Intel X48 based mainboards are already starting to sell in stores and get the users pretty confused. Is the new chipset worth our attention or it is hardly that much different from the Intel X38? Our review dedicated to a new Gigabyte mainboard will help answer this question.

The structure of Intel’s core logic solutions available in the market has always been very transparent and clear for the users. Within each chipset generation the company offered a basic discrete core logic set for mainstream systems, its enhanced modifications for high-performance market and a number of integrated models for inexpensive and budget systems. For example, at this time Intel P35 is a basic product, Intel X38 is positioned for the high-end systems, while the whole line-up of integrated solutions, such as G35, G33 and G31, are being offered for the budget segment. However, this well-balanced system we are all so used to may fall apart very soon, because the stores started to offer mainboards on a new core logic set – Intel X48. This product is not intended to simply replace the top Intel X38 chipset, but is positioned at an even higher level than the latter. As a result, the Intel chipset lineup will include not just a solution for enthusiasts – Intel X38, but also a solution for super-extreme enthusiasts – Intel X48.

We first heard the name of the new Intel chipset – X48 Express – last fall. However, it was pretty hard to figure out what would be so different between the new chipset and the Intel X38 one. On the one hand, the formal side to the picture is absolutely transparent. Intel web-site offers a detailed description of the new chipset, which indicates that the only distinguishing feature of the new Intel X48 will be the support of 1600MHz bus and hence DDR3-1600 SDRAM.

But on the other hand, the already existing Intel X38 based mainboards work with processors supporting 1600MHz bus, at this time represented only by one single model – Core 2 Extreme QX9770, without any evident problems. That is why we get the impression that Intel X48 is pretty mach a marketing product that cannot boast any advantages over Intel X38. Besides, Intel X38 and X48 North Bridges are pin-compatible, which backs up this statement additionally. And it means that mainboard manufacturers do not have to develop new mainboard designs and simply install the new core logic set on their already existing products.

However, there is also a different opinion. Some overclockers believe that Intel X48 is an improved solution with higher stability in extreme working conditions. That is why we can often hear that mainboards on the new chipset are more suitable for serious overclocking experiments. They claim that default support of 400MHz front side bus and hence DDR3-1600 SDRAM should ensure better results when the FSB and memory frequencies get raised beyond their nominal values.

So, our today’s article will help figure out what points of view are closer to the truth. Are Intel X48 based mainboards a real overclocker’s dream or the new chipset is just the result of Intel’s marketing efforts? This is the main question we are going to answer today with the help of Gigabyte GA-X48T-DQ6 mainboard based on Intel X48 chipset, which Gigabyte was so kind to provide for our tests.

Accessories Bundle and Specifications

We will start our discussion of the new Gigabyte GA-X48T-DQ6 mainboard features with a picture:

Frankly speaking, the board doesn’t strike you as an absolutely new product. Moreover, the components base and layout of Gigabyte GA-X48T-DQ6 reminds us of the Gigabyte GA-X38T-DQ6 on Intel X38 Express chipset. However, it would be not quite correct if we said that Gigabyte engineers didn’t work on the PCB layout at all. Yes, GA-X48T-DQ6 is very similar to the predecessor, however it features a revised cooling system as well as new processor, memory and PCI Express slots voltage regulator circuitry.

However, you can still find a mainboard with the same PCB layout as GA-X48T-DQ6 among numerous Gigabyte products. It is GA-EX38T-DQ6. Although to be fair, I have to say that it was announced after Gigabyte finished the design of the Intel X48 based solution. In other words, it would be correct to say that Gigabyte has an Intel X38 based solution in its line-up that uses a PCB layout of a more expensive Intel X48 based products, but not the other way around.

Considering that new and old mainboards on different chipsets turned out so close, you will hardly have any questions about the specifications of the new Gigabyte GA-X48T-DQ6.

Although the mainboard specifications are quite ordinary overall, its retail price is expected to be around $300. The reason for that is extremely high price of the new X48 chipset that Intel is currently selling to the mainboard makers for $70 a piece.

The mainboard comes in a traditional package for top Gigabyte products: it is a large golden shimmering box with the board inside and a smaller box with accessories.

Among the goodies are a common user’s manual, a bunch of leaflets and brochures, a CD disk with the software, I/O Shield for the case rear panel, four SATA cables, PATA and FDD cables. Besides, Gigabyte also included two brackets for the case rear panel with the total of four SATA ports and two standard Molex power connectors. Besides, there are also two pairs of cables that allow connecting regular SATA drives to these ports. There is also an additional bonus in the form of two small but extremely useful screws that allow transforming the GA-X48T-DQ6 cooling system from “crazy” into very convenient form.

PCB Design and Functionality

Frankly speaking, we review chapter devoted to peculiarities of the new Gigabyte GA-X48T-DQ6 mainboard could have been left empty. And it would have been absolutely fair to the board as well as to the reader. Since GA-X48T-DQ6 is practically identical to previously released Gigabyte solutions in terms of features as well as specifications, it doesn’t have any unique features to boast. Although there are a few not quite unique features that are still worth discussing here, so we will dwell on them in this part of our review. However, if you are very well familiar with Gigabyte mainboards on Intel X38 chipset, you can actually skip this section and move on to the next one.

First of all, I would like to say that Gigabyte GA-X48T-DQ6 is designed for dual-channel DDR3 SDRAM that can ensure better performance these days, especially during overclocking. This is the default memory type for Intel X48, but some manufacturers, including Gigabyte, also offer similar solutions with DDR2 DIMM slots, since this chipset, just like its predecessors boasts a universal memory controller. The set of supported dividers for memory frequency is also standard on GA-X48T-DQ6. So this mainboard, just like other mainboards on Intel’s “third series chipsets”, cannot clock the memory at any frequency that is more than twice as high as the FSB frequency. That is why the promised support of DDR3-1900 SDRAM that is posted on the mainboard box as well as Gigabyte’s company web-site is simply a marketing trick. In reality, this memory frequency can be achieved only if you overclock the FSB to 475MHz.

Since Gigabyte GA-X48T-DQ6 is positioned as a top of the line solution, it features two fully-functional PCI Express x16 slots compliant with 2.0 specification. These slots allow installing two graphics cards combined into a single Crossfire configuration. I would like to specifically point out that the graphics slots have been very conveniently located on the PCB. There is a significant gap between them, which will be enough for any cooling system. The remaining PCI and PCI Express x1 expansion slots will never be fully blocked in this case.

The chipset cooling system called Silent Pipe uses heatpipes and is designed according to pretty common contemporary schematics. One heatpipe connects a small heatsink on top of the chipset South Bridge with the heatsink on the chipset North Bridge. Another two heatpipes originate from the North Bridge heatsink and lead to the heatsink on processor voltage regulator MOSFET. However, some transistors from the CPU voltage regulator circuitry located above the LGA775 socket didn’t get under the Silent Pipe cooling system and are topped with individual miniature heatsink. However, you shouldn’t worry about it because they do not warm up that much during work. Moreover, the heatsink on the main MOSFET serves mainly not to cool down the processor voltage regulator, but to dissipate the heat generated by the chipset North Bridge and transferred from it via heatpipes.

Although the above described cooling system seems to be a pretty ordinary one, it is extremely efficient. The primary reason for that is the use of solid copper for heatpipes and heatsinks. They also paid special attention to ensuring proper contact between the heatsinks and the chips: they are all attached with spring screws. And finally, Gigabyte engineers ensured proper heat dissipation at the bottom of the PCB by placing a low-profile aluminum Crazy Cool heatsink plate right beneath the hot mainboard components.

However, as we have already pointed out in our previous articles, this heatsink may prevent you from installing some processor coolers properly, that is why you may need to remove it. It is great that Gigabyte GA-X48T-DQ6 allows doing it without any problems at all.

To implement some of the features the developers used a number of additional controllers. The mainboard is equipped with an IEEE1394 controller from Texas Instruments; two Realtek RTL8111B Gigabit network controllers that can work in Teaming mode; an additional Gigabyte SATA2 Serial ATA RAID controller supporting two additional ATA_300 and one PATA-133 ports. As a result, Gigabyte GA-X48T-DQ6 allows connecting very diverse additional devices. However, despite quite numerous connectors and chips on the mainboard PCB, it doesn’t cause any trouble during system assembly.

I believe that pretty much the only serious inconvenience is the fact that Clear CMOS contacts are located between the PCI Express x16 slots, right next to the battery. It will be really hard to reach them inside an assembled system with both graphics cards installed. Luckily, contemporary BIOS versions do not require frequent use of Clear CMOS, so this drawback is very unlikely to spoil the overall positive impression from Gigabyte GA-X48T-DQ6 mainboard. Especially, since all other connectors and pin-connectors have been arranged in a very smart manner: mostly along the lower edge of the PCB. You can see two connectors for four USB 2.0 ports, a connector for IEEE1394 port and even the connectors for COM and LPT that are close to oblivion these days. In the corner you can see six Serial ATA connectors, four provides by the ICH9R South Bridge and the remaining two, of different color, by the Gigabyte SATA2 additional onboard controller. The same controller is responsible for the PATA connector located nearby that has been conveniently turned parallel to the board. Another pair of SATA ports implemented via the chipset South Bridge is placed a little higher.

