cache. The processors operate on a standard Intel bus, which is familiar to us from Pentium 4 processors. The bus frequency is 400 MHz (QDR) or 533 MHz (QDR). There is also support for SIMD instructions, from MMX to SSSE3, EIST and Hyper-Threading (back!). Please note that the latter feature is only available on some models (with 533 MHz (QDR) bus).

Poulsbo, chipset for Atom

The SCH (System Controller Hub) chip is a “single-chip chipset,” that is, it combines the north and south bridges on one chip. The chipset is designed for Atom processors, and only it is compatible with new features such as using the bus in CMOS mode (we'll talk about this a little later). SCH is feature-rich - it contains a built-in GMA graphics core (based on the PowerVR architecture), HD Audio (simplified, supporting only two channels), a PATA controller (Ultra DMA 5, 100 MB/s), and also supports two PCI Express lanes (For Wi-Fi cards, For example). There are three SDIO/MMC controllers and support for eight USB ports with the ability to use one in client mode. The choice of the PATA interface is quite logical: flash memory card controllers usually use this format, for example, Compact Flash. Three SD controllers may seem like a strange choice, but some memory uses just such an interface (OneNAND, for example). The DDR2 controller in the SCH chip supports memory with a voltage of 1.5 V instead of 1.8 V according to JEDEC specifications. This small detail also helps reduce energy consumption.

Poulsbo Graphics Controller

For graphics, we received a new GMA 500 controller. It uses a unified architecture and supports shaders 3.0+. Interestingly, the graphics controller has hardware support for decoding H.264, MPEG2, MPEG4, VC1 and WMV9 formats. The GMA 500 clocks at 200 or 100 MHz, depending on the chipset version, and supports DirectX 10 (hardly a big deal, but worth mentioning), although the drivers only support DirectX 9. Please note that the graphics core is not of Intel origin. Unlike other GMAs, it is built on PowerVR technology.

Interesting TDP

For Atom Z500 processors, the thermal package (TDP) varies from 0.85 W (for the 800 MHz version without Hyper-Threading) to 2.64 W (for the 1.86 GHz model with “Hyper-Threading” support). SCH consumes approximately 2.3 W in its most advanced version, which gives the SCH + CPU combination less than 5 W. Compared to existing solutions, the progress is obvious: Via Nano, for example, is stated at 25 W for the 1.8 GHz version, and Celeron-M ULV - 5 W at 900 MHz.

Atom N200 and i945

For Atom targeting standard computers,Intel offers another line (Diamondville). Atom processors of the N200 and 200 lines are aimed specifically at standard computers, but more, of course, at cheap portable PCs such as the Eee PC and competing solutions.

The Atom N200 processors are similar to the Atom Z500, the only difference being the support for 64-bit EMT64 extensions, which is present in the N200 and 200, and the lack of EIST support. Thus, Atom 200 processors cannot change the frequency on the fly. The prices are very attractive: Atom N270, with a frequency of 1.6 GHz (533 MHz bus) and 2-W TDP costs only $44. And the 230 version, with 4-W TDP, will cost only $29 (at the same frequency).

Veteran Chipset: i945

The main problem with the Atom N200 processor is the chipset: Intel only offers i945 variants. This chipset, not only is outdated (it was released in 2005), has a major drawback: it consumes a lot of energy (22 W in the GC version). i945 chipset supports modern technologies: SATA (2), PCI-Express (1 line via ICH7), HD Audio, etc. It is clear that it works with DDR2 memory (two channels) and uses the integrated GMA 950 graphics core. As you can guess, using an old chipset (from the Napa platform) with a TDP that is 10 times higher than the thermal package of the processor is not the best idea. But nothing more interesting has been proposed yet. The laptop PCs use the i945GSE chipset, which consumes only 5.5W (4W north bridge and 1.5 W south bridge). It's clear that its performance isn't nearly the same - especially in 3D graphics, since Intel has lowered the GMA frequency (from 400 to 133 MHz).

GMA 950

Now let me say a few words about the GMA 950, the integrated graphics core in the Intel i945 chipset. It supports DirectX 9 and is capable of running the Aero interface, and is also widely available in laptops with a Core Duo processor. Performance is weak, there is no hardware support for decoding HD formats. Moreover, the graphics core is very sensitive to memory bandwidth, and the drivers are not optimized. Finally, Intel uses several frequencies for the graphics core - from 400 MHz for the i945G version (desktop PCs), to 250 MHz for laptops and 166 MHz for ultraportable models (with a proportional loss of performance). The version used by Atom processors (i945GSE) is limited to 133 MHz, although the i945GC chipset has a graphics core running at 400 MHz.

Atom architecture: another execution and "Hyper-Threading"

Atom processors use a new architecture, albeit with older technologies. This is the first x86 processor from Intel with sequential (instead of out of order) execution of instructions since the Pentium, which appeared back in 1993. All other Intel processors since the P6 use out-of-order execution.

