The I7 870 takes off skyward without accelerating. Motherboards

Almost a whole year has passed since Intel launched its new processors belonging to the Nehalem family on the market. However, we cannot say that over the past time these processors have been able to gain widespread recognition among users. Despite the fact that Nehalem can provide a truly higher level of performance than processors of the Core 2 generation, the demand for them has not been so great. Even according to statistics taken from the database of the CPU-Z utility, which is used primarily by enthusiasts who pay great attention to updating their own computers, the share of systems built on Core i7 processors is currently only a little more than 10 percent.

The reason for such sluggish interest in Core i7 is quite understandable: these processors are quite expensive and, moreover, use their own LGA1366 platform, which includes specialized motherboards and three-channel DDR3 memory, the construction of which turns out to be no less expensive than purchasing a processor. However, the low popularity of systems based on Nehalem family processors among individual users has not yet upset Intel much. The main goal of the company, which the new processor was supposed to hit, was to strengthen its position in the server market. And here, in the class of multiprocessor solutions, it really could become an excellent proposal. The new QPI interprocessor bus with a point-to-point topology plus an integrated memory controller in each processor have become the elements thanks to which the performance of server solutions based on processors of the Nehalem family has reached a fundamentally new level. Users of desktop systems cannot feel most of these advantages even if they want to, which is why the server processor for desktop systems represented by the Core i7 has not yet become a bestseller.

However, the previous generation of processors, Core 2, continued to successfully maintain its share in the desktop market, surpassing the competitor's offerings and providing ample performance for solving typical user tasks even after the release of Nehalem. Somehow, this situation has only begun to change now: AMD was finally able to master the 45-nm technological process, which gave it the opportunity to launch serial production of processors of the Phenom II family, capable (with certain reservations) of competing with Core 2 in terms of performance. Naturally, Intel could not accept this state of affairs, so the company chose the fall of this year to radically update its mid-level platform. This update is quite logical and therefore predictable: from now on, processors with the Nehalem microarchitecture will be used not only in systems in the upper price range, but also in mid-range computers. The high cost of the LGA1366 platform does not stop Intel at all, since for commonly used systems the company wants to offer a different LGA1156 processor socket, a different chipset and other motherboards. With minor changes in processor and system architecture, new hardware components should be more accessible to the general public. Actually, this article will be devoted to getting to know the platform consisting of such updated components.

Price reduction course

The first processors of the Nehalem generation, aimed at use as part of desktop computers and belonging to the Core i7-900 series, were built on chips codenamed Bloomfield. They combined four computing cores, a single third-level cache with a capacity of 8 MB, a DDR3 memory controller and a QPI bus controller. In this form, Core i7s were unified with server Xeons, but at the same time they caused certain inconveniences when used in desktop computers. For example, they required the use of three-channel DDR3 SDRAM, which is not only unusual for desktops, but also clearly redundant.

As a result, the platform built on Bloomfield processors looked like this:

For mass processors, Intel decided to develop new semiconductor crystals that do not differ from Bloomfield in key parameters, such as quad-core and the presence of an 8-MB shared third-level cache, but are more profitable in terms of the combination of characteristics and price. These “optimized” quad-core processors, also belonging to the Nehalem generation, were codenamed Lynnfied.

It should be noted right away that Lynnfield uses the same 45 nm process technology as Bloomfield processors. Although Intel plans to launch 32 nm technology this year, the company does not plan to test the new process technology on these processors. The first meeting with 32 nm semiconductors awaits us a little later in products known under the code name Clarkdale, which will make the Nehalem microarchitecture accessible to buyers with even more limited financial capabilities. So the differences between Lynnfield and Bloomfield do not seem so significant at first glance, but they are quite enough to reduce the total cost of the platform by almost half.

As you know, one of the main advantages of the Nehalem microarchitecture is the modular structure of the processor chip, which allows you to easily and without labor-intensive redesign change the set of functional blocks included in the processor. It was this opportunity that Intel engineers took advantage of when creating Lynnfield. In particular, first of all, it was decided to replace the three-channel DDR3 SDRAM controller with a more user-friendly two-channel one. Reducing the number of memory channels from three to two has a slight impact on system performance, but has a direct impact on the cost of the platform, making it possible to reduce the number of DDR3 modules in the system.

Another long-awaited simplification of the platform was also made. It is obvious that the QPI bus, which came to Bloomfield mainly because of its server roots, performs only a single function there - providing communication between the processor and the PCI Express bus controller, to which the graphics subsystem is connected. And from a functional point of view, absolutely nothing will change if in the presented structure we remove the X58 IOH chip and the QPI bus, replacing its controller in the processor with a PCI Express bus controller. This was also done. As a result, systems built on the basis of Lynnfield acquired the following, much simpler structure:

Instead of a high-speed QPI bus interface, Lynnfield now has a PCI Express 2.0 bus controller that supports 16 lanes and allows the operation of one or a pair of video cards using ATI CrossfireX and NVIDIA SLI technologies in 8x + 8x mode. In addition, a low-speed DMI bus has been added to the processor, which ensures interaction between the processor and the southbridge of the chipset.

As a result, the Lynnfield-based platform makes it possible not only to abandon three-channel memory in favor of two-channel memory, but also to do without the north bridge of the logic set altogether. This, naturally, makes it possible to simplify the design of motherboards. The result is quite predictable: in addition to the fact that Intel is going to sell Lynnfield processors cheaper than their older counterparts, buyers will be able to save on both memory and motherboards. As a result, it will now be more than possible to fit a platform that includes a Nehalem generation processor along with a motherboard and memory into a $400 budget.

At the same time, it looks quite funny that the Lynnfield semiconductor crystal turned out to be larger than that of the more expensive Bloomfiled: the implementation of the PCI Express controller required more transistors than were used for the QPI bus controller.

