Please report all spam threads, posts and suspicious members. We receive spam notifications and will take immediate action!
Page 1 of 2 12 LastLast
Results 1 to 10 of 11

Thread: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

  1. #1
    Join Date
    Sep 2010
    Norther Virginia

    Default Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    Here it is in PDF format if you want to print it or read it on the go.

    With the release of Intelís 2nd Generation processors, code named Sandy Bridge, come a variety of boards from a variety of companies that fulfill every need that can be imagined. Today I will take an in depth look into overclocking these new CPUs and take an in depth look at the performance offerings of GIGABYTEís new P67A-UD7.

    Previously I did in in depth analysis of the board and all of its various ICs and components P67A-UD7 in Depth Look at The Board

    The performance review includes: Voltage read points, SLI performance, Overclocking and CPU performance, SATA6G performance, power consumption, and a look at some board features.Continue to Overclocking Made Simple if you just want to learn to OC, if you want to only see performance review click here:

    Overclocking Made Simple:
    If you are brand new to overclocking or a veteran from the Pentium 4 days you are in for a crash course in the ďnew generationĒ of overclocking to go along with a ďnew generationĒ of processors. Turbo multipliers are a new name for ratios and K series chips are the only ones that want to overclock past 4 GHz.

    What you want to initially do is increase your turbo multipliers or CPU clock ratio. Now if your board is properly certified for Serial VID, you can usually tell by VDR12 Certified or if youíre VID changes after changing the multiplier. If this does not occur, skip down to the voltage section. If you board has proper SVID implemented you can keep reading (most do).

    First you need monitoring and Stability programs. You need CPU-Z, at least one Temperature monitor, a stability tester, and one voltage monitor. You will also need a K series processor to overclock and a P67 Chipset motherboard like this one:

    REQUIRED: CPU-Z will give you processor frequency info as well as Vcore
    REQUIRED: Stability program, Intel burn Test is popular as is LinX (they are basically identical), Prime 95 is also very good.
    Required: Pick one or more or your own monitoring software:
    HWMonitor will give you voltages and temperatures
    Motherboard Software (such as EasyTune 6(GIGABYTE), TurboV(ASUS)): will give you voltages and temperatures as well as in windows voltage and multiplier change options.
    Core Temp for Temperatures
    Real Temp for Temperatures
    Speed Fan: system monitoring
    Step #1: Do not overclock RAM yet, let it run at stock frequencies. RAM is not like it was on previous platforms, it can run at 1333mhz and you can do 5.2ghz, believe me I have. Letís tackle one thing at a time.

    Set the frequency you want for a high but modest overclock (not your final), something like 4.0- 4.5 GHz, set ratio or turbo frequencies (40x-45x), if you only have the option for turbo frequencies make sure they are all the same. If you are on a GIGABYTE board, and just want to use one Ratio setting, you can enable Ratio Change in OS option; this will disable the Turbo Settings (or you can also disable Turbo and OS Ratio Change).

    If you enable Turbo Mode (which you do not have to and I recommend not doing) you will need to set each multiplier. Note that on many motherboards you do not have the option to OC without turbo, its ok, 52x multi is 52x multi whether its turbo or not. On GIGABYTE boards, enabling turbo shouldnít make your multiplier go crazy, I have tested on the latest and even old BIOSes and you can use Turbo and the multiplier not drop. Turbo Mode or no Turbo Mode, your multiplier will stick steady. If for some reason it is jumping around, you have the option to disable Turbo Mode.

    I want to also mention that Turbo Mode, Real Time Ratio Change in OS, or even just disabling both make no change to your Overclock or affect its stability. The only thing that can help is with turbo mode you have the options to extend and set maximum TDP and TDC, I will talk about this later. If you have retail CPU please enable Internal CPU PLL Overvoltage option.

    Make sure you DISABLE all of the following advanced CPU features (to stabilize multiplier and voltage):
    C3, C6 States-DISABLE
    CPU Thermal Monitor-DISABLE

    Do not touch BLCK right now as it always needs to be lowered as you overclock higher, at stock or low OC (under 4.3 GHz you can increase BLCK no problem). Make sure you make note of Stock VID/Vcore/VCC (found under voltage options). Upon reboot if the system boots into BIOS check the new VID under voltages settings or CPU Vcore (VCC) under health monitoring, take this value and compare it to this:

    Stock-4GHz on Stock VID
    4.0-4.3GHz 1.300v-1.325v
    4.3-4.5GHz 1.325v-1.375v
    4.5-4.8GHz 1.375v-1.450v

    Above you can see the VID change from 1.23 to 1.370 for 3.4 to 4.5 GHz automatically, if this occurs SVID is working properly.

    If that fails to boot into BIOS, then set it to 4 GHz (40x), and SVID should work upon reboot. You should end up with a stable 4 GHz overclock with automatically increased Vcore/vcc. Those values listed above are optimal voltage for optimal overclocks up to 4.8 GHz. You want to stay within the low spectrum of those voltages for temperature and processor integrity sake.
    Both my 2600Ks will boot into BIOS at 4.5 GHz without anything else changed, only multiplier to 45x, VID changes from 1.18 to 1.365 on one and 1.23-1.370 on the other. While both boot at 4.5 GHz into BIOS, one only can do 4.8 GHz while the other can do 5.2 GHz (the one with higher VID), both D1 stepping chips. Retail D2 stepping chips should be better overclockers.

