by Ilya Gavrichenkov
01/22/2009 | 06:41 PM
We continue talking about processors built on new Nehalem microarchitecture. Following our theoretical material and an article devoted to performance of systems built on Core i7 processors, we decided to take a closer look at a matter that is of great interest to computer enthusiasts in the first place. This notorious topic is overclocking. And even though many users still do not see how they can benefit from overclocking their system, army of overclockers keeps multiplying. And the reason for that is not only the growing general interest to new technologies, but also the fact that many hardware manufacturers have finally turned their face to the overclocker community. Trying to win more users for themselves many computer hardware manufacturers add new functionality that may help reveal undocumented potential of the equipment. And even Intel who have been fighting the mere idea of overclocking a few years ago, today tempered justice with mercy. Now they no longer deny the possibility of using their processors in operational modes other than the nominal one. Moreover, they encourage overclockers by inviting them to versatile events and adapting their CPUs and mainboards for them accordingly.
Therefore, we believe that the launch of new microarchitecture has every chance to become a new catalyst for further overclocking popularization, because the systems built for Core i7 processors have become easier and also more interesting to overclock. Besides, we shouldn’t forget about the platform changes such as introduction of the new processor voltage regulator circuitry, relocation of the memory controller into the CPU and elimination of the FSB bus. They certainly make overclocking also much more affordable, because the influence of the most sensitive system component - the mainboard - on overclocking results reduces tremendously.
In order to increase the practical value of our today’s article, we decided not to use any engineering samples. Instead we took the already retailing mass production CPU, mainboard, memory and cooling solution. Our today’s main hero will be Core i7-920, the most affordable member of the Nehalem family. And in the end of our today’s investigation, we will be able to give you a recipe for successful overclocking that will allow you to make a $300 CPU run much faster than one of the most expensive processors in the market today: Core i7-965 Extreme Edition.
In our today’s article we will try to reveal all tricks and peculiarities of successful LGA1366 systems overclocking. However, we assume that our readers already have basic knowledge regarding the structure of Nehalem systems. If it is your first experience with the new platform, we strongly advise to take a look at our article called “First Look at Nehalem Microarchitecture” before you proceed.
Although overclocking Core i7 based systems is a fairly new procedure, it is not that complicated at all. We believe that overclocking systems built on new CPUs is not any harder than platforms built around previous generation quad-core Core 2 Quad processors. However, it is important to understand that since one of the major changes brought by the new Nehalem microarchitecture is an absolutely new platform design, Core i7 overclocking requires a totally different approach.
Generally, overclocking old LGA775 systems is performed by raising the processor bus frequency. As the processor bus frequency increases, so do the processor and memory frequencies are proportionally connected with the FSB speed by means of multipliers and dividers. The processor clock frequency multiplier is defined by the nominal CPU frequency, but can also be set lower if necessary. The only exceptions here make Extreme Edition processors that have an unlocked clock multiplier. It allows overclocking by simply setting the clock frequency multiplier at a higher value than the nominal. As for the divider connecting the FSB and memory frequencies is set by the chipset North Bridge, which features the memory controller in LGA775 systems. Contemporary chipsets support wide range of memory frequency dividers, so the memory frequency can be adjusted in a very flexible manner even overclocked independently of the CPU.
The situation is completely different in LGA1366 platforms using new Core i7 CPUs. These processors not only have 8MB shared L3 cache and an integrated memory controller, they use a totally new series interface to connect to the chipset. As a result, new generation systems have no traditional FSB bus that used to be the primary factor shaping up all the system frequencies. Now this key role belongs to the so-called base frequency – BCLK, which has no application whatsoever in its initial form. However, BCLK frequency is used to set the frequencies of all major functional parts of an LGA1366 system. Among them are:
Core i7 processors have four different multipliers used to obtain these four frequency values. In other words:
All four multipliers mentioned above are completely independent. The only exceptions are the memory frequency multiplier and the multiplier for the North Bridge built into the processor: [Uncore multiplier] should be at least twice as high as the [Memory multiplier].