Bloomfield core

Lynnfield core

The characteristics of the semiconductor crystals of these two processors correlate as follows:

However, this crystal size ratio did not stop Intel from making Lynnfiled processors cheaper than their predecessors. Thus, official Lynnfield prices will range from 200 to 555 US dollars, while various models of Bloomfield processors are currently sold at prices from 285 to 1000 dollars.

Lynnfield range

Lynnfield processors are inferior to Bloomfield processors on the market not only in the number of memory channels of the built-in controller. There is also a difference in clock speeds: cheaper Lynnfiled will operate at a slightly lower frequency. However, these differences do not seem too serious, so it is logical that Lynnfield will be sold under the same brand as Bloomfield - Core i7.

However, the younger Lynnfield model will still be classified in the lower-class Core i5 family, but it will also differ from all other Nehalem desktops in the lack of support for Hyper-Threading technology. Accordingly, Intel's logic in assigning names to its new processors becomes clear. Those quad-core models that look like eight-core ones in the operating system thanks to Hyper-Threading support belong to the Core i7 series. If the processors are presented in the operating system as quad-core, then they will be called Core i5. In this regard, it is not difficult to assume that the Core i9 family will consist of promising six-core processors with support for Hyper-Threading technology, known under the code name Gulftown. As for Core i3 processors, a different logic will work here: this family will include budget processors with reduced characteristics.

Lynnfield's lineup will initially include three products at 2.66, 2.8 and 2.93 GHz. It is important that, due to lower frequencies, these processors will have a reduced calculated typical heat dissipation, amounting to 95 W rather than 130 W, as with Bloomfiled. This will allow the new models to be considered as a full-fledged replacement for the Core 2 Quad, not only in terms of performance, but also in terms of power consumption.

The entire list of current processors with the Nehalem microarchitecture now consists of six products:

Considering that from a microarchitectural point of view, Lynnfield is very similar to Bloomfield and, in fact, differs only in the Uncore part, the coincidence of the main characteristics between the Core i7-900 and Core i7-800 processors, as well as the Core i5-750, seems not at all surprising.

Among the characteristic features of the structure of the new Lynnfield processors, which in many ways make them similar to the Core i7-900 models present on the market, it should be noted:

Congenital quadruple structure. A single processor chip includes four cores with a 256 KB L2 cache and a common shared L3 cache.
Elimination of the processor bus in its traditional sense by moving the PCI Express 2.0 controller directly into the processor. The controller built into the processor provides 16 PCI Express 2.0 lanes, to which either one (in PCI Express x16 mode) or two (in PCI Express x8 + PCI Express x8 mode) video cards can be connected.
A memory controller built into the processor that supports dual-channel DDR3 SDRAM. Moreover, each channel is capable of working with three unbuffered DIMMs.
Support for Hyper-Threading technology (only for older Lynnfield models belonging to the Core i7-800 series). Thanks to it, each core of the Core i7-800 can execute two computational threads simultaneously, as a result of which the processor appears in the operating system as eight cores.
Shared third level cache with a total capacity of 8 MB.
Built-in PCU microcontroller that independently controls the voltage and frequency of each core, with the ability to automatically overclock individual cores with reduced load on other cores.
Support for the new SSE4.2 instruction set.
Core i7-800 and Core i5-700 are manufactured using 45 nm technology, consist of 774 million transistors and have a core area of 296 sq. mm.

The close relationship between the Core i7-900 and the new processors of lower series is easy to notice from the readings of the CPU-Z diagnostic utility. In particular, for the Core i7-870 and Core i5-750 processors that arrived in our laboratory, they were as follows.

In general, everything looks exactly the same as our old friends - Core i7-900. The only thing that is somewhat confusing in the above screenshots is the display of the QPI processor bus frequency on them. Obviously, this is a program error, since Lynnfield processors simply do not have such a bus. The processor clock frequency is formed as the product of a multiplier and the frequency of the base clock generator, which traditionally for all Nehalem is 133 MHz.

Judging by the formal characteristics, the most noticeable difference between the new processors and the Core i7-900 is the new socket, LGA1156. As the name suggests, this connector has fewer pins than the usual LGA1366, which, in general, is not surprising and is easily explained by the reduction in the number of memory channels in the built-in controller and the replacement of QPI interfaces with regular PCI Express.

Reducing the number of contacts, together with some miniaturization of the contact pads on the processor, made it possible to reduce the physical dimensions of both the processor itself and the processor socket to sizes approximately corresponding to the dimensions of LGA775.

However, when looking at the new product from the back, it becomes clear that the LGA1156 and LGA775 processors are radically different. Although the dimensions are similar, the number of contacts on the belly of the Lynnfield is significantly higher.

Left - LGA775 processor, middle - LGA1156, right - LGA1366

Thus, the new Core i7-800 and Core i5-700 processors are not compatible with any older platforms and require the use of their own LGA1156 motherboards. Moreover, new processors also require their own cooling systems. According to regulatory documentation, mounting holes must be located on LGA1156 motherboards at a smaller distance from each other than on LGA1366 systems, but further than on LGA775 boards. To be honest, given the equality of typical heat dissipation between the older LG775 and LGA1156 processors, such differentiation of cooling systems is somewhat puzzling, but the fact remains: the Core i7-800 and Core i5-700 require their own coolers.

At the same time, the boxed cooler that we received along with the Core i7-870 processor deserves some attention. Despite the fact that this processor is the eldest in the Lynnfield family, the cooler for it is offered in fairly small sizes. The height of its radiator made of aluminum with a copper core is only 13 mm.

This gives a tangible idea that Lynnfield processors are not as hot-tempered as, for example, Bloomfield.

New chipset: Intel P55

The release of Lynnfield processors entailed significant changes in the structure of platforms. Designed for use exclusively in single-processor systems, these processors do not have the QPI interface, which is used in LGA1366 platforms to communicate between the processor and the chipset. Therefore, for the Core i7-800 and Core i5-700 processors, their own set of logic was developed, called Intel P55 Express.