GA-X48T-DQ6 also supports FDD: the FDD pin-connector can be found to the right of the DIMM slots.

There is quite a lot of free room around the processor socket. Most super-coolers fit onto this mainboard easily. There are only two cases when you may encounter problems during cooler installation: if you are using memory modules of non-standard height or if the power supply panel of your system case is very close to the top edge of the mainboard PCB. The thing is that LGA775 socket on GA-X48T-DQ6 has been moved relatively close to DIMM slots and to the upper PCB edge, so please, keep this in mind.

The mainboard back panel also leaves a highly favorable impression. It bears a very diverse set of ports and connectors. There are eight USB 2.0 ports, two different IEEE1394 ports, two Gigabit network ports, PS/2 connectors for keyboard and mouse, optical and coaxial SPDIF outs and six analogue audio-jacks.

By the way, the integrated sound on GA-X48T-DQ6 uses an eight-channel Realtek ALC889A codec with the specified signal-to-noise ratio of 106dBA. I would like to point out that this codec supports DTS Connect technology that delivers multi-channel digital sound stream via digital audio-outs on the mainboard if there is an external compatible receiver available.

Processor Voltage Regulator Circuitry and DES

We put the discussion of the GA-X48T-DQ6 processor voltage regulator circuitry into a separate section on purpose. The thing is that even though it seems at first glance that this mainboard uses a traditional six-phase processor voltage regulator, typical of top Gigabyte mainboards, it is in fact completely different from the predecessors. This board’s processor voltage regulator circuitry, just like the one on a few other new solutions from the same manufacturer, supports DES (Dynamic Energy Saver) technology that should additionally reduce the processor power consumption with a few skillful tricks.

The schematics of the GA-X48T-DQ6 voltage regulator are pretty standard. It is based on Intersil ISL6327 micro-chip that Gigabyte has been using in its mainboards for a long time already. Six-phase regulator is built with high-frequency power MOSFET, which ensures not only longer life but also higher efficiency and lower operational temperature. Besides, they use highly reliable high-quality Japanese capacitors with polymer electrolyte for the voltage regulator as well as on the rest of the board. It is pretty funny that Gigabyte’s marking people are still trying to present the voltage regulator as a 12-phase one referring to twice as many inductance coils. But, don’t buy into it.

The main distinguishing feature of this voltage regulator is its ability to switch the number of active phases during work thanks to the ISL6327 functions. This is how Gigabyte is going to ensure that their mainboard will be more economical from the power consumption standpoint. The thing is that it makes sense to use more phases for a processor voltage regulator only if the power consumption is high: in this case six-phase circuitry will be more efficient and reliable and will generate higher quality signal. It is not economical to use a lot of phases under low workload as they may eat up more power. These are the reasons that drove Gigabyte engineers to design a circuitry with variable number of active phases: from 2 to 6.

The same concept is also implemented on ASUS mainboards, however they allow switching only between 4 to 8 phases. Gigabyte promises higher efficiency thanks to five different circuitry statuses. However, this is not the only difference between ASUS and Gigabyte approaches. While ASUS offers a fully hardware solution that doesn’t require any special software, Gigabyte’s circuitry works only when you launch Dynamic Energy Saver Utility. This utility monitors processor power consumption and uses this data to adjust the number of active phases in the processor voltage regulator in real time. If it is not launched, the regulator will work according to traditional six-phase algorithm.

The information window of the Dynamic Energy Saver Utility provides the user with data on the current processor power consumption, number of active phases and the actual energy savings statistics. I would like to add that you can also see how many phases of the CPU voltage regulator are active from the row of multi-color LEDs on the mainboard PCB located to the right from DIMM slots.

Besides the phase switching within processor voltage regulator circuitry, Dynamic Energy Saver Utility also does a few other things aimed at saving even more power. Firstly, this utility reduces the processor Vcore by 0.05-0.08V. There is a CPU Voltage Level switch that determines how greatly the Vcore will be lowered; however, unfortunately, you cannot skip it completely. Secondly, this utility can also enable processor throttling if you want to, so that the CPU will skip every other clock cycle under low workload.

As a result, we can conclude that DES technology first of all saves power in idle mode. In case of full CPU utilization, throttling will be disabled and the voltage regulator will activate all six channels. The only way to save power in this case is the forced lowering of the processor core voltage, which is more of a trick than an engineering approach.

To prove everything we have just said, we measured the power consumption of a quad-core Core 2 Extreme QX9770 processor in case of different DES settings. Enhanced Intel SpeedStep was activated.

As we see, DES technology does reduce the processor power consumption and power losses in the voltage regulator circuitry. The obtained results prove this statement with all certainty. But do not forget that the mainboard drops the CPU Vcore below the nominal value, which may lead to potential loss of system stability and overall, can hardly be considered a technologically fair approach. The same thing can actually be done on any mainboard that allows manual adjustment of the processor Vcore parameter. Unfortunately, it is impossible to estimate how efficient changing the number of active phases in the processor voltage regulator circuitry in real time actually is, because once DES is launched, the processor Vcore inevitably drops.

So, the current DES version is pretty interesting from the theoretical prospective, but its practical implementation still arouses a number of serious questions. Firstly, the processor Vcore drops below the nominal regardless of the user’s wishes. Secondly, DES requires installation and non-stop operation of a special utility. And thirdly, this technology doesn’t work with overclocked processors.

In other words, DES will hardly be of interest to computer enthusiasts at this time. at least until Gigabyte engineers make a few changes to it.

BIOS Functionality

Gigabyte GA-X48T-DQ6 mainboard was tested with BIOS version F4 dating back to March 6, 2008.

BIOS Setup interface of Gigabyte GA-X48T-DQ6 mainboard is very similar to that of some other Gigabyte mainboards released recently. And it is actually not surprising at all, the developers created their brand name style, put everything together for maximum user comfort, gave up “hidden” options, added all necessary utilities and now the Gigabyte BIOS Setup became very functional, simple and easy to work with.

Almost all the options dealing with major system parameters configuration are gathered in a special BIOS section called MB Intelligent Tweaker (M.I.T.). This particular section will be of primary interest to overclocking fans.

Here you can set the processor multiplier and FSB frequency from 100 to 700MHz with 1MHz increment. You can also adjust dynamic CPU overclocking depending on its utilization using C.I.A.2 technology. Note that the BIOS supports 45nm processors and allows changing the multiplier setting with 0.5x increment.

BIOS Setup of Gigabyte GA-X48T-DQ6 also offers a technology for automatic graphics card overclocking called Robust Graphics Booster.

The memory subsystem frequency is set with special dividers: just like on the solutions using previous generation Intel chipsets. Gigabyte GA-X48T-DQ6 offers a complete set of FSB:Mem coefficients for Intel X48. Note that you also set FSB Strap frequency together with the divider. It is represented as a remark to each divider in the corresponding menu. It is nice to see the current and set memory frequency displayed right next to the FSB:Mem setting option, which makes configuring it a lot easier.

There is Performance Enhance menu right next to the option for DDR3 SDRAM frequency setting. You can choose one of the following: Standard, Turbo and Extreme. This option affects the major chipset speed parameter, a latency called Performance Level. However, you should remember that it only works if the memory timings are configured automatically.

Gigabyte GA-X48T-DQ6 BIOS allows adjusting the following memory latencies:

When you configure the settings, the mainboard displays all parameter values and allows setting any of them to Auto, which makes it a lot easier for inexperienced users to work with the BIOS Setup.

I would also like to point out that Static T_READ Value parameter has no direct connection with the memory timings. This is what Gigabyte mainboards BIOS calls the Performance Level, which sets the latency for the chipset North Bridge used to synchronize the FSB and memory bus frequencies. In other words, GA-X48T-DQ6 allows advanced users to adjust Performance Level manually, which makes this board an ideal solution for overclocking.

As for the voltage adjustment, the board supported the following settings:

All voltage settings except processor Vcore are provided as relative values. Actual voltages can only be set for the CPU. There is a very convenient information field showing the default Vcore of the system processor right next to it.

Besides everything we have already mentioned, Gigabyte also offers Loadline Calibration option that allows reducing the drop of the CPU Vcore under heavy workload. For example, when we overclocked quad-core Core 2 Extreme QX9770 processor to 4.0GHz by raising its Vcore to 1.4V, Vdroop on our board reached 0.088V under maximum workload and 0.04V in idle mode. Enabling Loadline Calibration helped ensure lower voltage drop of 0.056V and 0.024V respectively. I can’t say that it is a good result, but it is still better than nothing. Especially since in the past Gigabyte engineers used to disregard this useful option completely.

Besides the most important overclocking-friendly MB Intelligent Tweaker (M.I.T.) section, there are a few other pages of the BIOS Setup worth your close look.

For example, PC Health Status page where hardware monitoring functions are concentrated.

Note that the board allows adjusting the rotation speed of the processor fan depending on the CPU temperature.

Processor technologies can be configured in the Advanced BIOS Features section:

Gigabyte engineers also made sure that you can save your settings profiles. The board allows saving up to 8 profiles. All profiles when the board booted successfully will be saved automatically.