Next execution

Without going into too much detail, think of a processor as a device that receives instructions one after another and places them on a conveyor belt. In the next architecture, instructions are executed in the order in which they were received. And in an out-of-order architecture, the order of instructions issued to the pipeline can be changed so that they are executed as efficiently as possible. The advantage of an out-of-order architecture is that the number of waits can be reduced. For example, if you have instructions simple calculation, a memory access instruction and another simple calculation instruction, then in the regular architecture they will be executed one after another, and in the out-of-order architecture the processor can perform two calculations in parallel with long memory access, which saves time. But what is quite surprising is that usually the next architecture has a short pipeline, but Atom has 16 stages, which in some cases leads to disadvantages.

"Hyper-Threading"

"Hyper-Threading" technology appeared with the Pentium 4 processor. It allows two threads to run simultaneously, optimizing the pipeline load. Of course, this is not as efficient as two physical cores, but the technology forces the OS to think that the processor can handle two threads at the same time, and this can improve the performance of the computer. On an Atom processor with a long pipeline and the old regular architecture, "Hyper-Threading" works very effectively; the technology can significantly increase performance without a noticeable impact on TDP. Intel claims only a 10% increase in power consumption.

Computational core

Otherwise, Atom is equipped with two ALUs (integer units) and two FPUs (floating-point units). The first ALU performs shift operations, and the second ALU performs branches. All multiplication and addition operations, even with integers, are performed on the FPU units. The first FPU is very simple and limited to addition operations, while the second is responsible for SIMD and multiplication/division operations. For 128-bit calculations, the first branch is used in conjunction with the second (both branches are 64-bit).

Intel optimized core instructions

If you look at the number of clock cycles it takes to execute an instruction, you'll find something interesting. Some instructions are fast, others are (very) slow. "mov" or "add" instructions, for example, are executed in one clock cycle, as on the Core 2 Duo, and multiplication (imul) instructions take five clock cycles, as opposed to just three on the Core microarchitecture. To make matters worse, 32-bit floating point division, for example, takes 31 clock cycles, compared to just 17 (or almost half) for the Core 2 Duo. In practice - and Intel confirms this - Atom is optimized for fast execution basic instructions, that is, the processor sharply reduces performance on complex instructions. You can check this by simply running Everest (as an example), which has a tool for measuring instruction execution times.

Cache and FSB

Intel chose a very unusual Atom organization, but without sacrificing performance, which is important for a processor with a regular architecture.

24 + 32 KB: asymmetric cache

Atom's L1 cache is 56 KB: 24 KB for data and 32 KB for instructions. This asymmetry, quite surprising for Intel, is a consequence of the cache structure. Intel uses eight transistors to store one bit, as opposed to six transistors in standard cache. This technology allows you to reduce the voltage applied to the cache to store information. It seems that this move to eight-transistor cells was made late in the process, when the processor design was already close to completion, so in order to fit the cache within the previous boundaries, its size was reduced - this explains the 24 KB for data.

L2 cache 512 KB, shrinkable

The L2 cache capacity is 512 KB, it operates at the same frequency as the processor. The 8-way cache is classic and quite close in performance to what was used in the Core 2 Duo (its latency is 16 clock cycles compared to 14 for the Core 2). One of the new features is that parts of the cache can be automatically disabled if a program doesn't need a lot of cache memory. In practice, the cache switches from 8-way to 2-way mode, that is, from an available volume of 512 to 128 KB. This technique allows you to save a few more precious milliwatts.

FSB: two operating modes

The Atom processor uses the same FSB as other Intel processors since the Pentium 4. It operates in Quad Pumped (QDR) mode and GTL signaling technology. Interesting: Atom uses a different signal technology - CMOS mode. GTL is efficient (the bus can reach 1600 MHz QDR), but consumes a lot of power, and CMOS allows for lower bus voltage. Technically, GTL uses resistors to improve signal quality, but they are hardly necessary except at high frequencies. With an Atom processor and a bus limited to 533 MHz (QDR), you can go into CMOS mode - the resistors will be disabled and the bus voltage will be cut in half. At the moment, only the SCH chipset supports CMOS mode on the FSB.

Energy consumption: tests and theory

Power consumption is critical for this Intel platform, so many steps have been taken to reduce it. In addition to the chipset, which consumes a lot of energy compared to the processor, Atom itself has acquired many interesting features.

Bus and cache

As we already said, Intel has worked a lot on the bus and cache. A different mode for the bus (CMOS) was developed, and the cache can automatically turn off its sections depending on the load. Such features help reduce power consumption, as do the next architecture and 8T SRAM L1 cache cells.

State "C6"

In addition to reducing the processor voltage to 1.05 V, Atom has new mode waiting "C6". Recall that "C" modes (0 to 6) are low-power states, and the higher the number, the less power the CPU consumes. In "C6" mode, the entire processor is almost completely turned off. Only the cache memory of a few kilobytes (10.5) remains active to maintain the state of the registers. IN this mode The L2 cache is emptied and disabled, the supply voltage drops to just 0.3V, and only a small portion of the processor remains active to enable wake-up. The processor switches to "C6" mode in about 100 microseconds, that is, quickly. In practice, Intel says that "C6" mode is active 90% of the time, which reduces overall power consumption (it is quite clear that if you run a program that loads the processor, or even watch a video on Flash, the processor will switch to this mode will not pass).