The main feature of this chipset is its extreme simplicity. The advent of the Nehalem microarchitecture made it possible to remove the memory controller from the system logic set, but now it has come to the PCI Express controller. Since the processor itself is now responsible for working with this bus, it turned out that it “pulled over” all the functions traditionally performed by the northbridge of the chipset. Accordingly, the need for this chip was no longer needed, and the Intel P55 became Intel's first chipset consisting of a single chip - the Platform Controller Hub (PCH).

The connection between the processor and the chipset in LGA1156 systems is performed via the Digital Media Interface (DMI) bus with a bandwidth of 10 Gbit per second in each direction, which was previously used to connect the north and south bridges of logic sets. Therefore, there are no obstacles to using not only the new P55 chipset, but also the old ICH10 south bridge in conjunction with Lynnfield processors in LGA1156 systems.

However, we cannot say that the P55 PCH is fundamentally different from the ICH10 in its capabilities. In fact, we are talking exclusively about updating existing interfaces and slightly expanding their number. The following table can give a good idea of this.

The implementation of 8 PCI Express 2.0 lanes in the P55 is necessary to provide the ability to connect additional devices in addition to graphics cards. The processor in LGA1156 systems is only capable of providing a connection to PCI Express graphics cards, while the P55 PCH is responsible for working with all other devices. Actually, approximately the same scheme was implemented in the X58, where the role of the north bridge was assigned only to interaction with graphic cards. An important improvement of the P55 was that it introduced support for the PCI Express 2.0 bus. This means that devices supporting this specification will be able to communicate with the chipset twice as fast as before.

In addition, changes have been made to the USB controller. It did not support new and faster versions of the protocol, but it began to provide a larger number of ports, and also received the function of hardware disabling individual ports, which can be useful for security purposes.

For the announcement of LGA1156 processors, motherboard manufacturers have prepared a large number of products based on the Intel P55 logic set. As an example of such a board, we would like to give a photograph of Intel's own design, DP55KG, and note some characteristic details on it.

Actually, in addition to the presence of only one chipset chip on the board, what is also striking is that it has a rather “weak” heatsink installed on it, which goes against the ornate designs that manufacturers of boards for enthusiasts usually like to install on their boards. However, in this case, Intel did not save at all; the P55 PCH chip really does not need any kind of extraordinary cooling system. Although it is manufactured using a 65 nm process, its typical heat dissipation is only 4.7 W. For example: the typical heat dissipation of the northbridge of the X58 chipset is 24.1 W. Therefore, those motherboard manufacturers who again prefer to install massive chipset heatsinks on their products are, in fact, engaged in a senseless increase in the cost of their products and misleading users.

The second feature of LGA1156 boards is the simplified design of the processor socket. Compared to LGA1366, the processor mount is screwed to the board with three rather than four screws, and the closing mechanism now snaps onto one of the screws and does not include a metal frame framing the processor socket. However, this can hardly be considered any fundamental change: the point, most likely, is a banal saving of metal.

Also, the design of the second PCI Express graphics slot deserves some attention. Considering that when installing a pair of video cards, it will only be able to work in 8x mode, Intel “cut out” the second half from it, which, under any set of circumstances, will only be able to perform a cosmetic role.

First disappointment: Lynnfield memory controller

We decided to add a separate section to the article dedicated to the memory controller built into Lynnfield because it differs from the Bloomfield memory controller not only in the number of channels. The fact is that in the new LGA1156 the entire processor Uncore unit was changed, namely the circuit for generating bus frequencies and L3 cache.

Let us recall that Core i7 processors in the LGA1366 version use one 133-MHz base clock generator (BCLK) and several independent multipliers to generate frequencies that generate the frequencies of the computing cores, QPI bus, L3 cache and memory controller, as well as the operating frequency of DDR3 SDRAM. In the new Core i7, designed for LGA1156 systems, the fundamental design remains exactly the same, but the number of multipliers available for change has been reduced, for example, due to the elimination of the QPI bus, the independent multiplier for which has been eliminated as unnecessary.

As a result, Core i7-800 and Core i5-700 processors use only three multipliers:

1. CPU frequency multiplier. This multiplier is determined by the standard processor frequency and cannot be increased by the user above the nominal value. However, during operation it can change automatically, thanks to the Turbo Mode and Enhanced Intel SpeedStep technologies.
2. Memory frequency multiplier. Formally, Lynnfield processors support DDR3 SDRAM with frequencies up to 1333 MHz. However, the set of available multipliers for Core i7-800 processors includes 8x, 10x and 12x, which implies the possibility of using DDR3-1067, DDR3-1333 and DDR3-1600 memory in LGA1156 systems. Processors of the Core i5-700 series, unfortunately, do not support a 12x multiplier for the memory frequency, so for them the maximum memory frequency is 1333 MHz.
3. Uncore multiplier, which sets the operating frequency of the memory controller and L3 cache. For all LGA1156 processors this multiplier is fixed. Moreover, for the Core i7-800 it is 18x, and for the Core i5-700 it is 16x. Accordingly, in Core i7-800 Uncore processors the part operates at a frequency of 2.4 GHz, and in the Core i5-700 - at 2.13 GHz. Let us recall that in models for LGA1366 systems this frequency could vary at the user’s request, and by default it was twice the memory frequency.

Thus, the memory subsystem in Lynnfield turns out to be slower than that of Bloomield processors and this is explained not only by a decrease in the number of channels, but also by a lower frequency of the memory controller and L3 cache.

All this, naturally, affects the practical bandwidth and latency of the memory subsystem. If in LGA1366 systems we saw that reducing the number of active memory channels from three to two does not entail any noticeable drop in performance, then in LGA1156 systems the memory still works slower, although not by much.