When you overclock, the board normally doesn’t require using Clear CMOS jumper. If it cannot restart after you changed some of the settings, the MB Intelligent Tweaker (M.I.T.) settings get reset to defaults. However, unfortunately, the user received no notification about it and the board simply continues booting with non-optimized parameters.

Also, GA-X48T-DQ6 boasts a built-in Q-Flash utility that allows updating the BIOS without booting the OS.

Moreover, Gigabyte GA-X48T-DQ6 mainboard supports Dual BIOS technology. Therefore, there is an additional Flash memory chip with a BIOS copy on the mainboard PCB that will save the day if the main BIOS dies.

Overclocking Experiments

You have every right to expect mainboards on Intel X48 chipset to be a little better at CPU overclocking than their predecessors. Since this chipset supports higher FSB speeds, up to 400MHz, in nominal mode. However, only practical experiments will show if this is true or not.

To check out the overclocking potential of the new Gigabyte GA-X48T-DQ6 mainboard we put together a special testbed that included the board, of course, 2GB DDR3-1800 from Cell Shock (CS3222580), OCZ GeForce 8800GTX graphics card, Western Digital Raptor WD1500AHFD HDD and SilverStone SST-ST85ZF power supply unit. The CPU was cooled with Scythe Infinity cooler.

First of all, we decided to determine the maximum FSB frequency when the mainboard would run stably with a dual-core processor. We used a 45nm Core 2 Duo E8500 with 3.16GHz nominal speed (9.5 x 333MHz). The stability was checked with a standard one-hour run of OCCT Perestroika 2.0.0a application.

Practical experiments showed that our system can run stably with a dual-core CPU at 450MHz FSB without any special tricks on our end. If the CPU and the memory can work in this mode the mainboard will definitely support it. You won’t even have to increase the North Bridge voltage.

When the CPU clock frequency multiplier is lowered to 8x, we could hit a more important psychological maximum of 500MHz. However, in this case we had to push the FSB and North Bridge voltages 0.3V up to ensure stability of our test system.

The record-breaking FSB frequency with a dual-core processor overclocked on GA-X48T-DQ6 mainboard equaled 525MHz. It was achieved by raising the FSB voltage 0.35V higher and increasing the North Bridge voltage by 0.4V at the same time.

In this case the system remained absolutely stable. The CPU worked at 4.2GHz frequency with Vcore set at 1.4V. Memory ran with 1.9V voltage setting and 7-7-7-20 timings at 1680MHz.

I would like to say that if you use different dividers to set the DDR3 SDRAM frequency at a lower value, the workload on the chipset North Bridge will reduce dramatically. As a result, you will be able to achieve system stability without any significant NB voltage increase. In our case when the FSB:DRAM multiplier was set at 5:8, we had to resort to one more trick to ensure stability: we manually raised Static T_READ Value (Performance Level) to 12, which has some negative effect on the memory subsystem performance.

So, it is not quite correct to claim that mainboards based on the new Intel X48 chipset boast significantly higher FSB overclocking potential than those on Intel X38. However, we may be able to reveal certain improvements during quad-core processors overclocking. Let’s check it out now. For the next round of experiments we upgraded the above described testbed based on Gigabyte GA-X48T-DQ6 mainboard with a Core 2 Quad Q9300 processor from the Penryn family that works at the nominal speed of 2.5GHz (7.5 x 333MHz).

As we know, quad-core processors overclocking by raising the FSB frequency stumbles upon first problems much sooner than dual-core processors overclocking. That is why it is not surprising that we had to really fine tune the BIOS parameters to ensure that our GA-X48T-DQ6 will run stably at 450MHz FSB with Core 2 Quad Q9300 processor. To pass the stability tests we had to increase the FSB voltage to its maximum, i.e. by 0.35V, while the NB voltage was set 0.25V higher. Only with these settings Core 2 Quad Q9300 remained stable at 3.37GHz frequency.

Further overclocking was even harder to perform. To get our testbed to work at 460MHz FSB, the chipset North Bridge voltage had to be set 0.3V higher. And at 470MHz FSN the system stopped booting the OS with any settings.

As a result, the maximum overclocking for a quad-core Core 2 Quad Q9300 processor on Gigabyte GA-X48T-DQ6 mainboard was the modest 3.45GHz frequency.

So, looks like Intel X48 chipset used for Gigabyte GA-X48T-DQ6 mainboard failed to improve quad-core CPU overclocking. You can get pretty much the same results for quad-core CPUs on older mainboards using Intel X38 or even Intel P35 chipsets.

Benchmark Results

We compared the performance of Gigabyte GA-X48T-DQ6 mainboard on the new Intel X48 chipset against that of widely spread platforms on Intel X38. Since GA-X48T-DQ6 supports DDR3 SDRAM, we selected ASUS P5E3 Deluxe to be its main competitor in this test session.

Testbeds were configured as follows:

• CPU: Intel Core 2 Duo E8500 ((LGA775, 3.16GHz, 1333MHz FSB, 6MB L2, Wolfdale).
• Mainboards:
  • ASUS P5E3 Deluxe (LGA775, Intel X38, DDR3 SDRAM);
  • Gigabyte GA-X48T-DQ6 (LGA775, Intel X48, DDR3 SDRAM).
• Graphics card: OCZ GeForce 8800GTX (PCI-E x16).
• HDD: Western Digital WD1500AHFD (SATA150).
• OS: Microsoft Windows Vista x86.

Performance in Nominal Mode

The first series of tests were performed with the processor working at its default speed of 3.16GHz set as 9.5 x 333MHz. The memory frequency in this case was set at 1333MHz with 7-7-7-20 timings.

As usual, we will first check out synthetic benchmarks testing the memory subsystem performance. This parameter is crucial for the performance of mainboards for Intel processors. For our tests we used Lavalys Everest 4.50 utility.

Although we made sure that primary timings, Command Rate and Performance Level were the same on both test platforms, ASUS and Gigabyte mainboards configured secondary timings differently, which led to different results of the memory subsystem tests. Nevertheless, even now we can already state that Intel X48 cannot ensure higher performance than Intel X38. The memory controller in Intel’s “third series” chipsets is so well optimized that there is simply no room left for further improvement. That is why we can only expect Intel based platforms to demonstrate any performance growth when they launch new Nehalem processors with principally new memory controller built into the CPU core.

However, before we make any final conclusions, let’s check out the situation in complex benchmarks and real applications.

The results indicate that in most cases Gigabyte GA-X48T-DQ6 outperforms ASUS P5E3 Deluxe. However, don’t get too excited just yet. The thing is that it is not the superiority of the new Intel X48 chipset over Intel X38, but the fact that Gigabyte mainboard doesn’t set the nominal FSB frequency quite fairly using 335MHz instead of 333MHz.

Performance during Overclocking

Besides the tests in nominal mode, we would also like to compare the mainboards’ performance in overclocked systems. The thing is that relative performance of overclocker platforms is very often different from what we see in nominal mode.

For the second round of tests we decided to set the FSB frequency at 450MHz. We used the same Core 2 Duo X8500 processor overclocked to 4.275GHz set as 9.5 x 450MHz. The processor Vcore was increased to 1.4V to ensure better stability.

DDR3 memory was running at 1800MHz and the timings were set at 8-7-7-20.

During our overclocking experiments with Gigabyte GA-X48T-DQ6 we discovered one very interesting thing. In the above described test mode, this mainboard automatically sets Performance Level to 7 while ASUS P5E3 Deluxe uses default Performance Level = 8.

This could be the shaky advantage of Intel X48 over Intel X38: newer chipset can work with lower Performance Level settings. However, you can make ASUS P5E3 Deluxe work with Performance Level = 7, although you will have to increase the NB voltage quite noticeably in this case to ensure stability: from the default 1.25V to 1.7V. Gigabyte GA-X48T-DQ6 supports Performance Level 7 at 450MHz FSB and default NB voltage.

Unfortunately, even increased NB voltage didn’t help get Gigabyte GA-X48T-DQ6 to work with Performance Level 6. Therefore, we compared the performance of our testing participants’ with FSB overclocked to 450MHz in two modes for ASUS P5E3 Deluxe and only in one mode for Gigabyte GA-X48T-DQ6.

Let’s start with synthetics:

As we see, Performance Level parameter is not a determinative here although it does indeed have a serious effect on the results. As you can see, ASUS P5E3 Deluxe outperforms Gigabyte GA-X48T-DQ6 in read speed and latencies even with higher Performance Level setting. However, despite this fact we can’t help pointing out significant improvement: unlike previous Gigabyte’s solutions, GA-X48T-DQ6 doesn’t increase Performance Level during overclocking and demonstrates quite acceptable results.

Let’s check out the results obtained in complex tests and real applications:

The numbers speak for themselves. Overclocked Gigabyte GA-X48T-DQ6 shows very good speed in real applications. From the performance prospective, it can compete successfully against ASUS P5E3 Deluxe even with manually lowered Performance Level.