It should be noted that both Intel chipsets that can be used with Atom N200 processors consume a lot of power: the Atom 230 uses an i945GC, which consumes 22 W (4 W for the CPU), and the Atom N270 comes with an i945GSE, which burns 5.5 W (2.4 W for CPU).

On practice

Is the Atom processor so low-power in practice? As for the processor, yes. As for the platform aimed at cheap desktop computers (NetTop), the answer is also positive, but... Why "but"? Because the chipset consumes a lot of energy, and the TDP for the processor is stated to be 4 W or 2.4 W for the mobile version. Our test motherboard consumed 59 Watts in idle mode, we got 62 Watts when maximum load(with processor, 1GB DDR2 memory and 3.5" hard drive). It is quite clear that the given numbers refer to the complete platform (without a monitor), and not to one motherboard, and also include losses on the power supply (our model had an efficiency of approximately 80%). Power consumption can be called both small and large - a little for a desktop computer, but a lot for absolute values. We should mention that a recently tested 1.5GHz Via C7 motherboard with the same configuration consumed less power: 49W idle and 59W under load.

Tests 1: Atom vs Pentium E and Sempron

For our tests, we took a Mini-ITX motherboard from Gigabyte, equipped with an Atom 230 processor and an i945GC chipset. The board has one DIMM slot (DDR2) and one PCI slot - that is, you will not get a modern video card. What’s interesting is that the chipset, which, recall, consumes 22 W, is actively cooled, and for the processor a simple aluminum radiator.

Since this motherboard is designed for computers entry level, we took two solutions for comparison: Pentium E2160 (1.8 GHz), an entry-level dual-core processor based on the Core microarchitecture, and Sempron 3400+ (in this case, Socket 754). During our tests, the two processors were set to the same clock speed as the Atom (1.6 GHz). For the Pentium E2160, the GA-GM945-S2 motherboard was taken. It has the advantage of being built on (almost) the same chipset as the Atom motherboard - i945G. For Sempron we took an nForce4 motherboard.

Three motherboards were tested on the same OS - Windows XP Service Pack 2 with all updated drivers. We used DDR2-667 memory (1 GB) on the Intel platform, as well as 1 GB DDR400 DIMM on the Sempron platform. Finally, we took a 74 GB hard drive as a test Western drive Digital Raptor.

Test results

We decided to compare the three platforms at equal frequencies, conducting several real and synthetic tests.

In Cinebench R10, the Sempron processor was placed between the Atom and Pentium E, and the combination of Atom with Hyper-Threading technology proved its effectiveness (with Hyper-Threading, performance increases by 1.53 times). Please note that the gain on the Pentium E, equipped with two physical cores, is not particularly higher: 1.86 times.

In Sandra, a synthetic test, the difference between the three processors is impressive. Pentium E turned out to be noticeably faster. Note that the difference between Atom and Sempron may seem small, but the tests are multi-threaded, and the Sempron only has one core, while the Pentium E has two cores, and the Atom supports "Hyper-Threading", which gives a significant increase.

In the 3DMark 06 and PCMark 06 CPU tests, the Pentium E processor is quite confidently in the lead, and Sempron, as usual, is located in terms of performance between the Atom and Pentium E.

In this test, which is so loved by overclockers, although its code is old and not optimized, the Atom processor is much inferior to its competitors.

Finally, we carried out a test which consists of compression in WinRAR files with a capacity of about 1 GB. Since Sempron uses a different memory subsystem (DDR) and a discrete video card, we did not include it in this test. In practice, the difference between the platforms turned out to be smaller than in synthetic tests, but the Pentium E is still about twice as fast.

Tests 2: Atom vs C7-M and Celeron

We decided to compare our Atom platform with two other systems that can compete with the Mini-ITX test platform. The first system is a Via PC3500G motherboard with a C7 processor; the second is an entry-level processor often found in ultraportable computers - Celeron-M (Dothan).

Comparison with C7

The Via PC3500G motherboard has a micro-ATX form factor, it contains the CN896 chipset paired with a 1.5 GHz C7 processor. For our test, we clocked the Atom to the same level as the C7 (12 x 125 MHz, or 1.5 GHz). Memory, hard drive and OS were the same.

In Cinebench R10, as you can see, the Atom processor was faster than the C7, but not by much - at least with one thread. On the other hand, Atom's support for "Hyper-Threading" has led to a significant lead.

In PCMark 05 you can see that the Atom platform, even at an identical frequency, turned out to be faster than the C7 platform. There are several reasons for this. PCMark 05 is a multi-threaded test, like many modern programs, so Atom with "Hyper-Threading" has an advantage. Besides, Intel chipset significantly faster (or not as slow, to be more precise) than Via.