This is clearly visible from the results of synthetic tests of the memory subsystem. For example, for testing, we decided to compare the practical speed of the memory subsystem LGA1366 and LGA1156 processors Bloomfield and Lynnfield, operating at the same clock frequency of 2.93 GHz. Both systems used DDR3-1333 SDRAM with the same timings of 7-7-7-18.

For example, such numbers can be seen in the Cachemem test from the popular diagnostic utility Everest.

While the operating speed of L3 cache memory between processors turns out to be not very different, the same cannot be said about memory speed. The three-channel Bloomfield controller shows slightly higher performance on all memory operations than the two-channel Lynnfield memory controller. The only consolation here is that the memory subsystem of the new LGA1156 platform demonstrates slightly lower latency.

In general, these results are confirmed when using another utility that measures the practical parameters of the memory subsystem, MaxMem.

Bloomfield 2.93 GHz, three memory channels

Lynnfield 2.93 GHz, two memory channels

However, in this case, along with a slight drop in practical throughput in LGA1156 systems, there is also a slight increase in latency.

Thus, the new LGA1156 platform cannot be considered as a complete replacement for LGA1366. Of course, older Lynnfiled processors will be able to compete with younger Bloomfield models, but the fastest Core i7-900 in any case will remain the owner of unsurpassed performance. And this will be ensured not only by high clock frequencies and the ability to form CrossfireX and SLI systems according to the 16x + 16x scheme, but also by a slightly higher speed of the memory subsystem.

Lynnfield's Secret Weapon: Turbo Mode

One of the interesting innovations introduced in the Nehalem family of processors was a specialized Power Control Unit (PCU), which has the ability to monitor and manage the power consumption of individual cores. Thanks to the PCU, Core i7 processors now have Turbo Mode technology, through which dynamic and automatic overclocking of the processor is implemented. Let us recall that the essence of this technology comes down to the fact that when the processor’s capacity remains underutilized and its power consumption is far from the maximum values, the processor raises its own multiplier above the standard value. This technology is especially useful in cases where the load on the processor is not clearly multi-threaded.

Processors belonging to the Core i7-900 family were able to increase their multiplier by one, and with only one core actively working, by two. As a result, the frequency of such processors was often 133 or 266 MHz higher than the nominal value. Of course, such a frequency increase does not seem so significant, but even thanks to it, the weighted average performance of LGA1366 platforms with Turbo Mode activated was 3-5% higher than the performance of similar systems that do not use this technology. In other words, Turbo Mode technology was successfully tested on the Core i7-900 and has proven itself to be the best.

That is why Lynnfield processors have further developed this technology. The principles of its operation remain the same, but new processors have the ability to control their own clock frequency much more aggressively. In new processors, the multiplier can be increased by up to 5x, due to which, under favorable conditions, the clock speed of the Core i7-800 and Core i5-700 can increase by 667 MHz above the nominal value, and this is no longer a joke. However, it should be understood that the real value of such an increase is determined based on the current workload of the processor and its energy consumption. Thus, increasing the multiplier by 5x is possible only if a single processor core is loaded, but if two or three of the four cores are under load, then the multiplier can only be increased by 4x. But even if all four cores are busy, the multiplication factor can increase by 2x.

A more detailed idea of Lynnfield frequencies when operating in Turbo Mode can be obtained from the table:

One very important conclusion follows from this table. In situations where the load on the processor is not multi-threaded or weakly multi-threaded, the Core i7-800 and Core i5-700 processors may be faster than their predecessors from the older Core i7-900 series, since they can self-overclock much more strongly.

Moreover, in this case we are not talking about some ephemeral increase in frequency. In fact, most motherboards can put the processor into Turbo Mode permanently, so that the clock speed increases to its maximum value, regardless of the current power consumption of the processor. Therefore, when activating Turbo Mode, users will most likely be able to observe this situation (for illustration, we provide screenshots taken on a system with a Core i7-870 processor with a nominal frequency of 2.93 GHz):

The processor is in an idle state. Thanks to energy-saving Enhanced Intel SpeedStep technology, its frequency is reduced to a minimum value of 1.2 GHz.

A single-threaded load allows you to raise the frequency to a maximum of 3.6 GHz, which is 667 MHz higher than the nominal value.

With a dual-threaded load, the processor frequency is increased by 533 MHz to 3.46 GHz.

A quad-threaded load also allows the processor to continue running smoothly at 3.46 GHz.

But when the processor is fully loaded with eight computing threads, its frequency is set to 3.2 GHz, which, nevertheless, is higher than the nominal value by 266 MHz.

Not only the screenshots with frequency readings look impressive. To evaluate the real effect of Turbo Mode, we compared the performance of a system with a Core i7-870 processor with and without Turbo Mode enabled.

On average, the positive effect from the inclusion of updated technology in the second version is about 8%. In those applications that are not well optimized for systems with multi-core processors, this increase can reach even more significant values. As a result, Turbo Mode looks like an excellent trump card for the LGA1156 platform, increasing its attractiveness in the eyes of the user. Indeed, thanks to this technology, Intel partly solves the problems of software manufacturers who have not yet bothered to adapt the algorithms used to the concept of multithreading. The new Core i7-800 and Core i5-700 processors, which are essentially true quad-core models, can be converted into fast pseudo-dual-core or pseudo-single-core processors if possible. Moreover, what is especially nice is that this transformation occurs completely transparent to the system and does not require any intervention from the user.

How we tested: switching to Windows 7

The main heroes of today's performance tests are the LGA1156 processors Core i7-870 and Core i5-750, which are, respectively, the senior and junior representatives in the Lynnfield family. As competitors for these models, we chose processors with a similar price, designed for all other current platforms: LGA1366, LGA775 and Socket AM3.