Conclusion

It turns out that despite mainboard makers’ desires, Intel X48 based solutions cannot be regarded as really new products. The chipset doesn’t boast any significant advantages over Intel X38. as for the 1600MHz bus support that has been formally introduced in the new Intel X48, it is a pure marketing trick, since even older mainboards on Intel X38 can work just fine with the only CPU supporting this bus – Core 2 Extreme QX9770.

Moreover, Intel X48 features the same memory controller as Intel X38. It means that mainboards based on these two chipsets will only differ due to settings in the BIOS Setup of particular mainboards. Don’t expect new boards to work any overclocking wonders. According to our tests, maximum FSB frequencies during overclocking depend on the PCB layout and processor overclocking potential and not on the chipset. The only arguable advantage of the new Intel X48 that we still need to investigate more in our upcoming articles is the more flexible Performance Level adjustment, which may potentially ensure better performance of mainboards based on this core logic set.

Nevertheless, Intel X48 Express based mainboards are worth your attention since in most cases they are better finalized and enhanced modifications of the previous generation boards on Intel X38. Although, you will have to pay the price, as these solutions cost considerably more than their predecessors.

Everything we have just said is true for the particular mainboard we discussed today – Gigabyte GA-X48T-DQ6. In fact, this solution is not very much different from Gigabyte Intel X38 based products that have been available in the market for a while. The main improvements include redesigned processor voltage regulator circuitry with DES technology and better BIOS optimization. And while the new “economical” voltage regulator may not be very the practical yet for computer enthusiasts, the BIOS of the new GA-X48T-DQ6 is definitely a significant step forward and Gigabyte engineers surely have to be given proper credit for that. The biggest advantage of the new mainboard’s BIOS is no performance drop during overclocking that was so frustrating for many overclocking fans who used to work with Gigabyte platforms in the past.

As a result, Gigabyte GA-X48T-DQ6 can be considered a very good choice for a high-end system.

We often get a lot of questions about CPU overclocking techniques and tools. So, how do we overclock processors? Hopefully our new article will answer most of your questions.

Everyone can feel confused in a strange situation. When you find yourself in an organization for the first time, you don’t know where to go and whom to talk to. When you sit down at the steering wheel for the first time, you can’t make out what to do next with the vehicle. You don’t know what to do when you first turn on your PC or enter the Internet. You’ll get the experience over time and go right to the elevator or turn on the ignition or perform the habitual sequence of actions on your PC almost subconsciously. But you need a guide at first and this role is going to be played by this article.

Why do you want to overclock, anyway? I guess there are four types of overclockers.

There are beginner overclockers. The beginner has already got a PC and has no choice of the configuration. He has to deal with what he’s got already.

There are thrifty overclockers. They want to have as much performance as they can yet spend as little money as possible. The PC is assembled out of simple, cheap and outdated hardware parts the person can afford. The default performance of such a PC is below necessary level, but it can be usually overclocked to a more or less acceptable level. A thrifty overclocker is not necessarily a poor person. There are just a lot of other values in our lives besides computers. You may want to invest your money into your education, family or recreation instead of your PC.

Experienced overclockers have somewhat different objectives. Their goal is to get maximum performance and fun without overpaying for that. It’s silly to waste your money purchasing senior (i.e. expensive) components, yet it is also silly to limit your own opportunities by saving on trifles. This overclocker considers numerous factors as he picks up every component. He wants an overclocker-friendly mainboard, a highly overclockable CPU, and a quiet yet efficient cooler. The resulting performance of his PC will be very high, comparable to or better than the default performance of a system assembled out of top-end components. Although such an extremely high performance may not be vitally important, the overclocker finds his pleasure in the feeling of satisfaction from a job well done.

And finally, there are PC enthusiasts whose goal is to get the highest possible performance at any price. They use everything – top models of components and extremely low temperatures – to find themselves on a breathtakingly high peak, unattainable by the crowd. It’s a kind of sport where you can enter the top ten, five or three and that’s the only reward a PC enthusiasts needs.

Of course, these categories are not sharply outlined. There are no definite landmarks between them. Beginners become experienced overclockers over time while experienced overclockers may go in for extreme experiments. You can meet such an exotic combination as a thrifty enthusiast even!

Anyway, you have to begin with something, and I will begin with the first and most important point.

Theoretical Issues

Don’t skip this section! I realize it’s boring and uninteresting to search for and digest information. You just want to know the magic buttons you should press in order to get the necessary result – an overclocked PC. But such universal buttons do not exist! They are different in every situation and you have to have some knowledge in order to find them.

After all, if you don’t want to do any overclocking, you shouldn’t probably be reading this at all. And if you do, you will anyway have to learn everything from your own experience sooner or later, so why now learn it right now? You don’t want to pay for your knowledge with burned-out or damaged components and with wasted time and money when there is information waiting for you to apply it to your purpose.

Collecting your system info

First of all, you need to know what you will be dealing with. If you have assembled your PC with your own hands or at least taken part in choosing the configuration, you should already know what components it consists of. If you don’t, you should identify each component first. Learn your PC and its parts, browse through your mainboard manual. Run informational programs and a few performance tests and write down the technical characteristics, temperatures and voltages under load and in idle mode. This information is going to be helpful afterwards. Knowing the exact configuration of your system, you can make a precise enough guess at the possible level it can be overclocked to. The performance data will show you how your system performance has increased after overclocking. Sudden fluctuations of the voltages and temperatures are indicative of dangerous situations you can identify and avoid. With these preliminary tests you can also make sure that your system is stable in its normal operating mode.

List of useful software

Every overclocker has a wide range of software tools that fall into a few groups:

• Informational/diagnostic
• Monitoring
• Overclocking
• Stability check
• Performance benchmark

There is no sharp separating line between the categories. Informational utilities can often benchmark performance while monitoring tools can overclock as well.

Informational & diagnostic utilities can accurately identify your system configuration. The two most advanced suites are Lavalys Everest and SiSoftware Sandra. Their capabilities are not limited to reporting the configuration, though. These software suites can perform monitoring functions, benchmark performance, and test the system for stability. However, you don’t have to run these voluminous suites, especially as only a part of their functionality is provided for free. There are many less known programs of the same type, e.g. WinAudit or PC Wizard. Instead the universal suites you can use a set of small, free-of-charge special-purpose utilities. For example, overclockers make wide use of the CPU-Z utility that reports information about your CPU as well as mainboard and memory. For more control over the memory timings you can use the MemSet program.

The best of universal monitoring tools have always been developed by enthusiasts, by independent developers. Unfortunately, this is the reason why they are usually short-lived projects. MBProbe was the first abandoned program and then we lost Motherboard Monitor. Today, SpeedFan is the most popular monitoring tool.

The CPU should better be overclocked using BIOS options. If you can’t find them in your mainboard’s BIOS, you can try the universal Windows-based ClockGen program. You can also look up on the CD enclosed with your mainboard for any exclusive program that can overclock from Windows, control the fans, and provide monitoring options.

The range of programs for overclocking graphics cards is broad as well. RivaTuner is perhaps the best in its class, but you may also want to use PowerStrip, NiBiTor, ATI Tray Tools, ATI Tool, etc.

Neither program can give you a 100% guarantee of stable operation of an overclocked CPU but your chances grow up dramatically if you use two or three such utilities. The stability check can be performed with OCCT, S&M, Prime95 or any other program that can load your PC heavily. You can even run your favorite 3D game for that.

As for performance benchmarks, there are hundreds of them that can test the system at large as well as single components. The BenchmarkHQ website offers a long list of useful utilities.

Overclocking related materials

We touch upon overclocking issues in most of our reviews. If you are regularly reading our news and reviews, you should have already accumulated enough knowledge to guide you in your practical experiments. The CPU overclocking statistics can give you an idea of what you can achieve. You can also refer to our forum for discussions of overclocking related problems.

You can also refer to our forum for discussions of overclocking related problems.

You can read one of the latest articles called CPU Overclocking Guide. And you shouldn’t ignore an article if it doesn’t mention your specific CPU model. There are universal overclocking basics and if you’ve got a clear understanding of how to overclock a Pentium III, you will easily overclock any modern CPU.

Problem of Choice

If you are a beginner overclocker and you’ve already got a PC, it’s both good and bad. It’s bad because you can’t change anything while even one weak component, e.g. a low-wattage power supply, can be an insurmountable obstacle to overclocking the whole system. It’s good because you don’t have to face the problem of choice.

This is one of the most difficult decisions you have to make when putting together your overclocking-ready system. There are thousands of factors to be taken into account like the range of currently available components, the comparative worth of different models, the pricing, the ease of assembly, upgrade opportunities, and even the exterior design. It is rather simple and easy to overclock a PC, but it’s very hard to pick up an optimal combination of hardware parts.

Fortunately, discussing this goes beyond the subject of this article. I’ll only touch upon it again when the choice of parts will determine the choice of an overclocking method.

CPU Overclocking Basics

Overclocking means clocking a hardware part at a frequency higher than the default one. You don’t have to know why overclocking is possible at all. It may be due to a high margin of safety provided by the manufacturer, due to marketing reasons that made the manufacturer set lower default characteristics than possible, or due to the use of faster components than necessary. Whatever the reason, your task is to make good use of the offered opportunity.

In a PC, everything is standardized and synchronized. The standardization is obligatory for components from different makers to be able to work together at all. The synchronization ensures that the components work together smoothly. The frequency of the system bus (or Front Side Bus – FSB) is considered the basic frequency of the system. The rest of the buses the various devices and components are connected with usually work at lower frequencies that are generated from the FSB frequency by means of divisors. The CPU frequency is currently much higher than the FSB frequency and is generated by means of a multiplier.

For example, the Intel Core 2 Duo E6300 processor works at a bus frequency of 266MHz. Its frequency multiplier is x7 and the multiplication of the two yields the resulting CPU frequency, which is 266×7=1.86GHz. It means that the CPU frequency can be increased by increasing the FSB frequency or the multiplier.

Senior models of modern CPUs have an unlocked frequency multiplier and permit to increase it, but such CPUs are too expensive in comparison with junior models from the same family. It is not reasonable to buy one because overclocking can lift the performance of the junior models up to the level of the senior ones and even higher.

Thus, CPU overclocking usually boils down to increasing the FSB frequency. Taking the same Intel Core 2 Duo E6300 as an example, if you increase its FSB frequency from the default 266MHz to 400MHz, the CPU frequency will grow up to 2.8MHz, by 1000MHz almost. If the FSB frequency is increased to 500MHz, the CPU will be clocked at 3.5GHz, etc. This is in fact all the information you need to go into your mainboard’s BIOS Setup, increase the FSB frequency and overclock your CPU, but there are some things you should be aware of yet. You’ll learn more of them eventually, and some of them are unknown even to me because new CPU models have peculiarities of their own. But again, there are basic things you can rely upon in all your overclocking attempts.

Getting Ready to Overclock

Before you begin to overclock your CPU, you should do some preparatory work. Check for BIOS updates on your mainboard manufacturer’s website and read through the list of changes. There have been numerous occasions when a mainboard with poor overclockability would transform miraculously after a BIOS update. New BIOS versions not only correct errors, but sometimes add new parameters or expand the range of the available ones. You can read the version number of your current BIOS during the startup. If the information disappears too quickly, press the Pause key on your keyboard. Sometimes the version number is shown in BIOS Setup. You can also learn it from informational utilities or special-purpose BIOS updaters. You don’t have to write every BIOS version, from the oldest to the newest, into your mainboard. The freshest version includes everything from the earlier ones. And even if the latest version is not the most optimal for overclocking, it is at least sure to be free from the errors of the earlier versions.

So, you are in your mainboard’s BIOS and want to know what to do next? Perhaps you’ve got an “intelligent” mainboard that can do everything by itself, and you only have to specify the desired level of CPU overclocking or the FSB frequency. However, it’s better to take care of everything by yourself to avoid any problems. This will save your time and components and will ensure the best possible result.

You should reduce the memory frequency first. As I said above, everything is interrelated in the PC, so when you are overclocking and increasing the FSB frequency, the memory frequency increases proportionally. And if the memory works at a high frequency by default, it may become the limit to further CPU overclocking. It’s desirable to set the minimum possible memory frequency in the BIOS. Don’t worry about the performance reduction. You’ll increase it back through overclocking. Moreover, you can return to the memory after you’ve found the maximum frequency of your CPU and try to increase it speed back again.

Next you should set higher memory timings, at least the basic ones, for example 5-5-5-15-2T for the widespread DDR2 memory type. You should do this for the same reason as you reduced the memory frequency, i.e. to prevent the memory chips from interfering with the CPU overclocking. Memory can work at a high frequency with high timings or at a low frequency with low timings. Sometimes two or more combinations of possible settings are written into the memory module’s SPD chip. The reduction of frequency can be taken as a permission to reduce the timings if the latter are set up by the mainboard automatically. This combination of a low frequency and low timings may be quite normal for the default operating mode of the CPU, but low timings may become a problem when the memory frequency has increased due to your overclocking attempts.

If BIOS parameters are set at Auto, the mainboard can control them automatically. Mainboards control their parameters normally most of the time, but not always. That’s why you should avoid this and specify parameter values explicitly.

For example, you may want to fix the CPU multiplier at its default value. I know of cases when an intelligent mainboard’s BIOS reduced the multiplier at startup. Perhaps it was an error, but anyway.

Moreover, you may also want to explicitly specify the voltages so that the mainboard didn’t increase them at overclocking. The memory voltage should, on the contrary, be increased a little to make the memory chips stable. By the way, it’s sometimes not easy to learn the default voltage values. Many mainboards show the CPU voltage in a special information line. Sometimes the default voltage is the minimum possible one. The CPU voltage can also be read with a special program like CoreTemp or RM Clock.

You can also try to determine the voltage by a guess method. The mainboard normally selects the default CPU voltage, and you should view it with some monitoring utility or in the PC Health section of the BIOS. Then specify a voltage value explicitly, trying to guess it so that it coincided with the value measured when it was set automatically.

The Spread Spectrum feature should be disabled unless the mainboard disables it automatically when overclocking. This option is meant to minimize electromagnetic interference the operating computer generates but it may limit your system’s overclockability.

Some mainboards are able to overclock the graphics card in automatic mode. If the graphics card is under heavy load, its frequencies are increased a little. You should disable this feature. You can’t get a big performance increase this way, but problems are quite possible.

Overclocking the CPU

Now you know enough to try to overclock your CPU. The step-by-step instructions are simple: increase the FSB frequency in the BIOS, save the settings, boot up the OS and test your PC for stability while keeping your eye on the temperature reading. If you haven’t ever seen a BIOS screen in all your life and can’t find the necessary settings, browse our CPU Overclocking Guide. The frequency can at first be increased with a big enough step like 50MHz or even 100MHz, depending on your CPU model. You should have learned the frequency peak of your CPU and set the frequency accordingly, but the potential of a given sample may differ greatly from the average value. Then, reduce the step to 20, 10 or even 5MHz. Overclocking with a 1MHz precision only makes sense when you are trying to set a new record. For everyday operation, it’s better to have a reserve of stability to safeguard yourself against the natural fluctuation of such characteristic as temperature and voltage.

You go on increasing the frequency while the system is stable and passes all the tests. When errors occur, you reduce the frequency and thus find the limit for your particular CPU.

Can you overclock more? Yes, but you have to increase the voltages for that.

Do you have to increase the voltage?

It’s hard to give a short answer to this question. You should first decide what voltage needs increasing. This is verified through experimenting. Try to increase the CPU voltage by one or two minimum steps in the BIOS and then check out if the CPU can now work at a higher frequency than before the voltage increase. If it can, begin to search for the frequency limit under the new conditions. If it cannot, you’ve increased the wrong voltage.

Besides the core voltage, the mainboard itself can limit further overclocking if you have set a high FSB frequency. Try to increase the voltage on the chipset’s North Bridge a little. Try to change voltages in combination, e.g. increase the FSB termination voltage if your BIOS offers this option. You have fixed all the voltages at their defaults before overclocking, but now you can get a hint from your mainboard: set the voltages to Auto and see what ranges they are changing in.

How high can you increase the voltages? There are three factors that can stop you: the mainboard’s capabilities, a too high temperature, and a lack of purpose. If your system reacts eagerly to a voltage increase and the temperatures remain within normal ranges, why not continue? But if you need to increase your CPU voltage by 0.3V to overclock by 100MHz, I don’t think it makes much sense. The effect from such a small frequency increase won’t be conspicuous for a modern CPU but your system will be put under a stress and the temperature will grow up, too. The CPU temperature increases along with its frequency, but it does so at a much faster rate when you increase the core voltage.

What temperature is normal?

A CPU temperature of 40-50°C is normal. It can grow up to 60°C under load but you should avoid a temperature of 70°C and higher. Replacing the cooler is not the only option to decrease the temperature. If your PC is rather old, you may try to reinstall the cooler and update the thermal grease and the temperature should go down considerably. The temperature will be growing up inevitably in a small and poorly ventilated system case. Installing system fans should help.

Talking about temperatures I mean the CPU temperature in the first place, yet it is not the only one that you should keep your eye on. Keep track of the chipset temperature, especially if you have increased its voltage. The thermal sensor is built into the North Bridge in Intel’s new chipsets. So far there is no program that can monitor this temperature, but such software will surely come out soon.

Mainboards can regularly monitor two temperatures, CPU and system. The system temperature is not the chipset temperature. There is a thermal sensor on the mainboard, usually located near the I/O chip (from Fintek, ITE or Winbond), and its reading is shown as the system temp. Depending on the exact location of the sensor, this temperature may be an important indicator or may be quite useless (and even not changing at all).

Note also the temperature of the MOSFETs near the CPU, especially if you use a liquid cooling system. These usually become very hot under load, but few makers of liquid cooling solutions provide any means to cool them. Memory modules remain almost cold even after a considerably voltage increase but become hot when the memory chips are being accessed intensively.

Do you have to decrease the multiplier?

There is yet another way to increase your system performance. Almost all of modern CPUs allow decreasing the frequency multiplier. You can decrease it and increase the FSB frequency appropriately, keeping the resulting CPU clock rate unchanged. The FSB frequency affects the overall system performance. The higher it is, the faster data flows throughout the PC. Thus, a 3GHz CPU working at a 300MHz FSB with a x10 multiplier will generally be faster than the same 3GHz CPU working at a 200MHz FSB with a x15 multiplier.

This safe and free method of lifting the performance of your system a little more won’t suit everyone, however. The fact is, the power-saving technologies implemented in CPUs stop to work when you change the multiplier because they are themselves based on the reduction of the multiplier and voltage in idle mode. Such technologies play an important role in reducing power consumption and temperature, so this method will only suit people whose PC is loaded all the time, e.g. with distributed computing software. It’s going to be indeed free for them as they get an increase of speed without losing in anything.

Overclocking Specifics for Intel Core CPUs

CPUs with the Core micro-architecture are the most appealing nowadays. They overclock well, so I’ll pay them most of my attention here.

An annoying thing you have to keep in your mind as you are overclocking a Core processor is the so-called FSB Wall. This term refers to the maximum bus frequency this sample of the CPU can work at. Thus, it is convenient to begin to overclock a Core processor with finding its FSB wall. To do this, reduce the multiplier to the minimum of x6 and see what FSB frequency your CPU can be overclocked to. Your CPU may be not stable at this frequency with the default multiplier, but at least you get a notion of what your CPU can potentially do.

For example, CPUs with a default bus frequency of 200MHz can be but seldom overclocked to over a 400MHz FSB. You should take this fact into account when you are choosing your CPU. There is no sense in overpaying for a senior and more expensive model if you can take a junior one and overclock it but note that the top frequency of a junior CPU with a default multiplier of x8 is likely to be limited by the FSB wall at 3.2GHz or something. Most likely, you will stop at 3.0-3.1GHz, which is low. You don’t want such limitations, do you? So, you may want to consider a CPU with an x9 multiplier instead.

As for CPUs with a default FSB frequency of 266 or 333MHz, people often buy a junior model with a multiplier of x7, but such CPUs may be limited not only by the FSB wall but also by the capabilities of your mainboard and memory. It is better to use CPUs with an x8 or higher multiplier, yet you have to face a new problem with them, the FSB Strap.

The FSB Strap is not a feature of the CPU, but of the mainboard and chipset. It is the frequency the chipset switches into another operation mode at, increasing the latencies and bringing about a performance hit. Gigabyte mainboards on the Intel P965 Express chipset slow down as soon as you try to overclock the CPU. ASUS mainboards on the same chipset deliver excellent performance up to a 400MHz FSB and then use the FSB strap. When testing an ASUS Striker Extreme mainboard on an Nvidia nForce 680i SLI chipset we found a performance hit on transitioning from a FSB frequency of 420MHz to 425MHz (for details see our article called ASUS Striker Extreme Mainboard Review). Mainboards on the Intel P35 Express chipset seem to be free from this drawback, judging by first tests (for details see our article called Asus P5K Deluxe Mainboard: Second Encounter).

Some non-overclocker-friendly mainboards on Intel’s 945 and 965 series chipsets do not support the FSB strap at all (se our Biostar TForce P965 Mainboard Review) and CPUs with a default 200MHz FSB can usually be overclocked on them to a 300MHz FSB or something. This can be corrected with a CPU modification called BSEL mod. By isolating and connecting contact pads on the CPU’s “belly”, the mainboard is made to think that the default FSB frequency of the CPU is 266MHz rather than 200MHz and behave appropriately at overclocking.

So, you should be aware of the FSB strap thing beforehand and try to avoid non-overclocker mainboards. You should pick up your mainboard considering the default multiplier of your CPU so that you didn’t hit the frequency range with reduced performance. Perhaps you will even have to lower the maximum FSB frequency a little to avoid this range. On the other hand, you shouldn’t have much fear of this thing. If your CPU can overclock far beyond 500MHz FSB, you shouldn’t care about the frequency of the FSP strap because the high frequency of the CPU will make up for any performance hits.

Overclocking Specifics of AMD CPUs

AMD processors can be overclocked like any other CPUs but you should reduce the frequency of the HyperTransport bus that connects the CPU with the chipset. It’s usually enough to set an x3 multiplier or a frequency of 600MHz, which is in fact the same thing.

Besides, with AMD processors the memory controller is integrated into the CPU. It means the resulting speed of the system doesn’t depend much on the employed chipset and will generally be the same in most cases. It means you can take almost any mainboard, except for those models that suit poorly for overclocking due to their limited BIOS options, sloppy PCB design, or other such reasons. Read a review of your mainboard to learn if the mainboard you’ve chosen is good for overclocking.

There is another difference from Intel’s CPUs and it is due to the integrated memory controller, too. The memory timings play a more important role for AMD processors, especially with DDR SDRAM. You should test your system in different modes to see if it’s better to reduce the timings rather than to increase the memory frequency.

Note also that AMD Athlon 64 X2 processors on the 65nm Brisbane core are slower than their 90nm Windsor precursors due to slower cache memory and fractional multipliers (see our article called AMD Athlon 64 X2 4800+ Anew: AMD Masters 65nm Technology). The memory frequency on the AMD platform is based on the CPU frequency and integer divisors and the real memory frequency may sometimes be much lower than set in the BIOS, leading to a performance hit. That’s why dual-core processors on the Windsor core are preferable for overclocking. Their overclockability isn’t inferior to that of their technologically advanced but slower mates.

Life after Overclocking

If you think you can sleep well now that you’ve overclocked your CPU, you are mistaken. High CPU frequency is just a means to get a higher performance from your whole system and you have to do more to reach that goal. The CPU means a lot, but the memory timings almost always affect performance whereas performance in games is often determined by the graphics card.

You decreased the memory frequency when you were preparing to overclock the CPU. Now you should raise it back if possible. The general rule is the higher the frequency, the higher the memory performance is. So, you should leave the memory timings unchanged (you have lifted them up previously) but try to achieve the maximum possible memory frequency. It usually helps to increase the voltage, but not too high. It is undesirable to increase the voltage of DDR2 memory higher than 2.1-2.3V. Found the maximum frequency? Good. Now you should find the minimum possible timings for it. As opposed to the frequency, it’s better to have lower timings.

I give you general recommendations, so don’t be afraid to check them out in practice. You may find yourself forced to select an inconvenient divisor or increase the timings too much to get a high memory frequency. Perhaps it will be better, for the overall system performance, to reduce the memory frequency a little, but set lower timings. Run a few tests for different frequency/timings combinations and choose the best one.

Performance in games is largely determined by your graphics card. If you are a dedicated gamer, don’t forget to overclock your graphics card as well. Overclocking graphics cards is an extensive subject that calls for a separate article. It’s not enough anymore to increase the GPU and graphics memory frequencies to get the maximum possible performance. Today, you have to account for the graphics core having multiple subunits clocked at different frequencies. You have to watch for image freezes. You should reflash the graphics card’s BIOS to correct the timings and frequencies.

Now that your system is overclocked and, hopefully, delivers considerably higher performance, you can sleep well. I mean if you can suppress the desire to tell everyone of your success and check out the increased performance of your PC in real-life applications.

Good overclocking luck to you!

AMD Company expands the line-up of their processors on K10 micro-architecture. Now besides quad-core Phenom X4 they will also offer triple-core Phenom X3. Let’s see if AMD managed to design a successful solution this time without switching to new production process.

AMD’s situation in the processor market these days can hardly be called enviable. AMD fans pinned a lot of hopes upon new K10 micro-architecture, however in reality it didn’t help the company create worthy rivals to Intel’s solutions, even though this micro-architecture is indeed innovative and efficient. Evident strengths of this micro-architecture, such as native quad-core design and L3 cache shared between all cores, remained in the shade because of technological issues that prevented AMD from getting to produce processors with frequencies beyond 2.5GHz. As a result, quad-ore Phenom X4 processors that are already available today turn out unable to compete not only against new 45nm Penryn CPUs, but also against the old 65nm Intel processors.

Moreover, the performance gap between Phenom X4 and Core 2 Quad is so dramatic that even the possibility of performance parity between the two is pretty vague. It is evident that 65nm production technology currently used by AMD will not allow them to increase significantly the working frequencies of their Phenom processors. As for the transition to a more advanced 45nm production process, it is scheduled to take place in Q4 2008. However, 45nm Deneb processors that should come to replace 65nm Phenom will be able to hit only 3.0-3.2GHz frequencies right from the start. And it seems to be not enough to successfully compete against the top quad-core Intel CPUs, so AMD will have to put up with the fate of an inexpensive processor supplier for another while.

AMD does understand the situation very well and tries to introduce the platform concept that will help promote not just bare CPUs but complete kits including a processor, a mainboard and a graphics card. This approach may allow the GPU to make up for insufficient processor performance, which AMD marketing people are trying to stress. However, these kits may be of interest primarily to OEMs and system integrators rather than end users, who are used to putting their systems together from individual components selected basing on their own specific preferences. That is why it is not surprising that neither AMD Spider platform with discrete ATI Radeon HD graphics, nor Cartwheel with the integrated AMD 780G chipset caused any significant stir among computer enthusiasts.

In this situation AMD has to look for other ways to win consumers’ hearts. Their main strategy in attaining this goal has become aggressive pricing policy. Together with the launch of Phenom X4 9X50 processors using the new core revision free from the “TLB bug”, they also lowered the prices of their quad-core processors proportionally to their performance in reference to that of the competitor solutions. As a result, AMD is currently offering the most affordable quad-core solutions in the market. That will certainly find their way into some users’ systems with strategic positioning like that. The same things are happening to the dual-core Athlon 64 X2 processor family that loses to contemporary Core 2 Duo processors in every test. Therefore, retail prices of Athlon 64 X2 dropped so greatly that now these CPUs are regarded only as budget solutions.

Price drop is a good way to maintain proper sales volumes. However, in this case the forefront of the computer community loses interest to AMD solutions and the company is no longer regarded as a technology leader. So, AMD had to find another unique way of warming the public up to their processors. And today’s announcement of the unique Phenom X3 processor family with triple-core configuration is exactly a measure like that. Of course, one of the reasons for these processors to appear is direct economical benefit for the manufacturer, as they got the opportunity to use up “defective” dies for quad-core Phenom processors by disabling one of the cores. But on the other hand, the launch of Phenom X3 may also be regarded as another attempt to compete with Core 2 Duo that are superior to dual-core Athlon 64 X2 from all standpoints. Triple-core Phenom processors are positioned as intermediate solutions between Athlon 64 X2 and Phenom X4, and their price will make them direct competitors to Intel’s mainstream dual-core processors.

So, in our today’s review we are going to check out the new triple-core AMD solutions from this particular stand point. Contemporary software is more and more dedicated to multi-threaded environments that is why triple-core Phenom processors may turn out quite interesting as an alternative to dual-core Intel CPUs. Luckily, we will not be kept in the dark regarding the practical features and potential of the new Phenom X3 processors. AMD provided us with one of the first retail processors from the new series and today we are proud to offer you the results of our extensive tests.

Simple Arithmetic of a Triple-Core CPU

I believe that the new AMD Phenom X3 triple-core processor family (also known as Toliman) doesn’t need a special introduction, as there is hardly anything new about it, if you look closer. These CPUs are based on the same semiconductor dies that are used in quad-core Phenom X4 processors. AMD simply blocks one of the cores in them thus making use of the “defective” chips that failed to become the base for fully-fledged quad-core processors. The mere idea of disabling part of the semiconductor die to ensure that defective high-end chips can get a second chance is not that new at all. However, until recently, AMD and Intel disabled only part of the L2 cache memory.

Phenom X4 processors differ from Intel’s quad-core ones primarily by the monolithic design, which Intel solutions are made of a pair of dual-core semiconductor dies. Therefore, the chances are quite high that there may be a defect in one of the Phenom X4 cores. And it is certainly more probable than the defect in the higher level L3 cache memory. Here Phenom’s block structure was also to AMD’s advantage. Its cores are combined at the L3 cache level, which allows taking one of the cores out of service without changing anything in the micro-architecture and semiconductor die itself.

If we compare the specifications of Phenom X4 and Phenom X3 side y side we will get even more assured that they are closely related:

As a result, Phenom X3 processors turn out absolutely identical to their quad-core elder brothers except for the number of cores.

Today’s announcement mentions three Phenom X3 models with 2.1GHz, 2.3GHz and 2.4GHz frequencies. All these three processors are based on the new B3 stepping free from the notorious “TLB bug”. You should remember however that AMD also makes Phenom X3 processor models based on the old B2 stepping, however, they are not supplied to the retail market.

To avoid confusion in the significantly expanded line-up of Phenom processors based on new K10 micro-architecture, we decided to put together a table listing all the key features of the existing solutions:

We highlighted three new triple-core processors that will be the first Phenom X3 to become available in retail.

Note that all new Phenom X3 features 95W TDP, which indicates that they should work with a broad variety of Socket AM2/Socket AM2+ mainboards, including those from the value segment. In fact, all you need to ensure that new triple-core processors will be compatible with your mainboard is to reflash the BIOS.

The compatibility of the new Phenom X3 processors with available software seems a little bit more complicated. Since it is the first CPU with three cores, it may face some problems with applications being unable to detect and use odd number of cores correctly. However, these individual problems will hardly become very widely spread. During our test session, for example, we haven’t had any issues, except the old versions of the SiSoft Sandra diagnostic utility that didn’t work.

Nevertheless, I would like to draw your attention to the fact that Microsoft has released an update a few days ago for 32-bit Windows server 2008 and Windows Vista operating systems intended to solve problems caused by incorrect detection of the number of available cores. You can get the corresponding info regarding this update from Microsoft official web-site. This update eliminates potential problems with incorrect detection of the number of cores in triple-core CPUs, but it is not mandatory. Even without it our Windows Vista Ultimate testbed could see all three cores just fine.

Keeping in mind that Phenom X3 is actually not that much different from Phenom X4, the most interesting thing about it is the price. After a lot of hesitation, AMD decided to set the following official prices:

• AMD Phenom X3 8750 (2.4GHz) – $195;
• AMD Phenom X3 8650 (2.3GHz) – $165;
• AMD Phenom X3 8450 (2.1GHz) – $145.

So, triple-core Phenom X3 family is positioned as something in-between quad-core Phenom X4 and dual-core Athlon 64 X2. As a result, the new processors fit logically into AMD’s product map and become competitors to Wolfdale family of dual-core Intel Core 2 Duo processors, which prices were reduced this past Monday.

But will three cores of the new Phenom X3 be able to compete successfully against two Wolfdale cores? Our test session will try to answer this question. But before we move on to the results, let’s take a closer look at the triple-core processor sample we got into our lab this time.

Closer Look at Phenom X3 8750

Triple-core Phenom X3 8750 looks exactly like its quad-core counterparts. Only the marking gives it away: HD8750WCJ3BGH.

While the first number “9” in the model name indicates Phenom X4, AMD selected indexes starting with “8” for its triple-core processors. “50” in the end of the marking stands for the absence of the TLB bug, just like in Phenom X4, i.e. it means that the CPU belongs to B3 stepping. The second digit depends on the frequency, and the principle here is the same for triple- and quad-core processors. In other words, Phenom X3 8750 you see on the photo above should work at 2.4GHz. It is the top model in the line-up today.

The CPU features three L2 caches (one per each core), each 512KB big and a shared 2MB L3 cache. The built in North Bridge works at 1.8GHz frequency and supports dual-channel DDR2 SDRAM that can work in Ganged or Unganged mode. So, the CPU uses 1800MHz HyperTransport 3.0 bus, however, nevertheless, it is compatible not only with Socket AM2+ but also with older Socket AM2 mainboards.

Nominal Vcore for Phenom X3 is set in the interval from 1.05V to 1.25V. Just like their elder brothers, these new processors support Cool’n’Quiet 2.0 power-saving technology, which, however, is only available on Socket AM2+ mainboards.

Testbed and Methods

As we have already said, Phenom X3 processor family falls “in-between” Phenom X4 and Athlon 64 X2. Therefore, besides the entire Phenom X3 line-up, we have also tested the top dual-core AMD processor and the youngest Phenom X4 model.

On the competitor’s side, we will have dual-core processor from the same price range. After the recent price reduction these are three youngest models from the Wolfdale Core 2 Duo family including the new Core 2 Duo E7200 CPU. However, since there still are some availability issues with 45nm CPUs, we have also included older 65nm representatives of the Core 2 Duo family.

The testbed configuration is given below:

AMD platform:

• CPUs:
  • AMD Phenom X4 9550 (Socket AM2+, 2.2GHz, 4 x 512KB L2, 2MB L3, Agena);
  • AMD Phenom X3 8750 (Socket AM2+, 2.4GHz, 3 x 512KB L2, 2MB L3, Toliman);
  • AMD Phenom X3 8650 (Socket AM2+, 2.3GHz, 3 x 512KB L2, 2MB L3, Toliman);
  • AMD Phenom X3 8450 (Socket AM2+, 2.1GHz, 3 x 512KB L2, 2MB L3, Toliman);
  • AMD Athlon 64 X2 6400+ (Socket AM2, 3.2GHz, 2 x 1MB L2, Windsor).
• Mainboard: ASUS M3A32-MVP Deluxe (Socket AM2+, AMD 790FX).
• Memory: 2GB DDR2-1066 with 5-5-5-15-2T timings (Corsair Dominator TWIN2X2048-10000C5DF).
• Graphics card: OCZ GeForce 8800GTX (PCI-E x16).
• HDD: Western Digital WD1500AHFD (SATA150).
• OS: Microsoft Windows Vista x86.

Intel Platform:

• CPUs:
  • Intel Core 2 Duo E8400 (LGA775, 3.0GHz, 1333MHz FSB, 6MB L2, Wolfdale);
  • Intel Core 2 Duo E8200 (LGA775, 2.66GHz, 1333MHz FSB, 6MB L2, Wolfdale);
  • Intel Core 2 Duo E7200 (LGA775, 2.53GHz, 1067MHz FSB, 3MB L2, Wolfdale);
  • Intel Core 2 Duo E6750 (LGA775, 2.66GHz, 1333MHz FSB, 4MB L2, Conroe);
  • Intel Core 2 Duo E6550 (LGA775, 2.33GHz, 1333MHz FSB, 4MB L2, Conroe).
• Mainboard: ASUS P5K3 (LGA775, Intel P35, DDR3 SDRAM).
• Memory: 2GB DDR3-1333 SDRAM with 6-6-6-18 timings (Cell Shock DDR3-1800).
• Graphics card: OCZ GeForce 8800GTX (PCI-E x16).
• HDD: Western Digital WD1500AHFD (SATA150).
• OS: Microsoft Windows Vista x86.

Performance

General Performance

SYSmark 2007, which we consider the benchmark showing integral processor performance, reveals pretty interesting results. As we have expected, Phenom X3 turns out generally slower than the youngest quad-core AMD processor. However at the same time their performance is not any higher than that of Athlon 64 X2 6400+ that runs almost as fast as Phenom X4 9550. So, it turns out that if we base our conclusions only on the data from the diagrams above, we will see that there is barely any market niche for the new Phenom X3. And these processors may be interesting only in very few applications that can load fully all three cores.

In this respect, it is not at all surprising, that Phenom X3 loses to Core 2 Duo processors even to the cheapest E7200 and E6550 models. So it turns out that in a wide range of tasks, during regular, not any specific type of work, even three cores with K10 micro-architecture cannot compete successfully against two cores with Core micro-architecture. And the main problem with Phenom processors is, probably, their low clock frequencies.

However, let’s not draw any hasty conclusions and see how the new Phenom X3 perform in different types of applications.

3D Games

Before we move on to the gaming performance graphs we would like to remind you that to test processor performance in games we use low resolution of 1024×768. this low resolution allows us to focus on the “gaming” CPU performance and eliminate the GPU influence on the results, as the performance of this particular component is the primary determinative in real games.

The performance of Phenom X3 may differ in different games, but nevertheless, we can single out two types of behavior this processor demonstrates in gaming applications. In games, which performance is not very scalable when we have more than two processor cores (in other words, those that do not fully support quad-core processors), Phenom X3 demonstrates poor results. New triple-core processors lose to Athlon 64 X2 6400+, not to mention Intel processors in Quake 4, Half Life 2 Episode Two, and, strangely enough, Crysis.

However, there is also another group of games including Unreal Tournament 3, World in Conflict and Lost Planet: Extreme Condition. Performance in these games depends a lot on the number of available computational cores that is why the new Phenom X3 processors do not look that bad here. At least they do not yield to the top Athlon 64 X2 and sometimes are even capable of racing against Core 2 Duo CPUs not only from the previous generation, but also the new Core 2 Duo E7200.

Media Content Encoding

The situation with media content encoding is determined by the codecs optimization for multi-core architectures. Apple iTunes that is well optimized only for dual-core processors works much faster in Athlon 64 X2 and Core 2 Duo based systems. With DivX codec featuring average optimization for multi-threaded environments, Phenom X3 processors fall just a tiny bit behind dual-core Athlon 64 X2 6400+ with 1.5 times higher clock frequency. However, they are still too far behind the dual-core Intel processors. However, the popular x264 H.264/MPEG-4 video codec that loads the CPUs with multiple cores very nicely, allows to fully reveal the Phenom X3 potential. During our performance tests with this codec, triple-core newcomers not only outperformed Athlon 64 X2, but also caught up with the youngest Wolfdale processors.

Final Rendering

Final rendering is an excellent example of tasks with well paralleled load. Therefore it is not at all surprising that Phenom X3 processor family perform exactly the way AMD wanted them to. The performance of the new triple-core processors fits precisely between the youngest Phenom X4 and the top Athlon 64 X2. Moreover, triple-core Phenom X3 compete quite successfully against dual-core Core 2 Duo including their 45nm models. It is a pity though that this is more likely to be an exception rather than a rule.

Other Applications

Dual-core processors cope better with Adobe Photoshop than Phenom X3. Although there are a lot of filters in this application that can split the workload into parallel streams, the results indicate that triple-core AMD processors lack frequency.

Rendering in Adobe Premiere is similar to 3D rendering. Phenom X3 perform pretty well here.

WinRAR archiving also works faster on Phenom X3 than on the top Athlon 64 X2. However Wolfdale processors from the Core 2 Duo E8000 series with larger L2 cache show much higher results.

The popular computer algebra suite proves more efficient on dual-core processors with Core micro-architecture, although it makes good use of multi-core structures. You can see it clearly from the advantage triple-core AMD processors have over dual-core Athlon 64 X2 6400+.

The results obtained in a popular chess application are another consolation for AMD fans. Yes, there are applications where Phenom X3 processors can work as fast as the youngest Core 2 Duo, and you can certainly find enough of them if you wish.

Overclocking

Although triple-core Phenom X3 processors are based on the same B3 stepping as the quad-core AMD CPUs, we should pay special attention to their overclocking potential. As there are fewer cores working simultaneously, the heat dissipation should also go down and theoretically it should allow achieving better overclocking results.

I would like to point out that Phenom X3 8750 processor we had at our disposal, just like other CPUs from the same family, features a locked clock frequency multiplier. That is why they should be overclocked by raising the clock generator frequency. This is not as easy to accomplish as we wish. The thing is that as we have already mentioned in our special article, this frequency is tied up not only to the resulting CPU frequency, but also to the frequency of the built-in North Bridge, memory and HyperTransport 3.0 bus. That is why when you increase the clock generator frequency, you should remember to reduce the corresponding coefficients and dividers forming the North Bridge, HyperTransport and DDR2 SDRAM frequencies.

For example, by raising the processor Vcore to 1.45V we could increase the clock generator frequency to 260MHz from the default 200MHz without losing stability. However, the North Bridge and HyperTrasnport frequency multipliers has to be set at 7x instead of the nominal 9x, thus keeping the corresponding frequencies close to their nominal values.

In this case Phenom X3 8750 processors overclocked to 3.1GHz frequency and remained absolutely stable. We tested stability with a one-hour run of Prime95 25.5 utility. To dissipate the heat from the overclocked processor we used Scythe Mugen (Infinity) air cooler.

I would like to say that 3.1GHz frequency is the best result for a K10 based processor that we managed to obtain in our lab so far. Therefore, we can hope that Phenom X3 processors are more overclocking-friendly than their quad-core fellows. However, we will be able to draw final conclusions only once we collect more data from the tests of more than one processor sample.

Power Consumption

To get a complete picture we also measured the power consumption of our test platforms (without monitor) built around the participating processors working at their nominal speeds. The systems were configured exactly the same way as during performance tests. Enhanced Intel SpeedStep and Cool’n’Quiet 2.0 power-saving technologies were activated. Processors were loaded using Prime95 25.5 utility.

As we have expected, triple-core processors turned out more economical than their quad-core counterparts thanks to fewer cores. Besides, they boast lower power consumption than dual-core Athlon 64 X2 thanks to not very high clock speeds. However, from the economical standpoint new Phenom X3 family still fails to compete with Intel processors.

Conclusion

AMD Phenom X3 is definitely a very interesting processor. At least since it is the industry’s first CPU with triple-core sign and monolithic organization. And although it is our first experience with a non-standard processor like that, it didn’t cause any issues when working in a convention hardware and software environment. This CPU is fully compatible with the existing infrastructure, which indicates that AMD chose the right strategy in putting the defective quad-core Phenom X4 to good use.

As for the consumer qualities and marketing future of the new processor, the things are not as clear here. All major problems typical of processors on K10 micro-architecture found their way into triple-core solutions, too. That is why Phenom X3 processors, just like Phenom X4 lack clock speed so badly. However, they are still in a little better situation than quad-core processors, because AMD positions them as competitors to Intel’s dual-core Core 2 Duo.

However, Core 2 Duo and Phenom X3 do not always show us an interesting race. We only see it in applications where performance scales well for more than two cores. Unfortunately, there are very few applications like that, so in most cases Phenom X3 loses to Intel processors from the same price range. Still, these applications do exist and include final rendering tasks, some video processing and encoding tasks, and a few others.

So, we have to state that another AMD initiative has not too many chances to succeed. Phenom X3 may become a great niche product, however, they will hardly get very popular. Youngest Intel processors from Wolfdale family priced at the same level offer higher average performance, lower heat dissipation and power consumption and much better overclocking potential. AMD, however, will hardly dare drop the Phenom X3 prices much lower, because they use a monolithic quad-core die, which is pretty expensive to make. To be fair, I would like to add that if AMD decided to lower the prices even more, Phenom X3 may become a worthy alternative to Core 2 Duo E4000 and Pentium Dual Core.

In conclusion I would like to say that Phenom X3 can not always be recommended as a suitable upgrade for the Socket AM2 systems. The thing is that top dual-core Athlon 64 X2 processors can often offer better performance, although with higher heat dissipation.

So, Phenom X3 will hardly become a bestseller. It will most likely find its users, but there are still too many reservations to be made with not too many significant advantages.