Finally, we measured the power consumption of both platforms. Surprise: Thanks to the energy-efficient chipset, the Via platform consumed less power than the Intel platform. At idle, the PC3500G system consumed 49 W, while the GA-GC230D required 59 W. However, as the load increased, Atom began to consume only 3 W more, and the Via platform increased power consumption by 10 W, remaining, however, still below the Intel level. All measurements were taken from electrical outlet, that is, the result was influenced by losses on the power supply (efficiency 80%).

Comparison with Celeron M

To compare with the Celeron M, we took a laptop with this processor based on the Dothan core. We did not conduct PCMark tests, since the hardware of the two configurations is very different, and it is incorrect to compare the results. As with the C7, we clocked the Atom down to Celeron M levels (1.3GHz in this case).

In a synthetic test like Cinebench R10, you can see that the Celeron is about twice as fast at identical frequencies. In any case, the "Hyper-Threading" technology added some points to Atom.

As tests show, the Atom is between the C7 and Celeron M at identical frequencies. Considering that both processors are used in cheap PCs (Netbooks), the C7 with frequencies close to Atom, and the Celeron M at lower frequencies, it can be argued that the performance of Atom computers will be more or less identical to modern systems. On the other hand, in modern laptops Celeron M operates at high frequencies of 1.6 GHz and 1.86 GHz, so the superiority over Atom will be noticeable.

Overclocking and 3D

Finally, we conducted tests in two areas that are unlikely to be relevant to the Atom platform, but for us and readers they are very interesting.

Since our motherboard did not have PCI Express or AGP slots (and PCI video cards increasingly difficult to find), we limited the tests to the GMA 950. For comparison, we took a Gigabyte motherboard based on the same chipset with a Pentium E 2160 processor at 1.6 GHz, equal to the Atom. Both computers use the same GMA 950 integrated graphics core at 400 MHz, and the processors run at the same 1.6 GHz frequency. Both computers are equipped with one DDR2-667 DIMM.

As you can see, 3DMark 06 performance at 640 x 480 without filters is very poor. In addition, the Pentium E turned out to be significantly faster than the Atom.

But it should be remembered that in portable PCs Atom will be used in conjunction with the i945GSE chipset, and the GMA 950 in this version will operate at only 133 MHz.

Overclocking Atom

The Gigabyte Mini-ITX motherboard provides few options for overclocking: you can only change the FSB frequency, but from 100 to 700 MHz. On our CPU model, the multiplier is locked at 12, and the FSB frequency is 133 MHz. We were able to achieve stable operation at 1.8 GHz (12 x 150) without raising the voltage, as well as at 1.86 GHz (153 MHz bus) by raising the FSB voltage to Motherboard BIOS boards (+0.3 V for bus). Performance increased linearly, as did power consumption: from 62 to 65 W for 1.6 and 1.8 GHz, respectively. And after overclocking Atom to 1.86 GHz, the platform's power consumption was 67 W. The difference can be explained by the rise in bus voltage. It should be remembered that power consumption increases not only due to the CPU, but also due to overclocking the chipset.

Why is there no HD test?

Why didn't we test HD video playback? The first reason is that Atom processors are not designed for this. Intel is targeting them at low-cost NetTop computers designed for browsing the Internet rather than playing Blu-ray discs. However, just for fun, we tried to watch HD-DVD, but the Power DVD player refused to start without modern video card, capable of taking over part of the video decoding. We tried to play HD videos downloaded from the Internet, but here too we were disappointed. The result was affected by the type of player used, and the video quality did not match commercial HD discs. Decompressing a multi-megabit/s DivX 720p stream is one thing, but 36 megabit/s H.264 video is another.

Conclusion

What is our conclusion about the Atom platform? The impression is mixed. The processor itself can be considered a success - it is inexpensive, consumes very little power, and although its performance is not high, it is quite sufficient for the target market (inexpensive PCs intended primarily for browsing the Internet). In addition, the support for "Hyper-Threading" is nice. But the chipset paired with the processor is disappointing. Intel only offers two options, and they can be criticized. SCH Poulsbo seems efficient, but it hardly makes sense to install it in standard PCs due to its MID orientation (there is no SATA port, for example), and the i945GC and i945GSE chipsets are suitable for PCs, but they also have drawbacks - a small set of functions, very low performance of the integrated graphics core in 3D (and all more apps it is used), and the chipset consumes significantly more energy than the processor itself.

The feeling is that Atom is a trial attempt - it succeeds from one point of view and fails from another. Will they stand up? computer manufacturers and ordinary consumers on the Atom side? Without a doubt, and for two reasons: prices and marketing. The platform will allow you to assemble computers at very low prices, and Atom has already become a prominent brand. The opinion of an ordinary buyer about a possible configuration may be as follows.

"Eee PC 900 for $450 (good) with Celeron processor(bad) at 900 MHz (bad)."

Or like this.