  2. #2
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    Step #2 Step up the multiplier one by one until the system fails to boot. If this happens, clear CMOS. Many boards have auto recovery that works perfectly in this manner, such as this P67A-UD7. I was able to step up frequency until 4.9 GHz. At 4.8 GHz VID was 1.375v.
    Now if none of this is working or even if it is, letís move on to Voltages. (If you want to skip to Step #3 please click here)

    Now letís move to voltages, this is very important and I think anyone looking to overclock should have a thorough introduction to the variety of voltages that the user is able to control.
    Make sure to pay attention to my tid-bits on Vcore(VCC) and VTT(VCCIO), these are most important for overclocking, other voltages can help too.
    Here is a table I put together from Intelís product data sheet; it defines the processors maximum voltages and maximum amperage- These are OFFICIAL Min and Max Values:

    Here are the technical signal names as well as a short description as well as explanation of every one and what it has to do with overclocking:

    VCC: Commonly Called Vcore, the voltage supplied to the processors inside the CPU. This voltage is a large part of Sandy Bridge overclocking. Now from personal testing, and weeks and weeks of headaches and hard ships, I have a few things I would like to share about this voltage.
    I say maximum voltage for Vcc/Vcore is 1.50v for 24/7, 1.55-1.60v for extreme benchmarking, please stay below to 1.6, and donít use any type of Load Line Calibration past 1.55v. Please not that high temperature teamed up with high voltage will kill your processor faster than anything else, it creates a perfect environment for processor degradation.

    Load Line Calibration or Vdroop control is a setting that eliminates/reduces processor voltage droop under load and in many extreme cases many reverse voltage droop. Voltage droop is there so that under load conditions where current (amperage) is increased the processor stays within TDP (Thermal Design Power). There is also vdrop which is implemented by the motherboard manufacturer, vdrop is there and it drops every voltage on the board from what is set in BIOS, no matter the load. LLC can reverse Vdroop and Vdrop in many cases.

    Let me explain.
    Voltage x Amperage = Wattage.

    As you can see in the table above there is a TDC and a TDP. Now TDP is max heat output of the processor (wattage), letís say we stick within Icc (current/amperage) of 85amps at load. 85amps x voltage=95watts (TDP). So If voltage is increased for overclocking, then at idle the processor needs to droop the voltage so that amperage x voltage doesnít exceed TDP, you do not want to drop amperage because it is almost all the power (plus you donít have control over it), voltage is nothing but a signal without amperage behind it. Of course with Sandy Bridge you can set the upper limit on TDP and TDC and you wonít push that much amperage until critical point, plus Sandy Bridge can exceed TDP on its own.

    Many people are 100% against LLC and many are 100% for it, I sit on the fence as I find it useful, and havenít killed a processor from it, but other things can occur and the life span of the processor will be reduced by too much voltage, with or without LLC. LLC will improve stability, please read the lesson that I learned with 1.5+V on Sandy Bridge.

    Important Learning Lesson about high VCC and high LLC(Processor Degradation): Let me share a little about electron migration and processor degradation. When I first received these beautiful 2600Ks, one could boot 5.2 GHz with 1.475v with Load Line Calibration enabled to level 2(Vrise / Total Vdroop eliminated), real processor voltage was at about 1.5v. Temperatures were through the roof under load. 80C-100C was common and so was thermal throttling (I will talk about cooling problems in a little bit); using handy software I was able to monitor the throttling. Through the next few weeks the processor voltage for 5.2 GHz stable (even boot in some cases) eventually rose to 1.55v LLC Level 2, which gives a total voltage of 1.58v. Benchmarking with CPU intensive load (100%) was impossible because of the heat produced. Eventually I was able to back down frequency one multiplier to 5.1 GHz and use 1.475v with LLC level 1 (elimination of vdrop and slight elimination of vdroop (still droops under load)) with my air cooler: Eclipse II. Temperatures donít even touch upon 70C at 5.1 GHz. I have a heavy duty watercooler that can handle 220W load, and my i7900 series at 4.5 GHz with 1.55v (more calculated/measured power output than this chip). With these chips unique thermal capacitance features, thermal output at high loads is very intense, make sure you buy a good cooler. One thing to mention is that my chip is a D1 stepping processor and the PLL Overvoltage unlock feature does not work with these earlier revisions, but all retail chips will be D2, and should overclock significantly better with less vcc/vcore.
    VCCIO: more commonly known at QPI/VTT voltage, this is the VTT voltage. Formally known as Processor Power for I/O it is the voltage for the integrated memory controller as well as the PCI-E controller. While Intelís Maximum is 1.05 +/- 3% = 1.08v, you can go higher, much higher. I would recommend staying below 1.2v for 24/7 use, but depending on the quality of the IMC on your chip, I have seen 2133 MHz done on as little at 1.1v. I used 1.12v for overclocking my Dominator 1600 MHz to 1866 MHz, and it did it without any problems. Do realize that this voltage contributes heat as well to the whole thermal package.

    VDDQ: more commonly known as Vdimm or Vdram, this is the voltage for your memory. Formally known as I/O voltage for DDR3, Intel states maximum at 1.575. YOU should run this at whatever it says on your RAM. At the time I am writing this article, 1.575 is not the standard, but 1.5v has been stock voltage on many DDR3 RAM modules for a long time. While at 1.5v you can run at stock speed of 1333 MHz and SPD 9, 9,9,24 to run your RAM at a higher speed, such as 1600MHz, most RAM requires 1.65v. Do not be afraid, if it says 1.65v on your RAM stick, set it to 1.64 or 1.66v. For overclocking higher than what your RAM is rated for you can take this up, I have used up to 1.76v, but for my tests I used 1.72v to run my 1600mhz Ram at 1866mhz. I wouldnít run this voltage over 1.8v unless you are going for some crazy high clocks.

    VAXG: internal graphics processor voltage, not applicable to this processor.

    VCCPLL: Commonly known as CPUPLL, this voltage is for the internal clock generator for the CPU. Intel states maximum at 1.89v, and stock at 1.8v.

    What does the PLL do, you ask? Here is how you get 3.4 GHz. A constant frequency input (BLCK) is generated by the PCH (P67), the BLCK is then multiplied by the core ratio by the internal phase lock loop (PLL) and then you have a greater resulting core frequency. CPU PLL is the PLL that gives you 3.4 GHz at stock and 5.2 GHz overclocked.
    I would leave this voltage at stock, at one point I thought this voltage helped me lower Vcore, but it was just the processor playing tricks on me. In fact on X58 systems lowering the CPU PLL was thought to help lower temperatures and thus improve stability, on the other hand at very high frequencies this voltage is creased by many overclockers. I say increase it to 1.89v if you like, but donít go north of that, if you want to save power and keep temperatures low turn it down to 1.71v.

