by Ilya Gavrichenkov
09/07/2009 | 09:04 PM
Almost a year has passed since Intel launched their new Nehalem processors. However, we can’t claim that these CPUs have become very popular among computer users since then. Although Nehalem processors do provide higher performance than the Core 2 generation solutions, they didn’t enjoy very high demand. Even according to statistics from the CPU-Z utility database, which many enthusiasts refer to, the share of computer systems built around Core i7 CPUs is currently only a little over 10%.
The reasons for this relatively weak public interest towards Core i7 CPUs is pretty clear: they are extremely expensive and use their own LGA1366 platform including special mainboards and DDR3 SDRAM, which ends up being just as substantial investment as the CPU itself. However, Intel hasn’t been too upset about Nehalem processors being not very popular among individual computer enthusiasts so far. The company’s primary goal at this point was strengthening their positions in the server market. And this CPU turned out an excellent solution in the multi-processor segment. The new QPI bus with point-to-point topology connecting processors with one another as well as the memory controller integrated into each CPU become those essential components that put the performance of Nehalem based server solutions on a new quality level. As for the desktop users, they can’t really feel the effect of all these advantages that is the server processor for desktop applications represented by Core i7 has not yet become a sales hit.
However, the previous Core 2 processor generation continued to maintain its desktop market share outperforming the competitor’s solutions and offering quite sufficient performance for typical user tasks even after Nehalem launch. Only now things started changing a little: AMD has finally mastered 45 nm manufacturing technology, which allowed them to start mass production of Phenom II CPUs capable of competing against Core 2 in performance (with a certain proviso). Of course, Intel wasn’t too happy about it that is why they chose to offer a dramatic refresh of their mainstream platform this fall. This is in fact a pretty logical refresh that is why it was quite predictable as well: starting today processors on Nehalem microarchitecture will be used not only in high-end systems, but also in mainstream platforms. Moreover, Intel doesn’t seem to be concerned about the high price of the LGA1366 platform at all, because they are going to upgrade the mainstream segment with a new LGA1156 socket, new chipset and new mainboards. All these new hardware components should become more affordable for the mainstream user community due to slight architectural modifications in processors and systems. So, our today’s article is going to talk about the new platform composed of these hardware components.
The first Nehalem generation processors for desktop computer systems that belong to Core i7-900 series were based on Bloomfield processor dies. They featured four computational cores, a single 8MB L3 cache, DDR3 memory controller and QPI bus controller. This way, Core i7 processors were unified with the server Xeons, but at the same time caused certain inconveniences when used in desktops. For example, they required triple-channel DDR3 SDRAM, which is not only unusual but even somewhat excessive for desktop use.
As a result, Bloomfield based platform looked as follows:
As for the mainstream solutions, Intel decided to design new semiconductor dies that retain the same key features as the Bloomfield ones, namely, quad-core design and 8MB shared L3 cache memory, but offer more attractive combination of price and features. These “optimized” quad-core processors that also belong to Nehalem generation were codenamed Lynnfield.
I have to point out right away that Lynnfield processors are made with the same 45 nm technology, as Bloomfield processors. Although Intel is planning to implement 32 nm process later this year, they are not going to trial run it on the new CPUs. We should see the products of the new 32 nm process a little later in solutions known as Clarkdale, which should make Nehalem microarchitecture even more affordable for users with limited financial means. So, at first glance the differences between Bloomfield and Lynnfield don’t seem to be too drastic, however, they turn out more than enough to lower the total platform cost almost by half.
As we know, one of the major advantages of Nehalem microarchitecture is modular structure of the processor die that allows the developers to easily change the set of processor functional units without any massive redesign. Intel engineers too advantage of this particular feature when they created their Lynnfield solution. Namely, they first of all replaced the triple-channel DDR3 SDRAM controller with a simpler dual-channel one. Lowering the number of memory channels from three to two doesn’t have that big of an effect on the overall system performance, but directly affects the system cost, because requires fewer DDR3 modules.