"$450 Eee PC 901 (good) with Atom processor (good) at 1.6 GHz (good)."

In other words, Atom processors will appeal to the public more, even if the practical difference is small.

The platform turned out to be truly paradoxical: a successful processor (even if the performance in absolute terms is low) and a chipset simply unworthy of it. Overall, there's little difference between the older platforms, so let's hope Intel comes up with new chipsets that are better future-proof.

Advantages.

Price $29 for Atom 230;
low processor power consumption;
"Hyper-Threading" shows its best side.

Flaws.

Weak overall performance;
bad chipset;
very low 3D performance;
unbalanced platform.

Behind Last year In the universe of Intel Atom processors, a series of literally galactic cataclysms occurred, both destructive and creative. As a result, it was, one might say, completely rebuilt. In this post we will remember the history of Intel Atom, talk about latest events, related to them, and in conclusion we will get acquainted with new models from this family, more similar to Intel Xeon.

Intel Atom was conceived by Intel as budget solution with minimal power consumption for various types of mobile devices. The first Atom appeared in 2008, it was made using 45 nm technology, over time the process technology was reduced to 14 nm. The success of Atom processors varied greatly depending on their application. So, some of them definitely appeared in right time and became widespread in the then newfangled “netbooks” (“laptops for working on the network”). Such netbooks did not work quickly compared to laptops with Core processors, but they were cheap, compact, did not have a cooler (and the problems associated with it), and sold well. Let's remember the super popular ASUS Eee PC 901, and note that netbooks were produced by such reputable manufacturers as HP, Lenovo, Dell and Sony.

ASUS Eee PC 901

The fate of Intel Atom as an x86 competitor to ARM processors for smartphones and tablets was much less successful. Although there is a very noticeable result here - the release in 2015 of Microsoft Surface 3 with an Intel Atom x7-Z8700 processor.

It should be noted that Intel has done a lot in this key area - mobile Atoms latest generation, which appeared in 2013-2014, in terms of performance they are far from their first ancestors, and in terms of capabilities they are closer to Intel Core: they have a completely updated graphics core - Intel HD Graphics, the microarchitecture has been changed to out of order execution, vector graphics have been added SSE4 instructions. However, interest in Atoms on the part of manufacturers was moderate: despite decent energy efficiency indicators (as stated by highly respected resources), the operational advantages were not so significant as to start a large-scale movement to change the platform. The financial issue also played an important role here: Intel Atoms were still more expensive than their ARM rivals.

By 2013, about a dozen Atom smartphone models were announced, some of which were never put into production. In our country, the Megafon-branded Orange San Diego smartphone was sold under the Mint brand.

Megaphone Mint

Intel actively promoted Android platform x86 among developers: created development tools, published training materials, held events. Moreover, a unique binary translator was created that worked on all Atom-based Android mobile devices, and on the fly translated ARM code into x86 instructions with almost no loss of performance.

However, as mentioned above, few Atom-based devices were released (compared to the number of ARM devices on the market), which led to a vicious circle - independent developers were in no hurry to release new x86-exclusive applications for these few devices, and device manufacturers , in turn, were in no hurry to release new models due to the lack of unique applications. In addition, the theoretical competitive advantage of Atom did not work - the ability to run desktop applications on mobile devices of the same architecture. Firstly, applications still had to be ported simply because of the mismatch between desktop and mobile operating systems (Windows or MacOS -> Android) and form factors, and this usually turned out to be even more difficult than a possible transition from x86 to ARM; and secondly, during the time of ARM's undivided dominance in the mobile market, all companies that wanted to create mobile versions of their desktop products had already done this for ARM devices, so the advent of x86 only added to their hassle - the need to create and maintain versions of the application for different CPU.
Be that as it may, during the global reorganization of 2016, the Atom direction for mobile devices was cut down at the roots.

However, the work of the processor creators was not in vain. A new direction has emerged at Intel, which has gradually become one of the key ones: “Internet of Things”. It is the totality of “Internet of Things” components that is the optimal consumer of Atom family processors with their low power consumption and wide range of characteristics. Thus we have imperceptibly approached our time.

To date, Intel has released great amount Intel Atom models, but there are not many of them that are current. This is primarily the newly announced E3900 series (its comparison table you see above). The series is designed to fill the need for high-performance “Internet of Things” hubs (Moderate requests are designed to satisfy the Intel Galileo, Edison and Curie platforms).

However, this is not yet the limit of “pumping” the Atom. Here we come to a new announcement. The “server” Atom C2000 line from back in 2013 is being replaced by the C3000 series, which is designed to raise Intel Atom performance to new heights. The flagship of the series will be a 16-core model - there have never been so many cores in Atom before. At the same time, all “branded” features - energy efficiency and affordable prices for server models - remain unchanged. So far, information is available about one of the younger models in the series - the C3338 processor. We expect announcements of the rest in the second half of 2017.

July 31, 2012 at 12:41 pm

When Atom is faster than Core?