    Instead of messing with this, use the Unlock CPUPLL Overvoltage Option under frequency control; this should help you get past 5.2 GHz on D2 stepping chips and beyond, if you need more oomph after that, increase it. (This setting if enabled on a D1 chip will actually cause it not to boot, like mine)

    VCCSA: More commonly known as System Agent Voltage. Intelís maximum System Agent Voltage (Vccsa) is 0.971v and minimum is 0.879v. Stock is 0.925v. System Agent Voltage should NOT be touched, it is supposed to be a fixed voltage, and it powers many things that the VCC does not power. One of the most important is the Power Control Unit (PCU) which controls internal power allocation of the processor. This voltage is to be generated by a separate VRM than used for SVID. So on the P67A-UD7 this voltage is generated by a two phase buck analogue PWM, with 4 phases, this voltage and the VTT (Vccio) come from the same VRM(not surprisingly voltage read points are right next to each other as well).

    Overclocking Continued:
    Now Serial Vid was discussed in short in the in depth preview, itís a new way for the processor to basically communicate with the motherboards voltage regulator module (VRM). Serial Voltage Identification or SVID has the ability to automatically scale processor voltage up to 1.52v.

    Step #3: If for instance you are at 4.5 GHz and 1.35v that is excellent, I would try to boot at 1.35v and run some stability tests. If itís stable back off the voltage, take it down 5-10mv (1-2 voltage steps) at a time. Test and repeat until it is unstable, then push Vcore/VCC back up a few notches. Run stability tests until you are satisfied with your Overclock; make sure you set the tests to 100% load on all threads. Make sure to watch temperatures. If you are hitting 80C at 100% load then that is your Vcore/VCC limit, go no further until you get better cooling.

    Step #4: BLCK Options. Now letís face it, sometimes itís just not practical to OC to 5 GHz and use that every day, the heat output is harsh and the system is loud and you want a lower overclock, but you want to maximize performance along with it. Here is where BLCK adjustment comes in handy. You cannot increase BLCK at your maximum OC, it is better to lower OC 1-2 multipliers and then try to take up BLCK. Boot into BIOS, do not try to change BLCK in Windows, and try raising BLCK by 0.2mhz, I have found that even at 5.1ghz I was able to boost BLCK a fraction of a MHz If you want to push more than a fraction of a MHz, you will need to increase Vcore yourself. Try 0.1MHz per 5mv adjustment. I personally do not think it is worth it, I would rather the extra vcore go to multiplier, but hey every little bit counts. At stock I can do 107.7BLCK on my worse chip; I havenít tried the better chip yet.

    Decrease BLCK if you are getting stuck at 5.2 GHz and 100blck, many people use lower BLCK to allow for a lower vcore to be used for 5.2 GHz and beyond. Even try increasing PCH voltage one notch, I found that increasing it a tiny bit may help BLCK in some instances. Make sure you donít go over 1.1v for 24/7 use, I have also found the CPU PLL also helped a tiny bit, you can try up to 1.89v, but this also increases heat output. I was able to do 107blck no problem at stock, at 5.1 GHz I messed around with BLCK and brought it up to 100.8, this put me right up at the same 5.2 GHz frequency but I ended up with the same Vcore and even more heat. I found that adjusting BLCK up 0.2-0.4MHz was no problem below 5 GHz, but above 5 GHz you need to watch your BLCK, drop it or lower it to accommodate your OC the way you like.

    Step #5: If you are still unstable upon boot, and be stable, then try these options:

    A. Select CPU PLL Overvoltage to enabled, this allows most people to go beyond 4.8 or 5.2 GHz whatever your barrier is. If you have a D1 stepping processor, enabling this option will cause you to fail to boot.

    B. If not already using turbo options, please do so. Now set upper limit on TDP and TDC at 200-300 to removed TDP and TDC limitations. If your board has OCP (over current protection) you can disable this if you want a very high overclock.

    C. You can try to enable Load Line Calibration to its full extent.

    D. Try disabling Fan control under health monitor

    E. Try disabling extra features not in use, such as extra SATA6G controllers and USB3.0 controllers (make sure if you have mouse and keyboard in those ports to move them to USB 2.0 ports.)

    In the end if you are unable to get above 4.8 GHz then hey 4.8 GHz is still a heck of a lot of power for only the cost of a 2600K or even cheaper chip. I have two 2600K chips, both are the same batch and both are D1 stepping. One will OC to a max 5.2 GHz and one to 4.8GHz no matter what I do it wonít budge any higher. Goes to show you every processor is different.
    Here was my Max OC, seems to be the max for any chip that is D1 stepping, D2 stepping chips should use PLL overvoltage enabled to get past 5.2GHz if your chip can. Also notice max BLCK of 107.5 stable.

    5.2GH/z and 107.5 BLCK


    (You can see in the above picture that the CPU is throttling, Tmonitor shows throttling of each core separately.)

    Now along with the new turbo multipliers comes a technology that is aimed at pleasing the end user, and that is Sandy Bridges ability to shortly for (25 seconds implemented, but motherboard manufacturers extended this to allow TDC and TDP extension) exceed TDP to utilize thermal capacitance of the internal heat spreader and heatsink. What this means is that frequency can be increased as well as overall thermal load resulting from voltage and current increase, and that the system can absorb more heat as soon as the high frequency is engaged to take advantage of the heat up time of the internal heatsink and external cooler. Intel found that in the beginning of high thermal output load, the ďpackageĒ (cooler) can absorb much more heat than the processor puts out normally, so they incorporated the ability for the processor to give off more heat at the beginning and operate at max frequency. Now this is great and all because the processor can run cooler, but this also means that the processor is going to run extremely cool at idle, so cool that you will inadvertently take it for granted, and then when you apply load and the turbo multiplier increases, you will have a huge amount of heat that needs to be dissipated, instead of predictable heat output. Now donít worry, Sandy Bridge has an internal overheating turbo throttle, which will reduce the multiplier by one and then take a voltage as low as it can and then re-measure temperature to make sure it is getting cooler, if not then it will reduce frequency again.