They also made one more change towards simplifying the platform, which has been long called for. Obviously, the QPI bus that migrated to Bloomfield solely because of its server roots performs only one function: connecting the CPU with the controller for the PCI Express bus used to connect to the graphics subsystem. So, nothing will really change from a functional standpoint if we remove the X58 IOH chip and QPI bus and replace the QPI controller in the CPU with a PCI Express bus controller. This is exactly what they have done. As a result, systems based on Lynnfield processors can boast the following much simpler structure:
Lynnfield processors have a PCI Express 2.0 bus controller instead of the high-speed QPI bus interface. This controller supports 16 PCI Express 2.0 lanes and allows using either one graphics card or creating dual-card ATI CrossfireX and Nvidia SLI configurations in 8x + 8x mode. Moreover, the CPU acquired low-speed DMI bus connecting the processor with the chipset South Bridge.
As a result, Lynnfield based platform allows not only to give up triple-channel memory in favor of dual-channel one, but also to do just fine without the chipset North Bridge. This certainly helps simplify the mainboard design. And the outcome is quite predictable: not only Intel is going to sell Lynnfield processors at a lower price than the senior ones, but the users will also save some money on memory and mainboards. As a result, it is now possible to fit a Nehalem CPU, mainboard and memory into a sub-$400 budget.
At the same time, it is pretty funny that Lynnfield semiconductor die turned out a little larger in size than a more expensive Bloomfield one: the implementation of a PCI Express controller called for more transistors than the QPI bus controller.
The table below shows how the parameters of these two semiconductor dies compare with one another:
Nevertheless, this difference in die size didn’t prevent Intel from lowering the cost of Lynnfield processors compared with the predecessors. The official Lynnfield prices are going to be from $200 to $555, while different Bloomfield processor models currently range from $285 to $1000.
Lynnfield processors are inferior to the existing Bloomfield CPUs not only in the number of channels of the integrated memory controller. Their clock frequencies are also different: cheaper Lynnfield processors will work at a slightly lower clock speeds. However, these differences don’t seem too significant that is why it is quite logical that Lynnfield processors will sell under the same trademark as Bloomfield, namely, Core i7.
However, the junior Lynnfield CPU model will nevertheless be assigned to a lower class family called Core i5, but unlike other desktop Nehalem CPUs it will have no Hyper-Threading technology support. So, it explains a lot in Intel’s logics in naming their processor models. Those quad-core processors that the OS sees as eight-core ones due to Hyper-Threading technology support will belong to Core i7 family. If the operating system sees the processor as a quad-core one, it will be called Core i5. Therefore, we can suppose that the Core i9 processor family will consists of upcoming six-core CPUs supporting Hyper-Threading technology, which are currently known as Gulftown. As for Core i3 processors, the logics will be different here: this family will include budget CPUs with limited functionality.
At first Lynnfield lineup will include three solutions with 2.66, 2.8 and 2.93 GHz clock speeds. It is very important that due to lower clock frequencies these processors will also boast lower 95 W thermal design power instead of 130 W as seen by the Bloomfield CPUs. As a result, these new processors can be considered a worthy replacement for Core 2 not only in performance but also in power consumption aspects.
The complete list of currently available Nehalem processors includes the following six models:
Since Lynnfield is very similar to Bloomfield on the microarchitectural level and actually differs only in its Uncore part, we shouldn’t be surprised to see identical primary specifications of the Core i7-900, Core i7-800 and Core i5-750.
The following features typical of the new Lynnfield processor structure make them related in many aspects to the existing Core i7-900 CPU models:
You can see that the new Core i7-900 processors and the new junior CPUs are related from the diagnostic CPU-Z utility reports. Namely, they read the following for Core i7-870 and Core i5-750 processors that we received in our lab:
Overall, everything looks exactly like by our good old friends – Core i7-900. The only thing that was a little confusing on the screenshots above is the QPI bus frequency. It is obviously a program glitch, because Lynnfield processors simply don’t have this bus. As for the CPU clock frequency, it is obtained as the multiplier multiplied by base clock generator frequency, which is traditionally at 133 MHz for all Nehalem processors.
Judging by the formal specifications, the new Socket LGA1156 is that significant difference between the new processors and Core i7-900. As you can see from the name, this socket features fewer pins than the LGA1366, which is actually not surprising at all, because the integrated memory controller has fewer memory pins and the QPI interface has been replaced with the common PCI Express.