• Intel Blog

Stuck in a traffic jam behind the wheel of a car theoretically capable of reaching speeds of more than 200 km/h, and watching cyclists on tricycles overtake me, I thought... no, not about how to get everyone on bicycles, and not about solving transport problems of humanity through teleportation, and... about Intel Core and Intel Atom processors. Namely - Atom compared to Core is, in fact, a scooter compared to a car. It consumes less fuel and costs significantly less. But on the other hand, the speed of a scooter is just as noticeably inferior to a car (despite even the ways to “accelerate” the scooter above the factory settings). But, nevertheless, in traffic jams or on narrow streets the scooter is faster. No wonder the scooter got its name from the English “ to scoot" - to run away, as it was successfully used by English teenagers to escape from the police.
Now let's get back to the CPU. Let’s replace “fuel” with “electricity” and “speed” with “performance”, and we get a complete analogy of the behavior of Inel Atom and Intel Core. But then it is reasonable to assume that there are “traffic jams” and “nooks and crannies” in which Atom will overtake Core. Let's look for them.

So, according to all generally accepted performance measurements, Intel Core is significantly ahead of Atom. In the "Performance" section of the Wikipedia article about Intel Atom, a harsh verdict is read: " approximately half the performance of a Pentium M processor of the same frequency"
If we compare Atom specifically with Core, then according to tomshardware tests, the Intel Core i3-530 defeats the Intel Atom D510 with a crushing score:


At the same time, it should be noted that tomshardware is clearly biased towards Atom. So, for example, if the running time of some task on Core-i3 is 1:38, then this is exactly how it is reported - “one minute, 38 seconds.” And if Atom performs something in 7:26, then, according to the authors, this is “about eight minutes.” But the main thing is to compare processors with different clock frequencies (2.93 GHz Core i3 and 1.66 GHz Atom) and not make allowances for wind. That is, the Core result must be divided by 2.93/1.66~1.76, which gives the final result of Atom losing from 2.15 to 2.6 times.

Why is Atom slower?

Quick answer: because it is cheaper and more energy efficient, which is incompatible with high performance.
Correct answer: Firstly, because Atom retains the FSB bus, while Core i3 has a memory controller integrated into the CPU, which speeds up data access. In addition, Atom has four times smaller size cache memory, and if the data does not fit in the cache, then slower memory access affects performance in full.
And secondly, the Atom microarchitecture is not Core2, used in Core i3, but Bonnell. In short, Bonnell is a continuation of the Pentium ideas, it has only 2 integer ALUs (versus three in the Core), and most importantly, there is no instruction reordering, register renaming, or speculative execution inherent in Core ).
How is it clear that in order to help Atom overtake Core, you need to:

1. Take a nanoset, a small set of data, so that it fits in the cache.
2. Try using float data to load the FPU rather than the ALU
3. If possible, deprive Core of the benefits of out-of-order execution.

Since everything is clear with the first two points, you can run the first tests.
They were carried out on my existing Intel Core i5 2.53 GHz and the already mentioned Atom D510, and were a set of mathematical function calls for float data with a built-in performance assessment “number of functions per second”, i.e. the bigger, the better.
The tests included the calculation of trigonometric functions both directly (C runtime, “x87” test) and by series expansion; using the Cephes library code; as well as vector implementation through SSE intrinsic functions (tests ending in _ps). At the same time, taking into account the difference clock frequencies, the results were scaled by 2.53/1.66~1.524
Tests compiled Microsoft Visual Studio 2008 with release optimization by default.


The data obtained fully confirms the first place of Intel Atom from the end. That is, the goal has not been achieved, let's move on to the next point - we will complicate the work of the Out-of-order CPU.

Making the task more difficult

Let's create an artificial test that will contain unpredictable branches containing computationally heavy functions, so that the result of Core's speculative calculations is constantly discarded, i.e. turned out to be unnecessary work.
Like that:
int rnd= rand()/(RAND_MAX + 1.) * 3; if (rnd%3==0) fn0(); if (rnd%3==1) fn1(); if (rnd%3==2) fn2();

Moreover, the functions will consist of chained calculations, so that Core cannot, by reordering instructions and renaming registers, calculate any of such expressions in advance, “out of turn.” Here is a simple example of such code
for (i=0; i< N; ++i) { y+=((x[i]*x[i]+ A)/B[i]*x[i]+C[i])*D[i]; }
By the way, similar functions are used in the above tests cephes_logf and cephes_expf, where the advantage of Core is minimal.
But, despite all the obstacles, Core still turned out to be faster. The minimum separation between Core and Atom that I managed to get various combinations calculations and accidents - twice as much! That is, Atom is still lagging behind.

But if I had stopped there, you simply would not have known about it - the post would not have taken place.
The next step was to compile the tests using Intel Compiler. The version used was Composer XE 2011 update 9 (12.1) with default Release optimization settings - similar to the Microsoft compiler.

The graph below shows the results of the above tests, including the rand I added, compiled by both VS2008 and Intel Compiler.