    This reduce and recheck cycle happens at an astonishingly fast 2us (microseconds). All of this thermal throttling control is part of the Thermal Control Circuitry (TCC). I have tested and found that this throttling happens at 90c-95c. At 5.2ghz that was my problem, I have to use LLC Level2 and 1.53v and well that was just way too much heat at 1.56v real steady and the high amperage. That is why most of my benchmarks are run at 5.1ghz instead of 5.2ghz, 5.1ghz needed MUCH less voltage.

    Crazy Cold Bug:

    Above you can see Gflops performance increase drastically at 5.2 GHz, I used subzero air from Mother Nature to cool down the Heat Sink. The problem is that at 5.2 GHz my D1 stepping 2600K has a cold bug below 20C; the system has to be above 20C to boot, as well as bench. This cold bug is being circumvented with each new stepping Intel is releasing, so hopefully this will improve, because 20C cold bug is terribly high. My problem would be I could bench the processor, but as soon as the CPU load stopped my system would hang because temperatures drop below 20c. BUT the cold bug is not the same at every multiplier, at 51x there was no cold bug, I could go as cold as I want. This is very weird indeed, so watch out for it.

    Overclocking RAM: Now I am sure many of you know how to OC RAM. But you have to be aware that RAM is not like the processor, it cannot usually OC 50%+. If you have a kit with low latency you can probably push it up to the next level by increasing the latencies and RAM voltage. Make sure you increase VTT/VCCIO voltage along with it. You shouldnít need more than 1.25v VTT/VCCIO. I have seen 2133 MHz on as low as 1.1v. Be aware that RAM amount makes a huge difference. 4GB (2GBx2) will OC easier and on less voltage than an 8GB (4GBx2) or 16GB (8GBx2) kit. The amount of Ram Modules makes a difference as well because each pair of slots has its own IMC/Channel (not to be confused with dual channel) if you have four sticks of Ram you are going to need more voltage for the VTT/VCCIO than if you have only two sticks. The same goes for 1 stick vs. 2 sticks.

    I was able to overclock my Dominators from Cas 8,8,8,24 1T 1600 to cas 9,9,9,24 1T 1866, 4gb kit.

    Now X.M.P. profiles are very attractive to mainstream users, as it is a performance profile that increases speed, decreases latency, and increases VTT voltage. Now most DDR3 high performance Ram out there to date is built for X58 and P55 systems, watch out for the X.M.P. settings, if you want to use X.M.P. make sure profile VTT voltage is below 1.2v qpi/vtt(Vccio).

    Here is a list of Common BDOS Errors and what to do to get rid of them; these suggestions are from trial and error, and many BDOSes from hundreds of hours of overclocking. I have gotten many of these BDOSes and checked them out (tried to cause them) and I have modified that list, here it is.
    BSOD Codes
    0x124 = add/remove vcore or QPI/VTT voltage (usually Vcore, once it was QPI/VTT)
    0x101 = add more vcore
    0x50 = RAM timings/Frequency add DDR3 voltage or add QPI/VTT
    0x1E = add more vcore
    0x3B = add more vcore
    0xD1 = add QPI/VTT voltage
    ď0x9C = QPI/VTT most likely, but increasing vcore has helped in some instancesĒ
    0X109 = add DDR3 voltage
    0x0A = add QPI/VTT voltage

  3. #3
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    P67A-UD7 Performance Review

    I am now going to move on to the board performance review, first let me start by saying that the current BIOS is not the same as the UEFI BIOS that will be out shortly, and because of that, I feel there is no need to show off the BIOS, I provided screen shots of the BIOS for the most important Overclocking Areas above in the OC guide.

    Board Spacing:
    Focusing on the socket area we can immediately tell that this board has a very hefty VRM design. The board carries 24 phases VRM with Driver MOSFETS (24 of them) cooled by a low profile and very well designed elegant heatsink. The air cooler used in all the testing is a LinSpire Eclipse II, a 5 heat pipe efficient air cooler that tops the charts in performance figures.

    Here you can see how the cooler fits with a hefty aftermarket 120mm fan:

    As you can see above, the low profile design of the heatsink makes it possible to fit large fans. Around the socket Area we have these shiny cubes that are called Ferrite Core Choke, as well as a small bank of capacitors. In the Intel VRD 12 PWM spec sheet, there was an interesting sentence that read: A faster VR with a small bank of capacitors will provide better end user experience than a slow VR with a large bank of capacitors. Gigabyte took that to heart, and provides the entire P67A line of Ultra Durable 3 boards with extremely high quality Driver MOSFETs, Chokes, and Capacitors. Ferrite Coke Chokes are basically high quality inductors; they reduce electromagnetic interference, noise, flatten current spikes, and filter high frequency signals. They are perfect for motherboard power regulators. Driver MOSFETs make it possible to reduce the overall footprint of the VRM on the motherboard by combining traditional 3 chips into one, a high side MOSFET, a low side MOSFET, and a driver chip. Now they all fit into one integrated circuit. Driver MOSFETs can be fed a range of voltages, such as 12v and then reduce that voltage to 0.5v-2.0v for the CPU. Sticking with same powerful Driver MOSFETs from the X58A series of boards each phase can output 35amps of current continuously. The CPU has a TDC of only 110amps. Capacitors are used to reduce voltage ripple, to provide a constant clean power. Capacitors do this by storing and electrical charge, and then discharging when needed, the discharged electricity would in theory have less variance, and testing proves that capacitors do their job. By using such high quality parts for the rest of the design, GIGABYTE was able to reduce the number of capacitors needed. Gigabyte uses an Intersil PWM, which is why the voltage when set to auto can correctly set its own value. This is due to the fact that the VRM and CPU can talk to each other and decide upon the best VID for the current frequency, and you donít need to reboot for this to happen, the voltage will change on the fly.