Fewer contacts as well as smaller contact spots in the CPU allowed to make the processor and processor socket physically smaller, about the size of LGA775.
LGA775 (left), LGA1156 (center), LGA1366 (right)
However, if you look at the bottom side of the new solution, you will see that LGA1156 and LGA1366 processors are very different from one another. Although they are of similar size, Lynnfield has much more pins on the bottom:
LGA775 (left), LGA1156 (center), LGA1366 (right)
So, new Core i7-800 and Core i5-700 processors are incompatible with any old platforms and require their own LGA1156 mainboards. Moreover, new processors also need their own cooling systems. According to platform guidelines the cooler retention holes on LGA1156 mainboards should be spaced out at a smaller distance from one another than on LGA1366 mainboards, but at a farther distance than on LGA775 mainboards. Frankly speaking, since the typical heat dissipation of top LGA775 and LGA1156 platforms is the same, this differentiation of the cooling solutions is a little puzzling. However, the fact I undeniable: Core i7-800 and Core i7-700 require their own coolers.
At the same time we should say a few words about the boxed cooler that we received together with the Core i7-870 processor. Although this processor is the top solution in the Lynnfield family, the cooler for it is relatively small. Its aluminum heatsink with the copper heart is only 13 mm tall.
This is clear indication that Lynnfield processors are not as “hot tempered” as Bloomfield, for instance.
The launch of Lynnfield processors resulted into serious changes in platforms structure. Since they are positioned solely for single-processor systems they don’t have QPI interface that is also used in LGA1366 systems to connect the processor and the chipset. That is why Core i7-800 and Core i5-700 processors received their own core logic set called Intel P55 Express.
The main peculiarity of this chipset is its simplicity. Nehalem microarchitecture allowed to remove the memory controller from the chipset and now time has come for the PCI Express controller to follow. Since from now on the CPU is responsible for supporting this bus, all functions that have traditionally been performed by the chipset North Bridge have now migrated to the CPU. As a result, there is no need for the North Bridge chip anymore and Intel P55 Express became the first Intel core logic set made of only one single chip – Platform Controller Hub (PCH).
The processor and the chipset in LGA1156 systems are connected with the Digital Media Interface (DMI) bus with 10 Gbps bandwidth in each direction, which used to be employed to connect the chipset North and South Bridges before. That is why the new Lynnfield processors in LGA1156 systems can be used just fine not only with the new Intel P55 Express chipset but also with the old ICH10 South Bridge.
At the same time, we can’t say that P55 PCH is very different in functionality from ICH10. In fact, they simply updated the existing interfaces and extended their number. The following table gives you a better idea of the innovations:
We need 8 PCI Express 2.0 lanes in the P55 chipset in order to ensure that additional devices can also be connected to the platform besides the graphics cards. The CPU in LGA1156 systems can only provide connection with the PCI Express graphics cards and the P55 PCH is responsible for supporting other devices. In fact, we saw almost the same implementation in X58 Express, where the chipset North Bridge was only responsible for work with graphics cards. An important improvement in P55 Express was the introduction of PCI Express 2.0 bus support. It means that devices supporting this specification can exchange data with the chipset twice as fast as they used to before.
Moreover, the USB controller has also been changed. It doesn’t support new or faster protocol versions, but delivers more ports. Besides, it also allows hardware disabling of individual ports, which is valuable for security purposes.
Mainboard makers have prepared a lot of solutions on Intel P55 Express by the LGA1156 processors launch. As an example of a platform like that we would like to offer Intel’s own offering – DP55KG mainboard – and point out some of its primary architectural peculiarities.
In fact, besides the presence of only one core logic chip, we also notice immediately that it is topped with a pretty “weak” heatsink, which has nothing in common with those sophisticated solutions usually offered by the manufacturers of the enthusiast mainboards. However, Intel didn’t try to save some money here, because P55 Express PCH simply doesn’t need any state of the art cooling. Even though it is manufactured with 65 nm process, its typical heat dissipation is only 4.7 W. For example, the typical heat dissipation of the X58 Express North Bridge is 24.1 W. That is why those mainboard makers who will still use massive chipset cooling systems on their products, will simply increase the cost of their solutions without any reason and thus mislead the users.