Look carefully. This is not an optical illusion. For four tests, the green line points showing the Atom result for tests compiled by Intel Compiler are higher than the burgundy line points showing the i5 result for tests compiled by VS2008. That is, Atom actually turns out to be more than twice as fast on the same code as Core i5.

Do you think this is an advertisement for an Intel compiler?
Absolutely not. I don't work in the advertising department or in the compilation group.
This is simply a statement that your optimized code can run much faster on Atom than unoptimized code on Core. Or - unoptimized on Core will be slower than optimized on Atom.
These are exactly the same bumps and crannies that prevent the car from accelerating.
You can draw your own conclusions.

Part 1: Background, Theory, Core, Power

Before Atom

Intel has long been paying close attention to the mobile consumer sector and releasing products aimed at it. At first, these were processors selected for low power consumption with all other parameters being equal (except that the frequencies were lower and the case was smaller). Then they began to produce CPUs specially modified for such applications. The story can begin with the i80386SL chip, which for the first time had SMM (System Management Mode), the dynamic core was replaced with a static one (i.e., to save energy, the frequency can drop to zero), and cache, memory and ISA and PI (Peripheral Interface) buses. All these changes tripled the number of transistors (from 275,000 for a regular 386SX/DX to 855,000), but the engineers felt that such a budget was justified. In addition, there were also versions of i386CX and i386EX without built-in peripherals with three power saving modes.

A lot of water has passed under the bridge, each subsequent CPU (except for server ones) was produced in both regular and mobile (sometimes also in built-in) versions, but all manipulations mainly consisted of adding energy-saving modes to the core and selecting chips capable of operating at reduced voltage at lower frequencies. Meanwhile, competition from architectures designed specifically for mobile devices intensified: the 1990s brought the PDA (starting with the Apple Newton MessagePad), and the 2000s brought communicators, Internet tablets (the half-forgotten acronym MID) and ultra-mobile PCs (UMPCs). ). On top of that, it turned out that the main tasks for the user of such devices have small computing needs, so almost any CPU released after 2000 already had the necessary power for mobile use, except, perhaps, modern games (for which then mobile consoles with 3D graphics appeared).

There is a need to create a special architecture for a compact mobile device, where the main thing is not speed, but energy efficiency. At Intel, this task was taken on by the Israeli branch of the company, which had previously created a very successful family of Pentium M mobile processors (Banias and Dothan cores). In these CPUs, energy-saving principles were put at the forefront from the very beginning of development, so dynamic shutdown of blocks depending on their load and smooth changes in voltage and frequency became the key to the economy of the series. The Pentium M looked especially bright against the background of the Pentium 4 released at the same time, which in comparison seemed like hot frying pans. Moreover, operating at the same frequency, Pentium M outperformed the “fours” in terms of performance, which was the first time that happened in the practice of processor development - usually a mobile computer pays for its compactness with all other characteristics. However, the Pentium 4 itself was, let’s say, not very good as a universal CPU...

The success of the platform showed that not everyone needs such high speed, but saving more energy would be nice. At that time (mid-2007), Intel released the “dad” of our today's heroes - the A100 and A110 processors (Stealey core). These are single-core 90-nanometer Pentium M with a quarter of the L2 cache (512 KB in total), greatly reduced frequencies (600 and 800 MHz) and consumption of 0.4–3 W. For comparison, standard Dothan at frequencies of 1400–2266 MHz have an energy consumption of 7.5–21 W, low-voltage (LV subseries) - 1400–1600 MHz and 7.5–10 W, and the first introduced ultra-low voltage (ULV) - 1000–1300 MHz and 3–5 W. It is reasonable to believe that modern computer spends most of the time waiting for the next keystroke or moving the mouse another pixel; the main difference between the A100/A110 and the ULV subseries Intel made is the ability to fall asleep very deeply, when there is no need to count at all, due to which consumption during idle drops by an order of magnitude. And a greatly reduced cache (a large L2 at such frequencies is not really needed) helped reduce the size of the crystal, which made it cheaper. The size of the processor case has been reduced by five times, and the total area of ​​the CPU and chipset has been reduced by three times. As we will see later, such techniques were used in the Atom series.

Despite the fundamentally correct goal setting, the A100/A110 remained in little demand on the market. Either 600–800 MHz was still not enough even for a simple Internet tablet, or there were only two chips (which even model range hard to name) from the very beginning they were an experimental product for testing the technology, or the processor was simply not promoted by marketers, knowing that it was being replaced by something much more advanced... Less than six months after the release of the A100/A110 on October 26, 2007 Intel announced the imminent release of new mobile CPUs codenamed Silverthorne and Diamondville and Bonnell core - the future Atoms. By the way, the name Bonnell comes from the name of a 240 m high hill in the vicinity of Austin (Texas), where a small group of Atom developers was located at the local Intel development center. “Whatever you name the yacht, that’s how it will sail.” ©Captain Vrungel

In 2004, this group, after the cancellation of the Tejas project led by it (the successor to the Pentium 4), received the exact opposite task - the Snocone project to develop an extremely low-power x86 core, dozens of which would combine into a super-efficient chip with a consumption of 100–150 W (the future Larrabee , recently relegated to “demonstration prototype” status). The group included several microelectronic architects from other companies, including “sworn friend” AMD, and its head Belli Kuttanna worked at Sun and Motorola. Engineers quickly discovered that the various options for available architectures did not suit their needs, and while they were thinking further, at the end of the year, Intel CEO Paul Otellini informed them that the same CPU would also be 1-2 core for mobile devices. Then it was hard to imagine exactly how and with what requirements such a processor would be used after the 3 years allotted for development - management, with a high degree of risk, pointed to handhelds and 0.5 W of power. History has shown that almost everything was predicted correctly.