    This board's VRM is VRD12 CERTIFIED, it is very important certification that is hard to meet. It depends on VRM response time as well as other factors. This is something to look for when buying a board, because VRM response time indicates how fast the turbo multipliers can change. This board has no lag when using turbo mode. It is an Intel Certification.

    Let move on to the other side of the heatsink, the side where the RAM modules are. There is much concern with the spacing between the CPU socket and the RAM. While this might be a cause of concern, it is due to two reasons. The increase in the distance of the traces would decrease electrical performance and this design is by Intel specifications. Most RAM modules like my Dominator set have detachable heatsinks, here are a few shots.

    As you can see with the heatsinks off the RAM modules, clearance is a little bit less of an issue; a taller cooler would make a good fit if you want to add a second 120mm fan.
    Now here is a shot of the problem with a large 120mm fan, the fan sits on the Ram modules:

    Of course you can make it work, but it is still an issue. You can always move the modules into different slots to accommodate the fan.

    Now let me move onto PCI-E slot spacing. As you can see you have a plethora of PCI-E slots, the first 16x slot is the topmost normal size PCI-E slot, underneath it is a 8x slot, and below that is another 16x slot, and below that a PCI slot and another 8x and another PCI slot.

    Here is a look at spacing with two large GTX570 GPUs and a 3rd GT220 for 3-way SLI. Notice that you can fit another dual slotted card in there very easily.

    Here is a shot of the cards moved around, notice there is one PCI-E slot available with the setup above, and when moving the second GTX570 to the bottom 8X slot it is possible to still have a PCI device and a 3rd single slotted card:

    Now here is the cool thing about the PCH heatsink and angled SATA connectors. There is just enough clearance for these huge GPUs.

    Here is SATA clearance:

    There are no issues for clearance with long XL GPUs.

  4. #4
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    Voltage Read Points:
    I found these points to be the voltage read points for the board. Now I do not have the Data sheet for the ISL 6366 PWM for the CPU so I just indicated voltage read point off of one of the Core Chokes.
    CPU Vcore:

    CPU PLL:

    QPI/VTT + System Agent:

    DDR3 Voltage:

    Voltage Read Points VS Software and BIOS and Load Line Calibration:
    I took manual Digital Multi Meter readings of CPU Vcore, CPU Vcore LLC 1, CPU VCORE LLC2, QPI/VTT, and DDR3 Voltage. System Agent and CPU PLL voltages were only taken from what is set in BIOS to what I found on the board. Now I did not spend much time on this, so if you find a better read point let me know!

    Percent Difference:
    Average Vcore percent difference: 0.5% difference
    QPI/VTT percent average percent difference: 0.1% difference
    DDR3 percent average difference: 0.6% difference
    System Agent percent difference: 2%
    CPU PLL percent difference: 1.5%

    It looks like the Ite voltage monitoring super I/O chip is doing its job. The voltages you should be most concerned with are Vcore and VTT, and they are almost dead on (VTT is really dead on). Now this board is for enthusiasts, so I am sure many will use read points, but software is just fine. You can use the above data on Load Line Calibration to see what level you want to use, if any at all.

    Power Consumption:
    Here are figures at the socket (the wall) power draw by the PSU, keep in mind that the PSU is about 85-90% efficient, so actual total power draw is about 80-90% of what is stated. Also keep in mind that each of the two GTX570s used has a TDP if 210Watts, so you can subtract 400watts from the load figures to get a rough estimate of board/CPU power consumption. Other than the GPUs, and 1 HDD, RAM, and 2 high power fans, nothing else was powered by the PSU. For load, I used Intel burn Test 8thread maximum level, and Furmark after disabling OCP on the GTX570s, OCP was enabled for all the other tests.

    Now I am going to do some estimation. First multiply by 0.85 for the PSU, then subtract 420watts for the GPUs, then subtract 20 watts for the HDD and 10 watts for the Fans. Now these are rough estimates, and the wattage meter is not perfect, I would give these results about 3-5% error. BUT they look fine, stock numbers on the 2600K are good, the stock TDP is 95watts plus motherboard and Ram would be about 117 watts. On the i7 930, TDP is 130watts, and results are 143.3, now that is a little low, but the Kill-A-Watt meter I used isnít perfect.

    But what is perfect is the reduction in thermal leakage of the CPU, these CPUs run at much higher voltage and amperage (most likely) yet run much cooler. Just like the 6-core Gulftown 32nm chips have the same TDP as the 45nm Bloomfields, but the Gulftowns have 2 extra cores. These numbers look about right, and I did need a 1K PSU for the GTX570s. If you are looking to buy this system, it looks like the motherboard is very efficient, with all those power phases you are probably at 90-95% efficient at the CPU socket, based on the efficiency rating of the MOSFETs.

  5. #5
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    CPU/ Board performance testing VS X58 system:
    Here is a pic of the Sandy Bridge system in my case after testing was done.

    Overclocking settings per chip:
    Stock means loaded stock defaults, 1333 MHz ram cas 9 for P67A-UD7 and 1066mhz cas 9 for X58A-UD5. For the 2.8ghz 2600K memory speed of 1066MHz at 9, 9,9,24 was used to even out the bottom comparison line. In some memory benchmarks I used another setting as well with the 2.8 GHz 2600K at 1333MHz to compare speed to stock 3.4 GHz 1333 MHz; same timings were used for all runs, except max OC. I made note where any setting was changed.

    Test Systems:

    CPU Performance Benchmarks:
    SuperPi Mod 1.5 XS, this program calculates Pi 22/7 to the 1 millionth digits, itís a good indicator of overall CPU/RAM/motherboard performance, and usually is an indicator of pure speed. 5.2 GHz overclock was included.