The second peculiarity of the LGA1156 platform is the simplified processor socket design. Compared with LGA1366, the processor retention mechanism is fastened to the mainboard with three screws instead of four, and the lock is catching on to one of the screws and doesn’t require an additional metal frame around the socket. However, this is barely a serious modification: they must be simply trying to save some metal here.
I would also like to mention the design of the second PCI Express graphics card slot. Since in case of a dual-card configuration it will only work in 8x mode, Intel removed its second half, which can anyway serve only aesthetic purposes under any circumstances.
We decided to add a separate section into our today’s article devoted to the memory controller integrated into Lynnfield processor, because it differs from the memory controller in Bloomfield. The thing is that they in fact modified the entire Uncore block of the new LGA1156 CPUs, namely the way the bus and L3 cache frequencies are formed.
I have to remind you that Core i7 processors in LGA1366 package use one 133 MHz base clock generator (BCLK) and several independent multipliers that form the frequencies of the computational cores, QPI bus, L3 cache and memory controller as well as DDR3 SDRAM. New LGA1156 Core i7 processors have the same frequency forming principles, but fewer adjustable multipliers, because there is no QPI bus anymore, for instance, and hence its independent multiplier has been removed.
As a result, Core i7-800 and Core i5-700 use only three multipliers:
So, the memory subsystem in Lynnfield processors turns out slower than in Bloomfield processors and not only because of fewer memory channels, but also because of lower memory controller and L3 cache frequencies.
Of course, all this affects the practical memory subsystem bandwidth and latency. If in LGA1366 systems we saw that the use of two active memory channels instead of three didn’t really cause any serious performance drop, then in LGA1156 the memory subsystem is slower, though not too much.
You can see it very clearly in memory subsystem synthetic tests. For example, we decided to compare the practical speed of the memory subsystem in LGA1366 and LGA1156 Bloomfield and Lynnfield processors working at the same frequency of 2.93 GHz. We used DDR3-1333 SDRAM in both systems with the same 7-7-7-18 timings.
For example, we can see the following results in Cachemem tests from the popular Everest diagnostic tool.
Bloomfield 2.93 GHz, three memory channels
Lynnfield 2.93 GHz, two memory channels
While the L3 cache memory of both processors works at not very different speeds, we can’t say the same about the memory. Triple-channel Bloomfield memory controller shows slightly higher performance during all memory operations than the dual-channel Lynnfield controller. The only consolation here could probably be the fact that the memory subsystem of the new Lynnfield platform shows slightly lower latency.
Overall, the same results were demonstrated by another utility measuring the memory subsystem practical parameters – MaxMem:
Bloomfield 2.93 GHz, three memory channels
Lynnfield 2.93 GHz, two memory channels
However, in this case we see not only a slight drop in practical bandwidth in LGA1156 systems, but also a little increase in latency.
So, the new LGA1156 platform cannot be regarded as a fully fledged replacement to the LGA1366. Of course, top Lynnfield processors will be able to compete against junior Bloomfields, but the fastest Core i7-900 will anyway remain undefeated. And this victory will be guaranteed not only by higher clock frequencies and support of CoressfireX or SLI 16x + 16x configurations, but also by a little higher performance of the memory subsystem.
One of the most interesting innovations introduced in Nehalem processors is the special Power Control Unit (PCU) that can control and manage the power consumption of individual processor cores. Due to PCU Intel Core i7 processors got Turbo Mode technology that delivers dynamic and automatic CPU overclocking. I would like to remind you the idea behind this technology: when the processor is not fully utilized and its power consumption is far from threshold values the CPU raises its clock multiplier beyond the nominal setting. This technology becomes especially useful when the CPU workload is not clearly multi-threaded.
Core i7-900 processors could increase their multiplier by 1x and with an actively utilized single core – by 2x. As a result, the frequency of these processors was often 133 or 266 MHz above the nominal value. Of course, this is not a significant frequency increase, but even thanks to it alone the average performance of LGA1366 platforms with activated Turbo Mode was 3-5% higher than the performance of similar systems working without this technology enabled. In other words, Turbo Mode technology worked very well in Core i7-900 and proved absolutely worthy there.