As a result, the following hardware and software components took part in testing:

Processors:

AMD Phenom II X4 965 (Deneb, 3.4 GHz, 4 x 512 KB L2, 6 MB L3);
AMD Phenom II X4 955 (Deneb, 3.2 GHz, 4 x 512 KB L2, 6 MB L3);
Intel Core 2 Quad Q9650 (Yorkfield, 3.0 GHz, 1333 MHz FSB, 2 x 6 MB L2);
Intel Core 2 Quad Q9550 (Yorkfield, 2.83 GHz, 1333 MHz FSB, 2 x 6 MB L2);
Intel Core 2 Quad Q9400 (Yorkfield, 2.66 GHz, 1333 MHz FSB, 2 x 3 MB L2);
Intel Core i7-950 (Bloomfield, 3.06 GHz, 4.8 GHz QPI, 4 x 256 KB L2, 8 MB L3);
Intel Core i7-920 (Bloomfield, 2.66 GHz, 4.8 GHz QPI, 4 x 256 KB L2, 8 MB L3);
Intel Core i7-870 (Lynnfield, 2.93 GHz, 4 x 256 KB L2, 8 MB L3);
Intel Core i5-750 (Lynnfield, 2.66 GHz, 4 x 256 KB L2, 8 MB L3).

Motherboards:

ASUS P5Q3 (LGA775, Intel P45, DDR3 SDRAM);
Gigabyte GA-EX58-UD5 (LGA1366, Intel X58 Express);
Gigabyte GA-P55-UD6 (LGA1156, Intel P55 Express);
Gigabyte MA790FXT-UD5P (Socket AM3, AMD 790FX + SB750, DDR3 SDRAM).

Memory:

2 x 2 GB, DDR3-1333 SDRAM, 7-7-7-20 (Mushkin 996601);
3 x 2 GB, DDR3-1333 SDRAM, 7-7-7-20 (Mushkin 998679).

Graphics card: Sapphire Radeon HD 4890;
Hard drive: Western Digital VelociRaptor WD3000HLFS;
Power supply: Tagan TG880-U33II (880 W);
Operating system: Microsoft Windows 7 Ultimate x64;
Drivers:

Intel Chipset Software Installation Utility 9.1.1.1015;
ATI Catalyst 9.8 Display Driver.

Separately, it should be noted that in our testing we are switching to using the new Windows 7 operating system, which, although not officially announced yet, already exists in the form of the final RTM version. In our case, when it comes to testing processors of the Core i7 family, this is of particular importance. The fact is that this operating system has special optimizations that increase the performance of platforms with processors that support Hyper-Threading technology. In close collaboration, Intel and Microsoft engineers implemented SMT parking technology, which optimizes Windows 7 to run on processors with virtual cores. This is expressed in the fact that in most cases where under Windows Vista and XP Hyper-Threading technology could cause slowdown of applications, under Windows 7 this should not happen, since the scheduler of this operating system distinguishes between physical and virtual cores and prevents cases when Executing two threads on the same core can lead to performance degradation.

Performance

Overall Performance

Apparently, all our worries about the insufficiently fast memory controller in Lynnfiled processors were completely in vain. In terms of real-world performance, the Core i7-870 and Core i5-750 turn out to be brilliant products. In absolutely all scenarios, even the youngest of the new products, which does not support Hyper-Threading technology, turns out to be faster than processors of the Core 2 Quad family, and faster than the competing Phenom II.

As for the performance ratio of Lynnfield and Bloomfield processors, the picture is not so clear-cut, but, nevertheless, the Core i7-870 and Core i5-750 look quite decent, providing high speed in E-Learning and 3D scenarios, showing a clear lag from the “older” Intel platform only when processing video content.

Gaming Performance

LGA1156 also proves to be an excellent gaming platform. After we switched to using the Windows 7 operating system, all the previous problems with the speed of the Core i7 in games went away on their own. As a result, even a $200 Core i5-750 turns out to be head and shoulders faster than the Phenom II X4 and Core 2 Quad processors. At the same time, the Core i7-870 quite successfully storms the position occupied by the representative of the LGA1366 platform, the Core i7-950 processor.

Audio and video transcoding performance

The situation with the operating speed of popular codecs looks somewhat more interesting. First of all, you should pay attention to how fast LGA1156 processors are when transcoding audio files in iTunes. This is a clear demonstration of the benefits of the second version of Turbo Mode implemented in Lynnfield; After all, iTunes is one of those applications that creates only a two-threaded load.

DivX, as we know, was well optimized for quad-core processors, but Hyper-Threading technology is of no use to this codec. This conclusion can be made based on the fact that the Core i7-870 and Core i5-750 processors show fairly close results. However, even despite the fact that one of the most important technologies of Nehalem processors is “out of business,” their computing power turns out to be enough to outstrip all processors with other architectures. There is not much difference in the performance of the LGA1366 and LGA1156 Nehalem variants in this test.

But when using the x264 codec, Hyper-Threading allows you to get a fairly serious increase in performance. As a result, all Core i7s that support this technology are at the top of the chart by a wide margin. At the same time, the speed of the new Core i7-870 is only slightly short of the performance of the Core i7-950. As for the Core i5-750, which does not support Hyper-Threading, its lag behind its peers looks quite serious. But, nevertheless, even despite this, it is ahead of almost all other quad-core processors that do not belong to the Core i7 class, with the exception of the Phenom II X4 965.

Performance in video editors

Nonlinear video editing turns out to be a task that heavily loads the power of all processor cores and requires high throughput of the memory subsystem. As a result, Turbo Mode technology turns out to be powerless, which entails a fairly noticeable lag between Lynnfield and Bloomfield. Nevertheless, compared to other processors, the LGA1156 n platform still looks very attractive.

Performance in graphics editors

When editing and processing images in the freely distributed graphics editor Paint.Net, the Core i5-750 processor performs approximately on par with older Core 2 Quad processors, and the Core i7-870 is slightly faster than the Core i7-920. In Photoshop CS4, LGA1156 processors outperform their predecessors thanks to the successful implementation of Turbo Mode technology. Accordingly, their gap from the rest of the testing participants turns out to be more than impressive.