    Now 32 million digits

    As you can tell from the SuperPi results, the 2600K is one fast little sucker. This processor clock for clock beats out the i7 930, remember the 2600K at 2.8 GHz has the same RAM speed as the i7930 at stock.

    Now letís move on to Intel Burn Test, not only does this program test the stability of your system, it also takes all threads to 100%. It has a Giga Flops output that is more consistent than LinX. Now this test also has a result for 5.2 GHz, I need to clarify, that it was one pass, and the system froze up afterwards (b/c of cold bug), I had to use extreme cooling for that test(to maintain heat). I took a picture of the results with my camera. The other results are pretty consistent, all done on air.

    You can see that the 2600K scales VERY nicely, almost perfect. You should also not that this benchmark really likes BLCK, as it has a lot to do with how quick the benchmark gets done, that is why clock for clock the i7 930 wins. You should note that as the 2600K at 5.2ghz is a 1000mhz faster than the i7 930 at 4.2, but when you overclock the 930 you increase the BLCK, thus if the 930 was at 5.2ghz with a 21x multiplier, it would most likely beat out the 2600K. Giga FLOP (Floating Point Operation Per Second), means 70 G Flops is 70,000,000,000 operations per second.

    Next we are going to use CINEBENCH which gives us an overall estimate of processor computing power, and the result is comparable to other platforms.

    The 2600K scores 9.84 Points at 5.2GHz.

    Since the graphics cards were kept consistent throughout the benchmarks, OpenGL difference is made by the processor here. You can see Sandy Bridge really shines.

    Now letís move to ADIA64 Extreme Edition which is like the older benchmarking program Everest. Here the CPU test HASH, FPU Julia, and FPU Mandel are optimized for Sandy Bridgeís AVX instruction that should really take off in the next few years, as it really improves

    Next while we are working with the AIDA64 EE program, I used this program to also do memory benchmarks, here are the memory bandwidth benchmarks:

    Now remember that at 2.8 GHz the 2600K is at 1066 MHz memory. I also included 2600K at 2.8 GHz with overclocked memory to show the difference CPU frequency makes to memory bandwidth. As you can see it does improve memory bandwidth some. We are using the i7930 in dual channel configuration, triple channel should do a bit better, but I wanted to compare the systems on an even plane. As you can tell memory bandwidth is an area that Sandy Bridge performs excellent in without a doubt, just look at how even when down clocked it matches the i7930 at 4.2 GHz and 1600MHz.
    Here is the last AIDA64 EE benchmark, it is memory latency:

    As you can see again, CPU frequency has a great impact on latency with SandyBridge, the fact that I used/set cas 9 even when SPD was lower was so that I could get nice base line readings.

    Next we have real world performance benchmarks, First up is WinRAR:

    2600K is doing well, not better clock for clock, but its higher stock and overclocked frequency give it a huge edge over i7 930.
    Next we have the video encoding program X264, gives us an average FPS.

    As you can see the 2600K is faster clock for clock.

    Now here are the requested benchmarks, I had two requested, the first one is a Chess Simulator, it is very CPU intensive, and I took a score I think the person was looking for

    Again you can see that the 2600Kís frequency give it an advantage.
    Next we have a science benchmark, it is pretty old, but has some nice features, here you go, Science mark:

    Here you can see that the 2600K clock for clock is a very good contender, this is more of a real life benchmark as is the chess program, as it actually models molecules like the Arena program plays chess.

  6. #6
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    SLI Performance w/ NF200.

    Now I should say a little something before I show you results of X58 vs. P67w/nf200. That is that the X58 platform has native support for SLI, it does it naturally. P67 is not meant to replace X58, especially not X58 SLI. P67 natively has 8 lanes of 5GT/S PCI-E 3.0 which goes into 16 lanes of 2.5GT/s PCI-E 2.0 which X58 has 36 lanes of PCI-E 2.0. You can see that P67 is in a huge disadvantage, so motherboard companies have taken it on their own to deliver platforms with the ability for 2-way SLI and 3-way SLI. Now while you can run 8x, 8x SLI, it is slightly better to run with full 16X per lane (not noticeable), and if you want to do 3 Way SLI you have to have at least 24-32 lanes. So the NF200 PCI-E Bridge is added. This bridge takes 16 lanes and turns them into 32. The NF200 is especially good at allocating the lanes correctly, but it has its drawbacks, most of which is its addition of latency to the PCI-E bus.

    Theoretically a board with NF200 should perform worse than a board with native 36 lanes of PCI-E. In practice, well letís see what happens in practice. In theory with just going off what we know SLI with NF200 from 16 lanes of PCI-E should be much worse than true 40 lanes SLI.

    SLI Gigabyte GTX 570, two cards, both in 16x slots. Not OCed, at stock.

    First letís take a look at 3DMark Vantage, a very common benchmark used to generate scores that are used to compare systems as well as tell you how well your computer can play games. For this benchmark we used the performance setting. PhysX was left on auto, and it went to GPU, that is why CPU score is so high.

    At stock and clock for clock the P67A-UD7 bet out the X58A-UD5. It was hard to believe; I had to run everything 5 times, every one of these benchmarks was run 5 times. I couldnít believe the results, but do not worry it pans out in a little. The margin for the P67A win is extremely small; I would say they are equal.
    Now here is the same 3DMark Vantage with PhysX set to CPU.

    The same thing here, graphics seems to increase with CPU speed, but at clock for clock the P67A-UD7 is faster than the X58A-UD5. But again the margin is a few points, for my standards it has to be 75-100 PTS or above on GPU score for it to be better. I would say that in most cases they are equal.

    Now 3DMark 11 is a very new program, and well doesnít recognize SLI with GTX 570, some of those scores are worse than single card. But never the less I included them because I did all of them.