That is why Lynnfield processors continued on with this technology. Its working principles remained the same, but new processors got the opportunity to manage their clock frequency in an even more aggressive manner. The multiplier in the new CPUs can be increased by up to 5x, which means that in favorable conditions the Core i7-800 and Core i5-700 clock frequencies may increase by 667 MHz over the nominal values, and this is a pretty significant increase I should say. However, it is important to understand that the actual frequency gain can only be determined taking into account the current CPU utilization and its power consumption. For example, the multiplier can be increased by 5x only if one processor core is loaded with work. If two or three cores of the four are utilized, then the multiplier can only be increased by 4x. But even if all four processors cores are loaded with work at the time, the clock frequency multiplier can be increased by 2x or 1x.
You can get a better idea of Lynnfield frequencies with enabled Turbo Mode technology from the following table:
There is one very important conclusion that we can draw here. When the CPU load is not multi-threaded or lightly-threaded, Core i7-800 and Core i5-700 processors may be faster than their predecessors from the top Core i7-900 series, because they can overclock themselves way greater in this case.
And we are not implying some ephemeral frequency increase here. In fact, most mainboards can switch the CPU into Turbo Mode permanently, so that the clock frequency will increase to its maximum irrespective of the current power consumption. Therefore, with Turbo Mode enabled the users will most likely see the following (to illustrate what we have just said we would like to provide screenshots taken off the system on Core i7-870 processor with 2.93 GHz nominal frequency):
Not only the frequency screenshots look impressive. To estimate the real benefit from Turbo Mode we compared the performance of the Core i7-870 based system with Turbo Mode and without it.
The average improvement from enabling the new second version of this technology makes about 8%. This gain can be even higher in applications that are not very well optimized for systems with multi-core processors. As a result, Turbo Mode looks an excellent trump of the new LGA1156 platform making it even more attractive for the user. Due to this technology Intel helps those software developers who haven’t yet got to optimizing their applications for multi-threading concept. New Core i7-800 and Core i5-700 processors, which are in fact real quad-core CPUs, may turn into high-speed pseudo dual-core or pseudo single-core solutions if necessary. And it is especially nice that this transformation happens in the background and doesn’t require any actions on the user’s part.
The main heroes of our today’s performance tests are Core i7-870 and Core i5-750 LGA1156 processors that are the senior and junior representatives of the new Lynnfield family. They will be competing against the CPUs with the same price tag available for all other platforms: LGA1366, LGA775 and Socket AM3.
As a result, our testbeds were built using the following hardware and software components:
I would like to point out separately that we are switching to the new Windows 7 operating system in our tests. Although it hasn’t been officially announced yet, it already exists as the final RTM version. In our case, when we refer to Core i7 CPU benchmarks, it is especially important. The thing is that this operating system has special optimizations improving the performance of systems supporting Hyper-Threading technology. Intel and Microsoft engineers have worked closely together on implementing SMT parking technology that optimizes Windows 7 for CPUs with virtual cores. As a result, it should eliminate most situations when Hyper-Threading technology could slow down certain applications in Windows Vista, because the Windows 7 scheduler distinguishes between physical and virtual cores and prevents the situations when execution of two simultaneous threads on a single core causes performance to drop.
Looks like our concerns about insufficiently fast memory controller in Lynnfield processors were actually unjustified. In terms of actual performance, Core i7-870 and Core i5-750 are excellent solutions. In absolutely all scenarios even the junior newcomer without Hyper-Threading technology support is faster than Core 2 Quad and faster than the competing Phenom II processors.
As for the performance comparison between Lynnfield and Bloomfield, the results here are not so unambiguous, but nevertheless, Core i7-870 and Core i5-750 both look very good performing fast in E-Learning and 3D patterns and only falling behind the “senior” Intel platform during video content processing.