Rendering Performance

Final rendering is perhaps one of the most efficiently parallelized tasks. That is why the best results here are shown by quad-core processors that have Hyper-Threadng technology in their arsenal. In particular, the Core i7-870 is only slightly inferior to the Core i7-950, but at the same time it outperforms the older Core 2 Quad and Phenom II X4 processors by a very impressive 30-40%. As for the Core i5-750 processor, which does not support Hyper-Threading, it can boast of performance only at the level of older Socket AM3 and LGA775 processors.

Archiving and math calculations

Thanks to the high-speed integrated memory controller, Core i7 and Core i5 processors are clearly better at archiving than competing products. Even the youngest representative of this family, the Core i5-750, significantly outperforms both the Core 2 Quad series and Phenom II X4 processors, which, by the way, are also equipped with an integrated memory controller.

Testing in the Mathematica package does not bring any surprises either. In it, as in many other applications, new products are ahead of all competitors, with the exception of representatives of the LGA1366 platform, with which there is some parity.

In the Folding@Home distributed computing project, as in other similar computing tasks, Hyper-Threading technology solves a lot. The Core i7-870, which supports this technology, shows results close to those of the Core i7-950. The performance of the Core i5-750, which does not support Hyper-Threading, is at the level of performance of older models in the Core 2 Quad and Phenom II X4 series.

Single-threaded performance

To conclude the performance study of Lynnfield processors, we decided to see how they would look under a single-threaded load, when Turbo Mode technology was able to overclock them to the maximum. As tests, we chose the good old SuperPi, which calculates 8 million digits of the number π, and a single-threaded version of the final rendering test Cinebench R10.

The performance of the Core i7-870 and Core i5-750 processors in such somewhat artificial conditions can only be called triumphant. Thanks to Turbo Mode, their frequencies rise to 3.6 and 3.33 GHz, respectively, which allows them to be quite ahead of even more expensive processors for the LGA1366 platform. However, the remaining quad-core models, which, based on their cost, should be direct competitors to Lynnfield, are hopelessly behind even the Core i5-750, not to mention the older LGA1156 product.

In other words, although the LGA1156 platform is currently equipped with only three models of quad-core processors, you can switch to it without any fear, even if you do not use multi-threaded applications at all. Thanks to a very aggressive turbo mode, the speed of the Core i7-800 and Core i5-700 processors will be very impressive not only with multi-threaded workloads, but also with single-threaded or dual-threaded tasks. And for this we can thank Intel very much, because the second version of Turbo Mode implemented in Lynnfield clearly solves the problem of correctly choosing the number of processor cores for any model of system use.

Overclocking

The semiconductor core of Lynnfield processors differs little from the Bloomfield core. Indeed, the production of these processors uses the same technical process, and the main components of these processors are the same. Therefore, it would be strange if the newly announced LGA1156 processors were very different in their overclocking potential from the LGA1366 models. However, to test this hypothesis, we conducted experiments on a freelance increase in the frequency of the Core i7-870 and Core i5-750 processors available in our laboratory.

The tests were carried out on a platform based on the Gigabyte GA-P55-UD6 motherboard. For cooling, in all cases a Thermalright MUX-120 cooler (with a traditionally curved base) with an Enermax Magma UCMA12 fan (1500 rpm) was used. System stability under load was checked using the LinX 0.6.3 utility.

Overclocking processors in the LGA1156 version can be done only in one way - by increasing the frequency of the BCLK clock generator. Of course, along with the frequency of the processor cores, the Uncore frequency will also increase, but nothing can be done about this: Lynnfield processors have not only a fixed processor multiplier, but also a locked multiplication factor for the Uncore frequency. Also, as the BCLK frequency increases, the memory frequency also increases, but fortunately for it, the corresponding multipliers can be lowered.

Trying to overclock the Core i7-870 processor, we were able to achieve stable operation at 4.07 GHz.

To achieve this result, the processor voltage was increased to 1.4 V, which can be considered a relatively safe level for Lynnfield, provided there is sufficient cooling. However, in our case, the temperatures of the processor cores reached 93 degrees. And although this is a fairly high temperature, the processor worked absolutely stably and did not overheat. Thus, processors belonging to the Core i7-800 series are quite capable of operating at frequencies of about 4 GHz when using air cooling, like their older brothers in the Core i7-900 series.

The second part of the experiments was carried out with a Core i5-750 processor. This processor does not support Hyper-Threading technology, which is the reason for its lower temperature during full load. I would like to hope that thanks to this feature, overclocking the Core i5 will become a more successful event. However, on the other hand, it is complicated by the fact that the Core i5-750 processor has a lower multiplier, which requires increasing the BCLK frequency when overclocking to higher values. But, fortunately, the upper limit of the BCLK frequency observed in typical LGA1366 platforms, which was 210-215 MHz, can be easily exceeded in the case of LGA1156 platforms.

However, in the end we did not need to increase the BCLK frequency above 210 MHz. Our Core i5-750 instance was able to work stably only at a frequency of 4.1 GHz, to achieve which it is enough to increase the BCLK frequency only to 205 MHz.

The core supply voltage was increased to 1.4 V, but the temperature of such an overclocked processor did not exceed 82 degrees. It turns out that, despite the rather small difference in the overclocking results of the Core i7-870 and Core i5-750, the temperature of a processor that does not support Hyper-Threading technology under full load is indeed significantly lower. This means that when overclocking experiments with the Core i5-750 you can use relatively inexpensive cooling systems.

It should be noted that we overclocked by disabling the Turbo Mode dynamic multiplier technology. However, the possibility of overclocking with activated Turbo Mode technology should be considered quite interesting. After all, it is quite possible that in the case of low computing load on the processor, its frequency can be increased more than we managed in today's tests. Therefore, in the near future we will prepare a new material in which we will consider various aspects of overclocking LGA1156 processors in more detail.