The nominal BCLK frequency for any Core i7 processors equals 133MHz. However, the derived frequencies may vary depending on the CPU models. The table below describes nominal frequencies for the Core i7 lineup defined by the official specifications. Currently, the lineup includes three models:
Although the specs define all major frequencies for each processor model, Intel in fact offers a little more freedom with coefficients determining these frequencies. In reality, only the processor clock multiplier and QPI multiplier have strictly set maximums. All other multipliers of the mass production CPUs can be changed in pretty wide range. The table below describes value intervals available for different CPU models (the nominal values are marked with bold font):
So, Core i7 processors, except for the very expensive Core i7-965 Extreme Edition CPU, overclock in only one single way: by raising the base BCLK frequency. However, once you set it over the nominal 133MHz, the frequencies of all other system parts will automatically rise over their nominal value, namely the L3 cache, memory and QPI bus frequencies. Unfortunately, setting lower multipliers for secondary frequencies will help to compensate the proportional increase only for DDR3 SDRAM, because all processors except the top model already have the minimal multipliers set for Uncore and QPI. But the good news is that L3 cache and QPI bus have great potential at higher frequencies. Therefore, overclocking will be limited in most cases by the processor cores and not by their “support group”.
Every overclocker knows that one of the integral factors for successful CPU overclocking is the voltage increase for different parts of the platform. For example, when you overclock an LGA775 system, you often have to raise the processor core voltage, memory voltage, processor bus voltage and chipset voltage. Setting these parameters above their nominal values almost always improves the overclocking potential of the system. Although, you should also keep in mind that higher voltage in semiconductor components results in higher heat dissipation and logically, lower life span of these components. However, using a high-quality cooling system and increasing the nominal voltages to reasonable heights allows the user to find a compromise between so called “risk factors” and frequency potential increase.
The same is true for the new generation platforms. However, Core i7 based systems have different structure that is why their voltage management during overclocking experiments requires a completely different approach. Since the chipset North Bridge and processor bus have lost their determinative role, their voltages do not need to be adjusted in most cases, even when the frequencies get pushed up fairly high. However, the memory controller, which has migrated into the processor, and L3 cache now receive their power supply independently, so playing with it may actually pay back during overclocking.
So, there are four major voltages to work with during Core i7 overclocking. They are:
As you know, increasing the voltage during overclocking makes the heat dissipation grow according to square-law. Therefore, when you overclock Core i7, as well as any other processors, it is important to monitor temperatures very carefully. Maximum allowed temperature for Core i7 is 100°C. If it heats up more, the CPU will enable thermal throttling, i.e. the Vcore and clock frequency multiplier will be forced down to 12x. This feature protects the processor die against dangerous overheating.
There are several different utilities out there that allow monitoring CPU temperatures, for example, CoreTemp or RealTemp. Using these utilities during stability tests of an overclocked processor will help find the optimal voltage setting for the overclocked CPU. They may also let you know if you need to improve your cooling.
However, it is important to keep in mind that Core i7 processors report only the temperatures of their computational cores, which allows to be more or less certain that these processor parts will not get overheated. At the same time, there is no way to control the temperature of the North Bridge built into the CPU. Besides, Core i7 has no integrated means of warning you about the overheating of the L3 cache and integrated memory controller, so you have to be extremely cautious when raising the Uncore and memory voltages.
You may think that the theoretical info we shared above is more than enough to move on to Core i7 practical overclocking experiments. It is partially true. However, the harmonious structure of interconnected multipliers, voltages and frequencies gets slightly messed up by additional innovations introduced in the new generation processors. I am talking about Turbo Boost Technology – sort of dynamic overclocking integrated by Intel into their new CPUs.
I would like to remind you that Turbo Boost implies the ability of the processor to increase its frequency multiplier over the nominal value if it doesn’t push the power consumption over the set threshold of 130W. The current implementation of this technology allows Core i7 processors to have their multiplier increase by 2 if there is only one core loaded with work, and by 1 if most cores are utilized.
The CPU seems to be treating its own multiplier pretty frivolously, so you may think that it could hinder overclocking. However, it is not quite so in reality. On the contrary, those overclockers who decided to go for an LGA1366 system, get an additional tool that may come in handy.
The simplest way is to disable Turbo Boost in the mainboard BIOS. All mainboards for Core i7 processors have this option. Moreover, Turbo Boost technology is directly connected to another technology working with the CPU clock frequency multiplier – Enhanced Intel SpeedStep. As a result, turbo-modes can only be activated when EIST is on. Many overclockers are used to disabling power-saving technologies, which means they automatically lose Turbo Boost.
However, you may not ignore this turbo-mode but use it to your own advantage. The thing is that the BIOS of most LGA1366 platforms allow disabling the processor’s ability to control its power-related parameters without deactivating Turbo Boost. This trick makes it possible to statically increase the processor clock multiplier by 1 over the nominal independent of the workload and its current level of power consumption. As a result, Core i7-920 processor with the default multiplier at 20x can work with 21x multiplier, and Core i7-940 – with 23x multiplier, which its nominal setting is only 22x. Of course, this raise doesn’t look serious enough, but together with increased BCLK frequency it may actually help to achieve better overclocking results.
Here I would only like to add that we do not recommend using turbo mode directly according to its intended purpose, which is also quite possible. Although dynamic increase of the frequency multiplier that occurs during lowering of the workload may do no harm in nominal operational mode, it may cause serious instability during overclocking. The problem is that when you change the BCLK frequency to overclock your processor, the frequency increases more if the multiplier increases. As a result, the CPU may over-overclock itself past the stability threshold when Turbo Boost kicks in. So, you may end up with a processor that successfully passes all stability tests, but when it tries to switch to turbo mode under reasonable workload, the system loses stability.
It is quite logical that one of the major factors for successful overclocking of Core i7 as well as any other CPUs is the right choice of high-quality components. Of course, a mainboard plays the determinative role in an overclocker platform as one of the most important system components tying together CPU, memory, graphics card and peripheral devices. For our tests we chose ASUS P6T Deluxe first of all because ASUS Company is a well-known manufacturer of high-quality overclocker-friendly solutions.
However, we can’t say that a quick look at ASUS P6T Deluxe made any peculiar impression. At first glance it looks like an ordinary board, nothing special. You may even start feeling a little disappointed, because P6T Deluxe sells for almost $300, but doesn’t strike you with its looks. Although it is important to understand that mainboards for Core i7 processors all based on the only compatible chipset – Intel X58 Express, cannot be cheap by definition. Intel set the tempo themselves by selling their Intel X58 for over $50 to the mainboard makers.
Nevertheless, ASUS P6T Deluxe is far not the cheapest Intel X58 based board. The thing is that ASUS engineers decided not to save on small things. For example, the board uses high-reliability electronic components, 8-layer PCB (instead of 6-layer one), and a number of interesting additional controllers. But let’s check out all the peculiarities of ASUS P6T Deluxe one by one.
First of all, I have to point out that this board has three PCI Express x16 slots (compatible with 2.0 protocol version) for graphics cards. These slots may work in two modes: x16/x16/x1 with one or two graphics cards, or x16/x8/x8 with three graphics cards. This way ASUS P6T Deluxe allows using dual-card graphics subsystems without any limitations. It is especially nice that P6T Deluxe was certified by Nvidia, so it supports not only ATI Crossfire technology, but also Nvidia SLI. So, this mainboard may be used as a base for a high-performance gaming platform equipped with any type of graphics accelerators.
Besides three PCI Express x16 slots, the mainboard also has an additional PCI Express x4 slot and two regular PCI slots. One of them, however, will most likely be blocked dead by the graphics card cooling system.
ASUS engineers decided to go with a traditional location for the DIMM slots, they are to the right of the CPU socket. By the way, the reference design suggests this particular placement, and not the one offered by Intel Smackover board. Moreover, unlike Intel board, ASUS P6T Deluxe has six DDR3 SDRAM slots – two per channel. It means that P6T Deluxe may accommodate up to 12GB of memory.
I would like to specifically stress the fact that ASUS engineers paid special attention to the quality of signal lines laid between the CPU and the memory. Take a look: the axis going through the center of the DIMM slots also goes through the center of the CPU. Leveling it out like that may add extra stability to the memory subsystem, for example, during overclocking. By the way, there is a triple-phase voltage regulator for DDR3 SDRAM instead of the dual-phase one like on many other mainboards.
As for the voltage regulator of the CPU itself, its circuitry consists of 16 phases and two additional phases for the North Bridge built into the processor. This circuitry is unprecedentedly complex, and theoretically it ensures extremely “pure” power signal. However, thanks to the traditional EPU controller, this circuitry enables only four phases if the electrical load is low. This makes the voltage regulator very efficient. No wonder that having implemented this complex circuitry ASUS decided not to save on the components. They used long-lasting capacitors with polymer electrolyte, ferrite core chokes and high-frequency Low RDS(on) MOSFET not only for the voltage regulator circuitry but for the entire mainboard.
Transistors surrounding processor socket are topped with ASUS’ traditional aluminum heatsinks of copper color. The chipset South Bridge is cooled with a low-profile heatsink topped with a decorative cover with a lit manufacturer logo that glows when the system is on. The North Bridge, on the contrary, has a massive aluminum heatsink with a lot of sophisticatedly shaped fins. According to ASUS engineers, this shape is ideal for a passive chipset cooler, because the fins go along the airflow created by the CPU cooling system. However, the developers have also made it possible to install a standard 40-mm fan onto this uniquely shaped heatsink. They included special retention stands with the mainboard accessories.
All above mentioned heatsinks are connected into a single cooling system with heatpipes, which is a typical solution for all mainboard from the upper price segment.
I have to say that this cooling system will hardly prevent you from using any massive processor coolers on ASUS P6T Deluxe. But it is not only because all mainboard heatsinks are relatively short, but because LGA1366 mainboards reference design allows moving the processor socket a little away from the upper edge of the PCB. This may be bad news for the owners of old system cases with an air duct leading to the processor cooler. However, the layout change at the top of the PCB frees some additional space that the mainboard designers could use as they see fit.
Speaking of distinctive peculiarities of ASUS P6T Deluxe mainboard, we should point out that this board has an additional SAS controller, which we have never seen before in platforms for computer enthusiasts. ASUS equipped its board with a dual-port Marvell 88SE6320 controller supporting SATA and SAS hard drives and RAID arrays 0 or 1.
By the way, since ICH10R South Bridge used on ASUS P6T Deluxe mainboard doesn’t support PATA devices, ASUS engineers installed an additional Marvell 88SE6111 controller. It provided P6T Deluxe not only with PATA-133 interface but also with an additional eSATA port laid out on the back panel of the board.
If you look at the connector panel of ASUS P6T Deluxe mainboard, you will see that it has two Gigabit network ports implemented via Marvell 88E8056 controller. These ports may be used separately or together – in Teaming mode. There is also an IEEE1394 port and 8 USB 2.0 ports. Another Firewire port and six more USB 2.0 ports are laid out as onboard pin-connectors. Besides the ports and connectors mentioned above, the back panel also bears a combination PS/2 port for keyboard or mouse and audio ports: six audio-jacks and two SPDIF ports – an optical and a coaxial one, that are implemented with ADI AD2000B controller.
Note that ASUS P6T Deluxe has a bunch of very nice pleasing trifles. For example, it supports ExpressGate technology that allows loading SplashTop Linux from built-in flash-storage almost immediately after powering up the mainboard.
Another great little thing is the new Turbo-V application that allows managing all key parameters of the board (BCLK frequency and all voltages) directly from the operating system. It also allows you to work with settings profiles easily.
And finally, the board has Power On and Reset buttons that make life a lot easier for testers.
Although, in fact, we are much more interested in the functionality of the mainboard BIOS Setup related to processor and platform configuration, especially considering the topic of our today’s article. Most overclocking-related settings are gathered together in Ai Tweaker section of the BIOS, as usual.
You can see options for adjusting all major frequencies and voltages. BCLK base frequency can be set from 100 to 500 MHz; memory, Uncore and QPI frequencies in this case are selected from the list of values determined by corresponding supported multipliers. Here I have to stress that the BIOS Setup of ASUS P6T Deluxe mainboard is very convenient to work with: it offers you to make your selection not from the list of multipliers but from the list of actual frequencies obtained with certain multipliers.
Besides the opportunity to change the key frequencies, ASUS P6T Deluxe also offers rich functionality for configuring all the primary and secondary voltages. The list of supported parameters and their value ranges is given in the table below:
This abundance of voltages includes four most important ones (the ones we have just discussed above), so ASUS P6T Deluxe mainboard is quite fit for Core i7 overclocking.
By the way, besides the already mentioned settings, this mainboard also retained a very useful Load-Line Calibration parameter that should help fight harmful voltage drop in the zone between the processor voltage regulator circuitry and the CPU itself (Vdroop effect).
BIOS Setup of the ASUS P6T Deluxe also offers special functions for managing processor technologies. The corresponding section contains options enabling processor power-saving technologies and virtualization technology as well as Turbo Boost mode. Intel Turbo Mode Tech setting enables or disables Turbo mode (once EIST is on), and Intel C-STATE Tech parameter is responsible for activating processor’s own power consumption management.
Otherwise, the BIOS of ASUS P6T Deluxe mainboard hardly differs from the BIOS of other high-end ASUS boards. Therefore, there is only one more curious thing I would like to add here: ASUS’ brand name utilities, such as O.C. Profile (for work with Setup settings profiles) and EZ Flash (for updating the BIOS) integrated in the P6T Deluxe BIOS now can work not only with external data storage media, but also with hard drives formatted for NTFS file system (in read mode).
When you overclock LGA1366 systems we advise using special triple-channel overclocker memory. And the reason for that is not only the fact that this memory sells in triple-module kits. As we know, Core i7 systems also support dual-channel memory and the performance difference in this case would be insignificant. The main problem is Intel’s insistent recommendation not to raise the memory voltage over 1.65V not to damage the memory controller integrated into the CPU. All dual-channel DDR3 SDRAM kits available in the today’s market are optimized for old LGA775 platforms and require higher voltage settings.
In fact, regular DDR3 SDRAM working at its nominal 1.5V voltage and 1333MHz frequency could do just fine during overclocking. The minimal DDR3 SDRAM multiplier of 6x supported by Core i7 processors allows staying within the nominal memory frequency up until BCLK is increased to 222MHz. This potential is more than enough to overclock to its maximum even the youngest Core i7-920 equipped with an air cooler. However, many overclockers prefer to bundle their overclocked CPU with fast memory, which speeds up the entire system even more. Therefore if overclocker Core i7 platforms are often equipped with DDR3 SDRAM that can operate at least at 1600MHz frequency and 1.65V voltage.
Of course, you can find DDR3 SDRAM like that among “previous generation” memory, however, it is way easier to purchase the special DDR3 SDRAM intended for Core i7 platforms. All the leading overclocker memory makers have already launched such kits of their own. For our today’s test session we picked Kingston HyperX KHX16000D3K3/3GX.
We decided to go with this particular kit because it is currently one of the fastest triple-channel DDR3 SDRAM kits for LGA1366 systems available in the market. This triple-channel kit consists of three modules, 1GB each that can work at up to 2000MHz frequency with 1.65V voltage and 9-9-9-27 timings.
The characteristics of Kingston HyperX KHX16000D3K3/3GX memory make it a perfect choice for extreme overclocker systems. The thing is that clocking DDR3 SDRAM at frequencies close to 2000MHz in Core i7 platforms requires a serious voltage boost of the processor Uncore segment: up to 1.6-1.7V. Therefore, we wouldn’t recommend having the memory work at this high frequency without at least liquid-cooling on the processor. Especially since Core i7 CPUs have no built-in tools for detecting if the integrated L3 cache or memory controller gets overheated.
Nevertheless, it doesn’t mean you can’t use Kingston HyperX KHX16000D3K3/3GX memory in “mainstream” overclocker systems equipped with more widely spread air coolers. In this case this memory can work at up to 1780MHz with 1.65V voltage setting and 8-8-8-24 timings and at up to 1550MHz frequencies – with 7-7-7-21 timings.
However, if you don’t feel like spending quite a bit of money on expensive Kingston HyperX KHX16000D3K3/3GX supporting 2000MHz frequency, the same manufacturer offers very high-quality triple-channel memory with 1.65V voltage and lower nominal frequency settings: 1867, 1800 or 1600MHz. the same solution are also available from other overclocker memory makers these days.
It is extremely important to choose a highly efficient cooler for an overclocker LGA1366 platform. The thing is that overclocked Core i7 processors dissipate much more heat, so they require effective cooling in order to reveal their frequency potential to the full extent. The maximum CPU temperature when thermal throttling kicks in is 100°C and the default boxed cooler that comes bundled with Intel Core i7 processor cannot cool them properly even during moderate overclocking.
It looks like there are a lot of very efficient air- and liquid-cooling solutions out there these days, however, most of them are incompatible with LGA1366 platforms. The thing is that Intel changed the retention holes layout on the board: they have been moved farther away from the processor socket compared with the LGA775 platform. Luckily, some cooler makers took matters into their own hands very rapidly and started offering additional retention kits for their flagship solutions, so that they could be used in the new systems. We managed to get our hands on the new retention kit for the Noctua NH-U12P cooler. By the way, Noctua Company is eager to send these kits to everyone who needs one for free.
Noctua NH-U12P cooler has already proven highly efficient in our earlier tests. Therefore, we considered it worthy to become part of our overclocker Core i7 platform.
The new LGA1366 retention kit that got its own name – Noctua LGA1366 SecuFirm2 – allows fastening the well-familiar cooler very securely in the new platform. Its design is similar to that for LGA775, however, it uses a backplate and larger retention brackets. Unfortunately, this new retention kit retained all the drawbacks of its predecessor: it is great for one time installation, but replacing a CPU in you system may turn into a real pain. To uninstall the cooler you have to remove the fan, and if you need to replace the CPU, you have to unscrew one of the retention brackets blocking the processor socket lock-lever.
However, we forgave Noctua NH-U12P these small drawbacks. First, this cooling solution is one of the most efficient processor air coolers in the market today, and secondly, the manufacturer managed to extend the life of its successful product by offering a free retention kit for east upgrade.
The main hero of our today’s experiments will be the most popular CPU in Core i7 family. It is the youngest solution – Core i7-920. The demand for it is extremely high due to its low price. For example, Intel’s official price list sets its MSRP at $284, while the next faster model in the family is already priced at $562.
We have already discussed the specifications of this processor before, so, we would only like to repeat them in a table below for your reference:
Intel Core i7-920
3 DDR3-800/1067 SDRAM channels
Cores / threads
Enhanced Intel SpeedStep
Intel Virtualization Technology
Intel Turbo Boost Technology
MMX, SSE, SSE2, SSE3, SSE4.1, SSE4.2
I would like to point out that the currently shipping mass production Core i7 processors have C0 stepping – the same processor stepping as those engineering samples that Intel sent us before the launch. However, mass production Core i7 CPUs that started selling in stores turned out to be dramatically different from the samples sent out to testers. Mass production Core i7 have unlocked multipliers for the memory and North Bridge built into the processor. As a result, systems built on mass production Core i7 CPUs allow clocking DDR3 SDRAM at frequencies over 1067MHz even without increasing the BCLK frequency over its nominal value. In other words, although the official specifications state only DDR3-800 and DDR3-1067 memory support, Core i7 processors in reality also support faster memory types. During our test session we could get our platform to work stably with the memory running as DDR3-1333, DDR3-1600 and even DDR3-1867 SDRAM.
For example, the screenshot from the Everest diagnostic tool you see below was taken in DDR3-1333 mode:
Note that the utility indicates 1.2V processor Vcore setting. It is currently a standard setting. All Core i7 processors we have checked out so far used the same exact setting.
Since we got our hands on a mass production CPU in a retail box, we would like to say a few words about its packaging and bundle. Core i7 ships in a blue cardboard box that is way larger than the boxes for quad-core Core 2 Quad CPUs. Nevertheless, the bundle remained unchanged. There is a booklet with installation instructions and a cooler.
The default cooler bundled with Core i7 processors has barely changed compared with the cooling solutions enclosed with LGA775 CPUs. It consists of a massive cylinder-shaped aluminum heatsink with a copper center and a 90-mm fan. The changes have also been made to the retention mechanism: it is now made of four plastic studs that lock into the mainboard retention holes.
This cooler is efficient enough to keep the CPU stable in nominal operational mode. However, during overclocking it can no longer do its job right and doesn’t let the processor reveal its frequency potential fully.
As for the CPU itself, it doesn’t look any different from the samples we tested before. The only evident difference is the serial marking on the CPU heat-spreader. Besides the Intel Core i7 trade mark and 920 serial number, it also states clock frequency, L3 cache size and QPI frequency. There is also a PCG mark (Platform Compatibility Guide) indicating electrical parameters. The identification S-Spec number of our processor unit was SLBCH, which is currently used for all mass production Core i7-920 CPUs.
Summing up everything we have just discussed, here is the list of components that were used to build our overclocker system for further testing:
The information we have just given you in the first part of our article suggests the best ways of overclocking Core i7 processors. The main idea behind it is to increase the base BCLK frequency that results into CPU clock frequency increase. However, since BCLK is also connected to a few other system frequencies, such as North Bridge built into the processor, memory and QPI bus, you should better use lower multipliers for Uncore, DDR3 SDRAM and QPI frequencies during overclocking. This way you will be able to better uncover the potential of your processor and will prevent your overclocking experience from stalling because some other system frequencies get overboard. As usual, you can certainly improve your overclocking results even more by raising the major voltages above their nominal settings, but make sure not to get too excited about it, at least without efficient CPU cooling in place.
We decided to find the maximum frequency our Core i7-920 processor will be able to hit without any additional voltage increase. For this experiment we locked the processor core voltage and Uncore voltage at their standard 1.2V in the ASUS P6T Deluxe mainboard BIOS. Note that using Auto setting is not recommended for these voltages, because in this case the mainboard will increase them on its own during overclocking without your knowledge or control.
To avoid any unexpected surprises during overclocking, we disabled EIST and Turbo Boost technologies and locked the CPU multiplier at 20x – the default setting for Core i7-920 with 2.66GHz nominal clock speed. We set 8x multiplier for the memory. Therefore, since Uncore frequency should be at least twice as high as the memory frequency, we used 16x multiplier for it. To get the desired QPI frequency setting, we used the lowest available multiplier of 18x.
With these settings our BCLK frequency reached 175MHz without any stability issues. By the way, we tested our system stability using 64-bit Prime95 25.7 utility in Small FFTs and Blend modes. This program proved to be the best tool for detecting over-overclocking and reported errors when most other popular stability check tools (including OCCT, LinX and IntelBurnTest) showed a pass.
As a result, our processor overclocked to 3.5GHz, which is a pretty good result considering it worked at its default 1.2V Vcore. The maximum core temperature during the stability tests never exceeded 74°C.
Of course, we can improve this result by raising the processor core voltage. However, we have to warn you against overdoing it, because voltage increase leads to higher heat dissipation, which ends up being the primary obstacle to higher overclocking results. Namely, in most cases you shouldn’t push the processor core voltage over 1.35-1.4V if you only use air cooling, because the CPU may get overheated without reaching its maximum frequencies.