LGA1156 proves to be an excellent gaming platform. When we switched to Windows 7 operating system, all former issues with Core i7 performance in games vanished. As a result, even a $200 Core i5-750 is by far faster than Phenom II and Core 2 Quad. Core i7-870, however, quite successfully competes against the LGA1366 solution – Core i7-950.
The situation with performance in popular codecs looks a little bit more interesting. First of all we should pay attention to the fast performance of LGA1156 processors during audio files transcoding in iTunes. It is a clear demonstration of the advantages in second Turbo Mode version implemented in Lynnfield, because iTunes is one of those applications that create only two-threaded load.
As we know, DivX was well optimized for quad-core processors, only it doesn’t really new Hyper-Threading technology. We can draw this conclusion from the similar results demonstrated by Core i7-870 and Core i5-750 in this codec. However, even though one of the most important Nehalem processor technologies is sort of out of business here, their computational capacity is more than enough to outperform all processors with a different microarchitecture. We don’t see any serious performance difference here between LGA1366 and LGA1156 Nehalem modifications.
However, when we use x264 codec, Hyper-Threading ensures a pretty significant performance boost. As a result, Core i7 supporting this technology are far ahead of the opponents at the very top of the diagram. Moreover, the performance of the new Core i7-870 is just a little below that of Core i7-950. As for the results demonstrated by Core i5-750 that doesn’t support Hyper-Threading, it falls pretty far behind its fellow processors. Nevertheless, even despite that, it outperforms almost all other quad-core testing participants that do not belong to the Core i7 lineup, except Phenom II X4 965.
Non-linear video editing is a task that loads all processor cores very seriously and requires high memory subsystem bandwidth. As a result, Turbo Mode technology is helpless, which pushes Lynnfield quite far behind Bloomfield. Nevertheless, LGA1156 platform still looks pretty attractive against the background of all other processors.
During image editing in a freeware Paint.Net application Core i5-750 is about as fast as the top Core 2 Quad solutions, while Core i7-870 works a little faster than Core i7-920. In Photoshop CS4 LGA1156 processors outperform their predecessors due to very successful implementation of Turbo Mode technology. As a result, they are considerably far ahead of other testing participants.
Final rendering seems to be one of the best paralleled tasks. Therefore, the best results here belong to quad-core processors supporting Hyper-Threading technology. Namely, Core i7-870 is just a little behind Core i7-950, but at the same time it is 30-40% faster than Core 2 Quad and Phenom II X4. As for the Core i5-750 processors that doesn’t support Hyper-Threading, it can only boast the same performance as Socket AM3 and LGA775 processors.
Due to high-speed integrated memory controller, Core i7 and Core i5 processors can archive way better than their competitors. Even the junior member of this family, Core i5-750, is far ahead of Core 2 Quad and Phenom II X4 CPUs, which are actually also equipped with an integrated memory controller.
The results of Mathematica suite are pretty predictable. Just like in many other applications, the new CPUs outperform all rivals except LGA1366 solutions, where they retain certain parity of results.
Just like in many other computational tasks, Hyper-Threading support determines the outcome in the distributed computing project called Folding@Home. Core i7-870 supporting this technology runs very close to Core i7-950. As for the performance of Core i5-750 without the Hyper-Threading support, it is at the level of top Core 2 Quad and Phenom II X4 processors.
In conclusion to our Lynnfield performance tests we decided to see how they will perform under single-thread load, when Turbo Mode technology can overclock them to the maximum. For our experiments we decided to go with the good old SuperPi benchmark that calculated 8 million digits of the π number, and a single-thread Cinebench R10 final rendering test.
Core i7-870 and Core i5-750 processors in these somewhat artificial conditions performed outstandingly. Turbo Mode increased their clock speeds to 3.6 and 3.2 GHz respectively, which helped them to seriously outperform even the more expensive LGA1366 solutions. All other quad-core processors that should be direct competitors to Lynnfield from the price standpoint are hopelessly behind even the junior Core i5-750 CPU, not to mention the senior LGA1156 solution.
In other words, even though LGA1156 platform is currently available only with three quad-core processors, it makes sense to upgrade to it even if you are not using multi-threaded applications at all. Aggressive Turbo Mode implementation makes Core i7-800 and Core i5-700 performance really impressive not only under multi-threaded load but also in applications using only one or two threads. And we should give Intel due credit for that because the second Turbo Mode version implemented in the new Lynnfield CPUs helps to make sure that the right number of processor cores is involved at all times.
Lynnfield die is not very different from Bloomfield. True, these processors are mad with the same manufacturing process and their major units are the same. Therefore, it would be strange if the overclocking potential of the freshly announced LGA1156 processors could be very different from that of LGA1366 CPUs. Nevertheless, to check out this hypothesis we performed a number of overclocking experiments of the Core i7-870 and Core i5-750 processors.
The tests were performed on a Gigabyte GA-P55-UD6 platform. The CPU was cooled with Thermalright MUX-120 solution with traditionally curved base plate and Enermax Magma UCMA12 fan (1500 RPM). System stability under load was tested using LinX 0.6.3 utility.
There is only one way to overclock LGA1156 processors: by raising the BCLK clock generator frequency. This is when the Uncore frequency increases together with the CPU core clock. However, there is nothing you can do about it, because Lynnfield processors have not only a locked clock frequency multiplier but also a locked multiplier for the Uncore frequency. The memory frequency increases together with BCLK, but luckily its multipliers can be lowered.
When we tried to overclock Core i7-870 processor we managed to achieve full stability at 4.07 GHz.
To achieve this result we had to increase the processor core voltage to 1.4 V, which is a relatively safe level for Lynnfield, provided there is proper cooling. However, in our case CPU core temperature reached 93 °C. And although it is a really high temperature, the CPU remained absolutely stable and didn’t overheat. So, CPUs from Core i7-800 family can really work at 4 GHz frequency with air cooling, just like their elder brothers from the Core i7-900 series.
The second part of our experiments dealt with Core i5-750 processor. This CPU doesn’t support Hyper-Threading technology, which leads to its lower temperature under full load. We hope that this peculiarity of the Core i5 will make its overclocking more successful. However, on the other hand, Core i5-750 has a lower clock multiplier, which requires increasing the BCLK frequency during overclocking to higher levels. Luckily, the maximum BCLK frequency of 210-215 MHz that we saw on typical LGA1366 platforms, can be easily surpassed on LGA1156 platforms.
However, we didn’t have to increase the BCLK frequency beyond 210 MHz. Our Core i5-750 sample could work stably only at 4.1 GHz, which only needs 205 MHz BCLK.
The core voltage was increased to 1.4 V, but the temperature of the overclocked processors didn’t exceed 81 °C. So, it appears that despite relatively small difference between the overclocking results of Core i7-870 and Core i5-750, the temperature of a processor without Hyper-Threading support is actually way lower under maximum load. And it means that during overclocking experiments with Core i5-750 we can also use relatively inexpensive cooling systems.
I have to say that we performed our overclocking experiments with dynamic multiplier adjustment via Turbo Mode disabled. However, I have to admit that overclocking with enabled Turbo Mode is pretty interesting, too. It is quite possible that in case of low computational load on the CPU, its frequency may be increased higher than we managed to achieve today. Therefore, very soon we are going to publish a new article where we will discuss in greater detail all aspects of LGA1156 processors overclocking.
One of the most intriguing features of the LGA1156 processors is their thermal design power of 95 W. It is 35 W lower than the thermal design power of almost the same LGA1366 processors. If we add here the considerably lower heat dissipation of the chipset on LGA1156 platform, then we can expect the new Intel platform to become a pretty attractive solution in terms of performance and power consumption ratio. And most likely this platform will even be able to compete against LGA775, which we have long admired when it comes to energy-efficiency.
However, it is very reckless to take the manufacturer’s promises for granted in this case. Therefore, we tested the actual power consumption of all participating platforms. The following numbers show the total power consumption of the tested platforms (without the monitor). During our tests we used 64-bit LinX 0.6.3 utility to load the systems to the utmost extent. Moreover, to ensure that we estimate the power consumption in idle mode correctly we activated all power-saving technologies, such as C1E, Cool'n'Quiet 3.0 and Enhanced Intel SpeedStep.