Energy consumption measurement

One of the most intriguing characteristics of LGA1156 processors is their estimated typical heat dissipation of 95 W. This is 35 W less than the calculated heat dissipation of almost the same LGA1366 processors. If we add to this the significantly reduced heat dissipation of the logic set in LGA1156 motherboards, then we can expect that the new Intel platform may turn out to be a very profitable solution in terms of performance and power consumption. And it is likely that this platform will even be able to compete with LGA775 systems, which have long been our sincere admiration when it comes to energy efficiency.

However, in such matters it is quite reckless to take the manufacturer’s promises on faith. Therefore, we tested the real energy characteristics of the systems participating in the tests. The following figures represent the total power consumption of the test platforms assembled (without a monitor) “from a wall outlet”. During measurements, the load on the processors was created by the 64-bit version of the LinX 0.6.3 utility. In addition, to correctly assess idle power consumption, we activated all available energy-saving technologies: C1E, Cool"n"Quiet 3.0 and Enhanced Intel SpeedStep.

At rest, the power consumption of the LGA1156 platform looks more than excellent. Among all the systems participating in the tests, it is the platforms based on Lynnfield processors that provide the lowest power consumption. This is partly explained by the fact that the energy-saving technologies in these processors have once again been improved. In particular, at rest, the Core i7-870 and Core i5-750 have the ability to reset their frequency to 1.2 GHz and voltage to 0.85 V.

The situation also looks quite good when measuring power consumption under load. Systems with LGA1156 processors have become definitely more economical than LGA1366 platforms; their power consumption has decreased by more than 50 W. At the same time, a system with a Core i5-750 processor that does not support Hyper-Threading technology turns out to be even more economical than an LGA775 system. That is, our expectations were fully justified: computers with LGA1156 processors are quite capable of offering a better performance-per-watt ratio than all other options. Moreover, Intel also has special energy-efficient processors in the LGA1156 version, so this platform has every chance of becoming a favorite among those users who care about saving energy over time.

To obtain a more complete and comprehensive picture, a separate study of the power consumption of the tested processors and motherboards under load, in isolation from other computer components, was carried out. More precisely, the consumption was measured on the 12-volt power line, connected directly to the processor voltage converter on the motherboard, and on the power lines of the motherboard.

Surprisingly, the lowest consumption among today's test participants was demonstrated by the Core i5-750 processor, which turned out to be even more economical than the Core 2 Quad Q9400. The consumption of the Core i7-870 is slightly higher, but, nevertheless, this processor is quite capable of competing in its electrical characteristics with models of the Core 2 Quad family. However, the low consumption figures demonstrated by processors with the Nehalem microarchitecture are partly explained by the design of their power circuit. The fact is that only processor cores are connected to a dedicated 12-volt power line. The Uncore part of the processor is powered from the motherboard via a 24-pin ATX connector. That is why this time we additionally provide motherboard consumption figures.

Considering the above, the fact that LGA1366 and LGA1156 motherboards turn out to be the most power-hungry in electrical terms is not at all surprising. After all, the measured consumption values include a considerable “additive” introduced by the Uncore part of the processor. However, one cannot help but notice that the elimination of the northbridge in LGA1156 systems actually made the corresponding motherboards much more economical than boards for LGA1366 processors. The difference in the consumption of boards for Bloomfield and Lynnfield reaches a total of 20 W.

Conclusions

Overall, the new LGA1156 platform makes a very good impression. And although it is obvious that Intel’s main goal, which the company wants to achieve with the release of Lynnfield processors, is to bring the Nehalem microarchitecture into the mid-price segment, during testing we often had the feeling that we were not dealing with a cheaper version of the LGA1366 platform, but with its updated and improved version.

And this feeling cannot be called groundless. The new LGA1156 platform does have a number of advantages. First of all, it is easier to understand: it uses the usual dual-channel memory, and the Lynnfield processors themselves look like real desktop processors, and not server processors, shoved into desktop systems for marketing reasons. Secondly, systems with Core i7-800 and Core i5-700 processors compare favorably with their predecessors in terms of energy consumption. At the cost of some restructuring of the platform structure, Intel engineers managed to achieve that platforms built on Lynnfield processors have power consumption comparable to the consumption of LGA775 systems, which until now have served as the standard for a successful combination of efficiency and good performance. Thirdly, a huge trump card of the new products is the Turbo Mode technology, which allows Lynnfield processors to work efficiently even when the load created is not clearly multi-threaded.

However, if we take into account not subjective, but objective factors, the LGA1156 platform will still not be able to change the current state of affairs. Despite all its obvious advantages, LGA1366 processors and motherboards will remain in high demand in the upper price segment. After all, only they allow you to build multi-GPU systems using two graphics cards operating in PCI Express x16 + x16 mode, or even more than two graphics cards. Only LGA1366 platforms will be compatible with future Gulftown six-core processors. And only in the Core i7-900 series of processors are products belonging to the Extreme Edition class available, which have not only unsurpassed performance, but also additional overclocking capabilities.

However, the Core i7-800 and Core i5-700 series processors look like excellent replacements for the LGA775 processors in the Core 2 Quad family, offering much better performance for about the same price. The emergence of the LGA1156 platform means a real revolution in the mid-price segment. This platform automatically relegates Core 2 and Phenom II processors to the list of obsolete solutions that can only be relevant as offerings sold in the “under two hundred dollars” price segment.

In other words, from now on we can talk about the real advent of the Nehalem era. Processors built on this microarchitecture have become not only affordable, they have matured and acquired additional attractiveness. As a result, without any doubt, new products in the Core i7-800 and Core i5-700 series will be able to quickly gain wide popularity and become bestsellers this season.

Check the availability and cost of LGA 1156 processors

Check the availability and cost of LGA 1156 motherboards

Check availability and cost of LGA1156 coolers

What is this chip capable of? Its market segment

The technical specifications for the Cor i7 870 continue to remain relevant and can easily solve any problems, including the most complex ones. At the same time, there is no particular need to increase its clock frequency (that is, to “overclock” this silicon crystal). Its four computing modules initially operate at a very respectable clock frequency, and the fast volatile memory subsystem is practically no different in its organization from current processors. As a result, we get one of the most productive chips of 2009, which, 7 years after the start of sales, continues to be relevant.

CPU equipment

The 870 was sold in two different trim levels. The more simplified and accessible of them was designated as “Trail”. In this case, the user received instructions for use along with a warranty card, a branded sticker of the family of semiconductor crystals, and the chip itself. The second configuration option was called “Box”, and it was a boxed version of the central processing unit. It included, in addition to the previously mentioned components for the previous version of the package, also a standard cooler, a branded box and thermal paste.

The first configuration option was better suited for computer enthusiasts and overclockers who could afford to buy an advanced cooling system and thereby increase the final performance of the computer system. But the box version was better suited for ordinary users, for whom the basic performance of this solution was quite enough.

Socket

The Intel Core i7 870 was designed for installation in Intel Corporation's most advanced 2009 socket for entry-level computer solutions - LGA1156. Moreover, it was one of the most productive processors on this platform. The main set of logic in this case consisted of only one south bridge - P55. A was transferred to the chip body. A little later, other chipsets for this platform appeared - H57 Express and H55 Express. Essentially, it was the same P55, but supplemented with a graphics subsystem. The most optimal solution in conjunction with this CPU was to use the P55, the capabilities of which were quite sufficient to unleash this high-performance semiconductor solution.

Process

The Core i7 870 central processing unit was manufactured according to the technological process standards (45 nm), which was advanced in 2009. 7 years have passed since then, and the technological process (14 nm) is already used in the production of advanced computing solutions. Despite such a big difference, this high-performance chip continues to be relevant and can solve any problem even today without any problems.

Cache

The high-performance Core i7 870 CPU had to be equipped with a three-level cache. Otherwise, he would not have been able to solve problems of increased complexity even in 2009. At the same time, the memory capacity should also be of good size.

As a result, the development engineers equipped it like this:

The first level for each core was represented by 32 kb modules divided into 2 parts. One of them specialized in storing CPU instructions, and the second specialized in storing data. That is, one computing module could use 64 kB in total. Well, since the total number of cores was 4, we get 4*64 = 256 kb.

The second level no longer had such strict specialization regarding the type of stored information, but was divided between individual cores. The capacity for each computing unit was 256kb. Well, in total its total size was 1 MB (4 cores of 256 KB each).

The third cache level was common to the entire CPU. Its size was 8 MB.

Random Access Memory

The i7 870 processor was focused on using the most advanced type of random access memory device at that time - “DDR3”. The manufacturer himself recommended using “DDR3-1066” or “DDR3-1333” brackets with this CPU. In practice, it was possible to install faster modules of this type of RAM, but they could not function faster than DDR3-1333. The maximum amount of RAM that the processor could address in this case is 16 GB.

Another important technological nuance is the integration of a dual-channel RAM controller into a semiconductor crystal. This engineering solution has significantly increased the performance of the final computer system.

Operating temperature of the semiconductor crystal. Thermal package

The maximum permissible temperature for the Cor i7 870 was 72.7 degrees. As a rule, in practice this meant that the cooling system had failed for some reason. In normal operating mode, the temperature of this chip is in the range of 40-55 degrees. If you overclock this flagship processor, then the temperature regime will change. In this situation, it will be between 50 and 60 degrees. It is possible to overcome the symbolic value of 60 degrees only when running the most demanding software. But even in such a situation, the temperature is unlikely to go beyond the permissible limits and will not exceed 70 degrees. The thermal package of this chip was equal to 95 W.

Frequency formula

This chip was equipped with technology such as "Turbo Boost". Therefore, its clock frequency varied depending on the degree of complexity of the problem being solved, the amount of computing resources involved and the thermal state of the CPU. Reference with increased heat dissipation in the Core i7 870 - 2.93 GHz. The next value behind it is 3.2 GHz.

In this case, all 4 modules work at once in both modes. Only in the first case there is a decrease due to overheating of the chip or due to the simplicity of the problem being solved, and in the second - in all other situations when 4 cores are needed to work at once. The next value is 3.47 GHz. When switching to this mode, 2 computing modules are switched off in the processor. The maximum frequency in this case is 3.6 GHz. The computer system works with only 1 thread.

Architecture

The code name for the computing units on this chip is Nehalem. This central processor included 4 cores at once. In addition, a distinctive feature of the semiconductor solutions of the “Cor I7” series is the support of a proprietary technology code-named “Hyper Trading”, which made it possible to obtain 2 logical ones based on one physical core (that is, at the software level). This semiconductor crystal in this case was no exception. As a result, at the software level it showed 8 computational threads at once.

All this allows him, even now, to cope with any application software (both optimized for multi-threading and not) without any particular difficulties.

Overclocking

Theoretically, it was possible to overclock the Intel i7 870 using the system bus. But in practice, the manufacturer himself recommended using chips from the “Kor I7 9xx” series, which also had the “Black Edison” prefix, for these purposes.

In the case of the Cor I7 8XX, the level of performance was initially excessive, and creating an additional significant load on the computer system was completely impractical. Now the performance of this CPU can be assessed as sufficient. It allows you to run any software that is currently available.

The I7 870 takes off skyward without accelerating. Motherboards

Price reduction course

Lynnfield range

New chipset: Intel P55

First disappointment: Lynnfield memory controller

Lynnfield's Secret Weapon: Turbo Mode

How we tested: switching to Windows 7

Performance

Overclocking

Energy consumption measurement

Conclusions

Other materials on this topic

What is this chip capable of? Its market segment

CPU equipment

Socket

Process

Cache

Random Access Memory

Operating temperature of the semiconductor crystal. Thermal package

Frequency formula

Architecture

Overclocking