However, we still couldn’t push the frequencies that much higher only by setting the CPU Vcore at 1.35V. Even though the system passed CPU stability tests with BCLK at 180MHz, it failed the memory stability tests. We may have reached the maximum overclocking for the North Bridge built into our processor, as it works at its own frequency connected to the BCLK and uses its own voltage setting. By the time the base frequency hit 175MHz, Uncore frequency already reached 2800MHz, which is evidently the best L3 cache can do at its default voltage. Therefore, we increased the Uncore voltage to 1.35V for further experiments. We have also increased CPU PLL voltage to 1.88V just in case.
These measures made our CPU stable at 190MHz BCLK. It overclocked to 3.8GHz and the frequency of its integrated North Bridge reached 3040MHz.
In this mode system passed all stability tests without problems, but further frequency increase resulted into Prime95 failure, and even additional processor voltages increase didn’t help. Looks like 3.8GHz frequency is the maximum our test Core i7-920 processor can do, even though the temperature of the hottest CPU core during our stability tests reached only 86°C, while the critical temperature limit is set at 100°C.
In fact, frequencies around 3.8GHz are the most widely spread maximum for Core i7-920 CPUs overclocked with air coolers. This is the conclusion we can make judging by our test results and basing on the feedback from the first users who purchased these processors. By the way, the engineering sample of the Core i7-920 CPU that we also have in our lab demonstrated the same overclockability. Even with the Vcore increased to 1.4V it overclocked only to 3.8GHz.
In this case we are talking specifically about the cores potential and not about the integrated North Bridge. To confirm if this was true we lowered the processor clock multiplier to 19x and got its running stably with the same voltage settings but higher 200MHz BCLK frequency.
Let’s sum things up now. We managed to overclock our Core i7-920 processor with the nominal 2.66GHz frequency to 3.8GHz. In other words, we got more than 40% clock frequency gain by using air cooling and potentially safe voltage settings for our processor. The recommended settings that were checked out on two different processors are given on the screenshot below (with an ASUS P6T Deluxe mainboard):
Note that among other things we owe our today’s success to high-quality memory that can work at 1600MHz frequency with 1.65V voltage setting. However, our memory could work at even higher speeds, up to 2000MHz. Unfortunately, our attempts to achieve higher speeds failed. When raising the memory frequency, we also had to increase the frequency of the North Bridge built into the processor. But to our great disappointment, it refused to work at 4GHz even after a serious increase of the Uncore voltage.
Let’s see what we ended up with here. Our today’s test session proved that new Core i7 processors overclock not any worse than their predecessors. We took the first mass production Core i7-920 CPU we came across and managed to increase its clock speed by 40% reaching 3.8GHz. We didn’t involve any special cooling solutions, just used a mass production air cooler. We also didn’t push the voltages to any dangerous limits.
This way we can say that Nehalem microarchitecture doesn’t limit the clock frequency increase in any serious manner. Today’s Core i7 CPUs are manufactured with 45nm production process using hafnium based high-k dielectric metal gates. This process makes it possible to raise the frequencies without any significant increase in CPU heat dissipation and power consumption. And it means that new generation processors will eventually become as popular among overclockers and enthusiasts as Core 2 Quad solutions.
Another important conclusion that we can make from the results of our today’s experiments is connected with the relative simplicity of Core i7 processors overclocking. Although these processors use several independent multipliers for various parts of the system and also use separate power supply for the cores and the built-in North Bridge, overclocking procedure is more than logical, so there is no room for any sort of misunderstanding.
The main success factors in Core i7 overclocking are evidently high-quality platform components. ASUS P6T Deluxe mainboard we discussed today turned out an excellent base for an overclocker system offering the most valuable overclocker qualities: excellent stability and predictability. The system memory is also of great importance. To ensure processor stability it has to work at relatively high frequencies without increasing its voltage to extreme heights. Kingston HyperX KHX16000D3K3/3GX kit coped perfectly with its task, so we have every right to recommend it to overclockers who decided to cast in their lot with Core i7 processors.
We are not going to stop our Core i7 overclocking experiments here. In one of the next articles we are going to discuss the performance of overclocked platforms using new generation processors and will try to answer the question what processors available in the today’s market would be the best choice for an overclocker.