In idle mode the power consumption of the LGA1156 platform looks better than impressive. Among all today’s testing participants it is the Lynnfield platforms that consume less power of all. It happened due to the fact that processor power-saving technologies have once again been improved in the new CPUs. In particular, Core i7-870 and Core i5-750 processors can drop their frequency to 1.2 GHz and their core voltage to 0.85 V.
The situation looks pretty good when we measure the power consumption under load. LGA1156 systems are definitely more economical than LGA1366 platforms, their power consumption became more than 50 W lower. Moreover, the Core i5-750 system that doesn’t support Hyper-Threading turns out even more energy-efficient than LGA775 platforms. In other words, our expectations came totally true: LGA1156 systems cam offer better performance-per-watt than any other solutions. More to that, Intel also plans to release special energy-efficient LGA1156 processors, so this platform has every chance to become the ultimate favorite for those who care about their electrical bills.
To get a better picture of the situation we also tested the power consumption of the processors and mainboards under heavy load without taking into account the rest of the system components. To be more exact, we measured the power consumption along the 12 V power line connected directly to the processor voltage regulator on the mainboard and along the mainboard power lines.
Strange as it might seem, but the lowest power consumption among all the testing participants belonged to Core i5-750 processor, which turned out more energy-efficient than even Core 2 Quad Q9400. However, low power consumption demonstrated by CPUs on Nehalem microarchitecture can partially be explained by their voltage regulator design. The thing is that the isolated 12 V power line is only used for processor cores. As for the Uncore part of the CPU, it is powered from the mainboard via a 24-pin connector. That is why this time we also decided to include the power consumption readings taken off the mainboard.
Keeping in mind everything that has been just said, it is no wonder that LGA1366 and LGA1156 mainboards are most energy-hungry. Uncore part of the CPU contributes significantly into the results of our power consumption measurements. Nevertheless, we can’t help noticing that elimination of the North Bridge in LGA1156 systems did in fact make the corresponding mainboards more energy-efficient. The power consumption difference between Bloomfield and Lynnfield mainboards reaches 20 W overall.
All in all, LGA1156 platform makes a very good impression. And although it is evident that Intel’s main goal with the new Lynnfield family launch is the transfer of Nehalem microarchitecture to the mainstream price segment, we often got the feeling that it was not a lot-cost LGA1366 platform modification we were dealing with, but its improved and upgraded version.
And we can’t say that this feeling was absolutely ungrounded. The new LGA1156 platform really has a number of advantages. First of all, it is easier to perceive: it works with common dual-channel memory and Lynnfield processors are none other but real desktop and not server solutions stuck into the desktop systems for purely marketing reasons. Secondly, Core i7-800 and Core i5-700 processors have much more attractive power consumption than their predecessors. By slightly changing the platform structure Intel engineers managed to bring the power consumption of Lynnfield platforms to the level of LGA775 platforms that have so far been the best example of great performance per watt ratio. Thirdly, Turbo Mode technology is the biggest advantage of the new processors that allows Lynnfield CPUs to remain efficient even when the created load doesn’t have the definitive multi-threaded nature.
However, if we take into account objective rather than subjective factors, LGA1156 platform will still not be able to change the set state of things. Despite all its evident advantages, LGA1366 mainboards and CPUs will remain popular in the upper price segment. Only these systems allow building multi-GPU configurations using two graphics cards working as PCI Express x16 + x16 or even more cards. Only LGA1366 platforms will be compatible with the upcoming six-core Gulftown processors. And only among Core i7-900 processor family we can find Extreme Edition solutions that offer not only unprecedented performance but also additional overclocking friendly capabilities.
Nevertheless, Core i7-800 and Core i5-700 processors seem to be an excellent replacement for the LGA775 Core 2 Quad CPUs offering much higher performance at the same price. The launch of LGA1156 platform brings a real revolution to the mainstream segment. This platform immediately turns Core 2 and Phenom II CPUs into outdated solutions that can only be of interest in the sub-$200 category.
In other words, from now on we can claim that the true Nehalem era has come. CPUs based on this microarchitecture became not just affordable. They matured and got truly attractive. So, no doubt that the new Core i7-800 and Core i5-700 processors will quickly become very popular and turn into real sales hits.