Device CE4100

Interestingly, after the Atom, in the summer of 2008, the EP80579 (Tolapai) was released for embedded applications with a Pentium M core, 256 KB L2, 64-bit memory channel, full set peripheral controllers, frequencies 600–1200 MHz and consumption 11–21 W. And almost immediately after it - the Media Processor CE3100 (Canmore) model for the digital home and entertainment: Pentium M architecture, 800 MHz frequency, 256 KB L2, three 32-bit memory controller channels, 250 MHz RISC video processor and two 340 MHz DSP cores (digital signal processor) for audio. How these things were purchased is not clear, because after the announcement nothing was heard about them, including from Intel. Apparently, not very much... After the heyday of Atom, in September 2009, Intel tried again and released the CE4100, CE4130 and CE4150 (Sodaville) with an “atomic” core with a frequency of 1200 MHz, two 32-bit DDR3 channels, updated peripherals and technology 45 nm. Once again, little has been heard about these highly integrated systems-on-chips (SOCs) since then. Maybe the market is not ready to meet a hero?
Left CE4100, right CE3100

Atom Theory

First, let's look at the main characteristics of the processor from the consumer's point of view. There are three of them: speed, energy efficiency, price. (True, energy efficiency is not a very “consumer” characteristic, but, nevertheless, it is the easiest way to judge some important parameters final device.) Next, remember that in an ideal CMOS chip (all modern digital chips are manufactured using this technology), energy consumption is proportional to the frequency and the square of the supply voltage, and the peak frequency depends linearly on the voltage. As a result, by halving the frequency, we can halve the voltage, which in theory will reduce energy consumption by 8 times (in practice, by 4–5 times). Thus, mobile processor must be low frequency and low voltage. How then will he be fast? To do this, it needs to execute as many instructions as possible during each clock cycle, which most often means increasing the number of pipelines (degree of superscalarity) and/or the number of cores. But this leads to a sharp increase in the transistor budget, which increases the area of ​​the chip, and hence its cost.

Thus, it will not be possible to win on all three points, even theoretically (which explains the presence of such a variety of processor architectures on the market). Therefore, somewhere you will have to give up positions. A historical excursion says that it is necessary to pass it quickly, which will make it possible to make the CPU core as simple as possible. This is exactly the path that engineers from Austin took. After considering the options, they decided to return to the architecture of 15 years ago, first and last time (among Intel processors) used in the first Pentiums. Namely: the processor remains superscalar (i.e. we will have 2 instructions per clock cycle - but not 3-4, as in Atom’s contemporaries), loses the mechanism for shuffling instructions before execution (OoO), but acquires something that the Pentium did not have - hyperthreading technology (HyperThreading, HT), which allows, on the basis of one physical core, to emulate the presence of two logical ones for the OS and software. To explain why this choice was made, the reader is advised to first recall all the possible ways to increase CPU performance. Now let’s evaluate them from the perspective of energy consumption and transistor costs.

Using a multi-processor configuration in a pocket or laptop device is unacceptable, but multi-core is fine if the speed of one core is not enough. At first, Intel did this in the same way as in the first 2-core Pentium 4 - by placing a pair of identical 1-core chips on a common substrate and a common bus to the chipset. Of the other shared resources, there is only the supply voltage, which is selected from a maximum of two requests. That is, the cores can separately change their frequencies, but fall asleep and wake up synchronously. In December 2009, Intel released the first integrated versions of Atoms, where there are 1-2 cores and a northbridge on one chip. The board still has a south bridge connected to the CPU via the DMI bus, which is slightly faster and more economical than the previous combination. We won't be offered more than two cores soon, so the main speed focus is on their internals.

At this stage, Intel engineers were also not very concerned about the issue of increasing the frequency ceiling, although no one was going to abandon the principle of conveyoring and decoding x86 commands into internal micro-operations (mops) - this would have been too radical a step back. But transition predictors, data preloaders and other auxiliary systems for filling the pipeline have become very important, because an idle conveyor that cannot execute other commands bypassing the stuck one means precious watts wasted - and Atom has all the necessary “supports” made only slightly worse than the Pentium M and more modern Core 2, except that the buffer sizes are smaller (again for the sake of economy). Ultimately, the main battle plays out around performance per clock.