    You can see that the X58A-UD5 beats the P67A-UD7, and then ties it later on. Once again doesnít beat P67A by my standards. I believe that this program likes X58A-UD5 better than P67A-UD7; I can believe it does as well because it doesnít recognize my processor or SLI on my GPUs. BUT it is a benchmark never the less and has pull. I think we have to wait a few months because this benchmark becomes are legitimate as 3DMark Vantage. Right now we can tell that 3DMark Vantage P67A wins and 3DMark 11 X58A wins, I conclude this by saying they are pretty much equal.

    Now letís shift over to another DX11 benchmark, Unigine Heaven, it is just beautiful the way it benchmarks, actually nice to watch. Futuremark should sit down and watch. The best part is that Unigine is freeÖ. Again Futuremark.

    Unigine also gives us a score, but I was looking for FPS. You can see they are pretty much on par, also notice that at 4.2 GHz the i7930 ties the stock 2600K. That is a lot of great gaming power the 2600K has.

    Now we move onto actual gaming benchmarks, I only did two, the first one is HAWX DX11, and it has its own benchmark program.

    You can tell that this benchmark is pretty much set on giving you one value, unless you increase GPU performance. (These cards have their chastity belt in tact, later i will have a GTX 570 SLI review, where I break open the belt ). This is a good benchmark for GPU, as it gives a very consistent number, now the scaling got cut off, but the difference is that 2600K max FPS is 65 and 930 stock gives us 63. That is a TWO FPS difference, almost nothing. See that the average FPS did not change, that means that this is almost 100% GPU dependent, as the GPU configuration did not change.

    The last SLI benchmark is real world gaming, Call of Duty 4: Modern Warfare. This benchmark was done using fraps, on record for 5 minutes and then spit out an average FPS, min, and max. I start recording on the first level of the first mission. I did this test the most because I actually enjoyed playing the game (I never play games), so the results have been averaged, 5-6 runs for each CPU setting.

    SATA6G performance:

    Now up until now there were only really a few options if you had an SATA6G capable SSD. But the only SATA6G capable SSD to date is the Crucial C300, luckily for you guys I have one, and have done a comparison of the old Marvell SE9128 which was the best SATA6G controller until now that Intel integrated SATA6G into the PCH. There were a few problems with the Marvell controller, and they lied within early firmware releases, 4Kb Random data speeds were slower than they were on ICH10R (Intel SATA3G). Throughout this past year GIGABYTE worked hand in hand with Marvell to work on the firmware for the Marvell SE9128, through the past 6 months there has been a huge jump in overall performance of the controller on my X58A-UD5. Well those updates where right on time, as Intelís SATA6G does live up to all its glory. Right now, only Intel RST 10 drivers will work for this, or else you are stuck with MSAHCI.sys which is Windows stock driver for AHCI enabled drives.

    Here are side by side pictures of Marvell SE9128 with latest updated firmware VS Intel PCH67

    Marvell on the Left with MSAHCI.sys, and P67 on the right with iaStor driver

    Now as you can see the results are pretty close, if you know a lot about SSDs you know you are looking at a 20% increase on OS performance. How you ask? Well you see where it says seq, that stands for sequential, that means large files, when you use an OS, sequential comes in play when loading the OS, games, or other very large files. Yes this drive only breaks SATA3G border on sequential, but that is not why people buy this drive(well yea its why some do), people buy this drive for its fast 4Kb speeds(4K). That is the file size most used in random windows OS performance. Such as opening a program, using a word processor, loading the internet and browsing web pages. General OS performance is about 90% 4K speeds.

    Hard drives donít touch 20mb/s, let alone 10mb/s. This is what makes SSDs fast. A Sandforce drive can do about 20-25mb/s, and a C300 can do 25-32mb/s. ASSSD is very hard to get a score on the high side of a the secturm, but what you are looking at is the 4K performance of ICH10R on PCH P67. in another program like crystal disk the scores would be higher, 4k read would be able 30mb/s. The reason I am only using AS-SSD is because of its consistency over Crystal Disk. Look at the jump in 4K writes, that is impressive too. Intel finnaly did it, they gave us a perfect SATA6G controller. If you guys want more benchmarks on SATA6G just let me know.

    ASSSD is a program built for SSDs unlike hdtun and hdtach and atto.

  7. #7
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    GIGABYTE Programs

    Gigabyte packages many interesting programs, in my In-Depth preview of the board, I took a look at Smart 6 and its ability to store your passwords in the BIOS and Backup BIOS. Now I am going to take a look at the new EasyTune6 (ET6) and Dynamic Energy Saver 2(DES2).

    Easy Tune 6: Now there is not much change from versions of this program for other boards, but the things that have changed are worth noting. First off, easy overclock options, allow you to OC automatically to 4.1ghz.

    Quick boost is a very nice option, but as I stated in my OC Guide you can most likley OC to 4.5ghz without having to mess with anything but the cpu multiplier.
    If you are going to more Extreme overclocks, this next image might be useful. Here are the OC options for in-windows OC through easytune.

    EasyTune6 also has a new layout for temperature and fan monitoring, it draws nice graphs for you for your voltages:

    Now moving on, EasyTune also allows you to OC your GPU:

    Lastly we have Dynamic Energy Saver. This litte program is powered by the Intersil PWM, and you can see all 24 phases/Ēengine cylindersĒ in action if DES2 is turned on. Now you might be wondering why all those lights on the board went off when Windows Loaded, that is because they are all software controlled. This program will allow yout o turn those lights back on, and seeing the phase LEDs work is pretty cool. This board also has ACPI power state LEDs, they can be turned off in BIOS if needed.

    DES2 allows you to see how much power your system is drawing, now I donít know how accurate it is, but I bet it shows CPU power draw pretty well as it is run by the PWM that controls the power output to the CPU.


    I sum performance up into five categories: Performance, Functionality, Overclocking, Value, and Appeal.

    I once had a professor back in college, and he gave us a written exam, I aced everything on it I mean aced it. Well when I got my grade back he had given me a 90%, and when I confronted him he gave me such a crazy answer I just accepted it and took my A. He said, "Yes you got everything right, but 100% is for God, 95% is for me, and 90% is for you." What a nut, but I am not too far off, so no 10 point out of 10 for any manufacturer, because in reality, nothing is perfect.

    Scale of 1-10, I donít give 10s.

    Motherboard and Processor performance is excellent, that is all I have to say. Yes the increase frequency is what makes the processor very good, but also the fact that it operates at those high frequencies at stock is impressive, and the fact that the motherboard can take it there is just great. SLI performance is much better than expected. With two cards this puppy is as fast if not the smallest bit faster than X58 SLI, I believe this is because of the PCI-E controller built into the CPU is much better than that of the X58 IOH. Although it has less PCI-E lanes, it seems to really boast some amazing performance. Even with the latency increase from the NF200 it still matches X58. I would say that SLI performance is equal and that is much better than expected, because P55 SLI was really not equal to X58 SLI performance. Intelís PCH P67 really has a very nice SATA6GB/s controller that delivers exactly what people want, excellent SATA6GB/s performance. GIGABYTE has worked closely with Marvell, and even peripheral on-chip SATA6G is impressive compared to on-die. In terms of Turbo mode there is absolutely no lag from multiplier to multiplier that people have been complaining about on other boards. With this board you donít even have to use turbo, and the multiplier apparently stays fixed at what you set. Max stable BLCK I reached was 107.5MHz. Score: 9.9

    This board has more connectivity than I have ever seen. It even provides the ability to change SATA internal ports into eSATA, and vice versa. USB 3.0 is all over the place, two internal connectors and many in the back, I have never seen so many SATA ports. Instead of using the LAN connectivity of the PCH P67 GIGABYTE chose to use dual Realtek NICs, so you can team them and use them in parallel or serial. The heatsinks get pretty hot (the NF200 one does) and this is a good thing, as they are actually working. They are also very low profile, and there were no spacing issues. There is however a little spacing issue with RAM problem but it is just a matter of looks, and actually is like that for better performance. Score: 9.7

    This board is just sick, if max BLCK is quality of the board then this board is extremely high quality at 107.5, 107.7 set. I was able to take my D1 stepping 2600K to 5.2 GHz, the only problem was the heat from the processor, and well that is the processors problem, these might be lower TDP at stock, but OCed they really know how to release heat. 5.1 GHz is my 24/7 OC and that is more than enough. At a minimum auto OC of 4.5GHz you will have a really hard time finding any system that OCs that well. I was able to just walk up the multiplier to 5 GHz and the VID slowly increased. This proper implementation of SVID is just AMAZING. On what other system can you auto overclock like that? X58 systems will just over-volt, might even damage your processor too, so only try this on P67 with a K series processor. This board has CPU PLL Overvoltage option which only works for D2 stepping chips, but that is an option every Overclocking board should have. I have to give this board a 9.9 as it took my processor to its limits; I wish I had a D2 stepping so I could go for 5.5 GHz! Score: 9.9

    This board has a pretty hefty Retail price of $300-$350, but if does have an NF200 chip which is what you are really paying for. If you do not plan on SLI there is really not much of a reason to buy this board other than the power phases and features. I feel as if you have dual video cards, and a hefty PSU to power them, then $300 for a board is not that much to pay. Considering this is a mainstream product, that price is a bit high. You are paying for very high quality components; itís hard to find any board better built. The fact that Gigabyte manufacturers their own boards is a great thing, they are one of the last of a dying breed of motherboard makers. Score: 9.5

    This board is such a great change from the baby blue color of X58 and previous Gigabyte boards. I really like it, as do many people I know. One thing that is one of my pet peeves is that many black PCB boards look brown up-close, such as my P6X58D-Premium. Itís because of the copper of the traces and PCB that bleed through. With My P67A-UD7 there was none of that. This board is really black, and matches my GTX 570 SLI and Dominator RAM very nicely. I think GIGABYTE actually used an extra layer of black matte PCB to cover up those traces without bleed through. The Gold accents on the new heatsinks is very nice, they really did a great job with the looks of this board. Right now this board doesnít have a UEFI BIOS, which many users want. But the fact is you donít really use your BIOS much, itís not an OS. You use it to Overclock, and the truth is a traditional style BIOS is much better for overclocking in my opinion it easier to use a keyboard. Because of lack of UEFI I have to take of a fraction of a point, but the hardware is there so it will be implemented shortly. I should mention you CAN boot from 3TB+ HDDs.
    Score: 9.7

    Total Score: 9.75, this board is Excellent in all areas. It doesnít fail to exceed all my expectations.

    High Overclock Ability
    VRD 12 Certified (Very good SVID implementation)
    Excellent SLI performance
    Excellent SATA6G and USB3 connectivity
    Very stable board
    Nice board and heatsink color scheme

    UEFI not implemented yet (for most of you, I prefer no UEFI as I feel it just isnít stable enough yet)
    No IDE (overclockers like IDE)
    PCI-E 1x slot is blocked by NF200 Heatsink

    I would like to thank everyone at GIGABYTE for making this review happen!!! Without you guys we wouldn't have this amazing board!

    If you have any questions or comments, do feel free to PM me in private if you would like.

  8. #8
    Join Date
    Feb 2009

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    Wow, great review! Thanks for the time spent putting this together.

  9. #9
    Lsdmeasap's Avatar
    Lsdmeasap is offline GIGABYTE Guru
    Join Date
    Jan 2008
    PCB Island

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    Up there!!

    Nice work man, I can see this took a lot of hours to do!
    Last edited by Lsdmeasap; 01-15-2011 at 10:29 PM.

  10. #10
    Join Date
    Sep 2010
    Norther Virginia

    Default Re: Ultimate Sandy Bridge OC Guide + P67A-UD7 Performance Review

    hey thanks a lot man, i appreciate it1

Page 1 of 2 12 LastLast

Thread Information

Users Browsing this Thread

There are currently 1 users browsing this thread. (0 members and 1 guests)

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts