Step 2) Optimize Memory Frequency

System Memory Ratios

The next step is to find the limit of your memory.  In order to do this, we need to prepare you to meet your memory overclocking goal.  Start off this step be entering the BIOS and setting the bclock frequency back to 150MHz.  Because we have already been through the range of bclocks from 150MHz to 200MHz (or whatever YOU reached), we already know that the bclock will be stable with these settings.  So that effectively allows you to go back through the same range again, changing only the variable of memory, so that you’ll KNOW that that any instability is memory related.

To do that you need to change your System Memory Multiplier (SPD) to a ratio that will allow you to push the blcock back up to your goal, whichout exceeding the memory’s rated speed.  Also, manually configure the memory timings according to the manufacturer’s specifications.  There will probably be many sub-timings in your BIOS that you could change, but only change the timings specified by the manufacturer, the others leave on auto.

Lets look at the sample systems for an example, for those systems you would configure the System Memory Multiplier (SPD) as follows.

• Sample #1 – set the System Memory Multiplier (SPD) to 6
• Sample #2 – set the System Memory Multiplier (SPD) to 8
• Sample #3 – set the System Memory Multiplier (SPD) to 10

For the sample system #1, the goal is 180MHz bclock, the multiplier is 8, and the memory is rated at DDR3-1066.  If we were successful in reaching the 180MHz bclock goal, the memory would be at 180×6 = DDR3-1080.  This is about an 1% OC and within the 10% I recommend limiting yourself to for this step.

For the sample system #2, the goal is 200MHz bclock, the multiplier is 8, and the memory is rated at DDR3-1600.  If we were successful in reaching the 200MHz bclock goal, the memory would be at 200×8 = DDR3-1600.  This is the memory’s rated speed and within the 10% I recommend limiting yourself to for this step.

For the sample system #3, the goal is 200MHz bclock, the multiplier is 10, and the memory is rated at DDR3-2000.  If we were successful in reaching the 200MHz bclock goal, the memory would be at 200×10 = DDR3-2000.  This is the memory’s rated speed and within the 10% I recommend limiting yourself to for this step.

In the same way, set your multiplier according to your goal, so as to prevent more than a 10% overclock on your memory.  As a generalization, memory rated at lower speeds typically has more overclocking potential than higher rated memory.  My suggestion for now is to focus on getting the IMC overclocked to the memory’s rated speed (cause remember, the IMC is only rated to run at DDR3-1066).

Instead of overclocking memory, I place a higher value on tightening up timings.  Take sample #2 from above, it has DDR3-2000 CAS9 memory.  You’ll get better day-to-day performance out of that memory by tightening the timings down to CAS8 level, as opposed to overclocking the speed to DDR3-2100…..not only that, but tightening the timings is almost always easier than increasing the speed.  Again, I’m not going to get into detail here because tangible differences on a daily system will not exist.

DRAM voltages

There are two voltages we need to focus on in this section, the QPI/VTT voltage you’ve already become familiar with, and the DRAM voltage itself.  QPI/VTT is already high enough to support your bclock and uncore goals, but as we push up the speed of the IMC (which is also part of the uncore), the QPI/VTT voltage may need a bit more.  This is especially true if you are running with all 4 memory slots populated, or if you’re using newer 4GB DIMMs, these configurations can put a lot of added stress on the IMC and require extra QPI/VTT voltage to retain stability.  For all three sample systems, set the DRAM voltage manually to 1.65V (or 1.66V if 1.65V is not possible).

Testing for highest stable memory frequency

Now start up your selected test program, for example OCCT (RAM) or IntelBurnTest (maximum).  Run the test for just a short amount of time, I usually try to run 3 loops with IntelBurnTest (maximum).  Then reboot the system and return to the BIOS.

1. If the test ran without error, raise the bclock by 10MHz, reboot into your OS and run the test again.
2. If the test failed, raise the QPI/VTT voltage by 0.0125V, reboot into your OS and run the test again.

Continue to repeat this testing following the two procedures above, until you meet one of the following three criteria:

• You reach the same bclock speed you identified in step one and successfully pass your stability test.
• You reach your maximum safe QPI/VTT voltage (as identified above)
• Raising the QPI/VTT voltage is ineffective in stabilizing the system.

Just as you did in step one, if needed you can follow the same procedure for fine tuning here.

After you have reached your desired goal, be sure to test the overclock with another one hour pass of your selected stability test to ensure the entire uncore is happy with the current configuration.  However, I would add one additional stability test at this point, memtest86+.  IntelBurnTest (maximum) and OCCT (RAM) are both very good at identifying IMC and memory error, but neither is as strenuous testing the memory modules as memtest.  It’s a good way to verify that the memory modules are not defective, and will operate without error.  Run at least one pass, or for 1 hour, whichever is greater.

Congratulations! - You now have a relatively stable bclock frequency and memory frequency.

Overclock the Memory?

I’m not going to go into detail in overclocking memory in this guide.  It can be tricky and time consuming and with the differences in memory pricing, my advice is to buy RAM rated to run at the speed you want to run it at.  The one exception was with sample #1, in that case we ended up with a very modest overclock that would be easily attainable, and might even yield a tangible performance increase.

When you get up into the higher frequencies, memory overclocking becomes a bit futile.  Instead of overclocking memory, I place a higher value on tightening up timings.  Take sample #2 from above, it has DDR3-2000 CAS9 memory.  You’ll get better performance out of that memory by tightening the timings down to CAS8 level, as opposed to overclocking the speed to DDR3-2100…..not only that, but tightening the timings is almost always easier than increasing the speed.  Again, I’m not going to get into detail here because tangible differences on a daily system will not exist.

Step 3)  Stabilize CPU Frequency

Almost there

The last step in this guide is often the first step for users who run into problems and then troubleshoot for days afterward.  Leaving it to the last step makes the task much simpler.  You now have the following settings locked in; QPI/VTT voltage, IOH voltage, memory voltage, uncore clock ratio, memory ratio, and memory timings.  That means when we are looking for our highest CPU frequency, there are only two variables we need to play with: bclock and CPU voltage.

Please start by entering the BIOS and adjust the bclock back down to 150MHz, then adjust the CPU clock ratio as follows.

• Sample #1 – set the CPU clock ratio to x22
• Sample #2 – set the CPU clock ratio to x22
• Sample #3 – set the CPU clock ratio to x24

Load-line calibration (“LLC”)

This actually goes by a few different names, but they are all meant as a means to reduce or prevent v-droop.  It does typically ease the overclocking process at the cost of violating Intel’s design specs.  However, overclocking in its essence violates Intel’s design specs, so you’re not breaking any new ground with this feature.  I highly recommend enabling this feature.  High end Gigabyte motherboards have two settings, select “Level 2”.  For more insight on the theory of LLC, refer to this excellent explanation at anandtech.com.  There was also some real world testing recently; feel free to check out Bobnova’s LLC investigation here.

CPU Vcore

That brings us to the first thing that most users want to play with after powering up their new system for the first time: CPU voltage, aka “Vcore”.  As you can see, this is actually one of the last things you should be changing.  I would recommend starting at a nice and easy 1.2V.  Surprisingly enough, many users are able to achieve very good overclocks with this modest amount of CPU voltage.  I would be surprised if sample #1 needed any more than that, sample #2 probably need a bit more, and sample #3 will probably need quite a bit more.

Testing for your highest stable CPU frequency

Once the operating system has fully loaded, start up RealTemp.  Now start up CPU-Z and verify that your overclocked settings have been properly applied, and that you are running at your desired CPU, bclock, and memory frequencies.  Each of the sample systems should look like this.

• Sample #1 – CPU = 3300MHz, uncore = 3300MHz, memory = 450MHz
• Sample #2 – CPU = 3300MHz, uncore = 3300MHz, memory = 600MHz
• Sample #3 – CPU = 3600MHz, uncore = 3300MHz, memory = 750MHz

***Note***, if you have SpeedStep (“EIST”) enabled, the CPU speed will fluctuate in CPU-Z and the load changes, please verify values given above WHILE RUNNING YOUR STRESS STEST.

Now start up your selected test program, for example OCCT (CPU) or IntelBurnTest (maximum).  Run the test for just a short amount of time, I usually try to run 3 loops with IntelBurnTest (maximum).  Then reboot the system and return to the BIOS.

1. If the test ran without error, raise the bclock by 10MHz, reboot into your OS and run the test again.
2. If the test failed, raise the CPU Vcore voltage by 0.025V, reboot into your OS and run the test again.

Continue to repeat this testing following the two procedures above, until you meet one of the following three criteria:

• You reach the same bclock speed you identified in step one and successfully pass your stability test.
• You reach your maximum safe Vcore (as identified above), or maximum safe temperature.
• Raising the Vcore is ineffective in stabilizing the system.

As you did in step one and two, follow the same procedure for fine tuning with 2MHz increments.

Step 3.5) iGPU overclocking

So, if you are using the iGPU in your processor and would like to gain an extra nudge of performance from it, consider a mild OC for it.  The iGPU in these processors are typically good for 900MHz or more with modest increases in iGPU voltage.

I like to use the FurMark utility with all default settings except check the fullscreen box.  In the default “benchmark” mode, the test lasts only one minute and actually puts a pretty decent load on the GPU.  Just keep in mind that the iGPU freq shown in the BIOS is not the actual GPU freq.  If you want the iGPU to run at approximately 900MHz take 120,000 and divide it by your bclock frequency….this is the value you need to set the iGPU frequency to in the BIOS to achieve 900MHz actual iGPU speed.  I’ll use the sample systems for an example.

• Sample #1 – (120,000/180=667) set the iGPU freq to 667MHz
• Sample #2 – (120,000/200=600) set the iGPU freq to 600MHz
• Sample #3 – (120,000/200=600) set the iGPU freq to 600MHz

So, starting where you were at the end of step 3, go into the BIOS and increase the indicated iGPU freq by 10MHz.  Enter Windows and run FurMark.

1. If the test ran without error, raise the bclock by 10MHz, reboot into your OS and run the test again.
2. If the test failed, raise the iGPU voltage by 0.025V, reboot into your OS and run the test again.

Continue to repeat this testing following the two procedures above, until you meet one of the following three criteria:

• You reach the same iGPU speed listed directly above and successfully pass the FurMark benchmark test.
• You reach your maximum safe iGPU voltage (as discussed above).
• Raising the iGPU voltage is ineffective in stabilizing FurMark benchmark test.

As you did in the first three steps, follow the same procedure for fine tuning with 2MHz increments.

Is it stable?

So, once you find your highest CPU frequency by meeting one of the criteria above, and run OCCT (mix) or IntelBurnTest (maximum) for one hour minimum.  I’d like to refer to all this testing as “reasonably stability”.  In my experience, the true test of stability is the regular daily operation of your system, doing what you intended it to do.

The reason we use testing programs like OCCT or IntelBurnTest is to simulate a worst case scenario.  I have never seen any real world application come close to generating the heat that IntelBurnTest (maximum) generates.  But there are a few that can come close to OCCT, like folding or crunching for distributed computing projects, encoding high definition videos, or very heavy multitasking.  But even though OCCT or IntelBurnTest stress tests tend to generate more heat, they do not always find 100% of instabilities. In my opinion, the ultimate test is to use the system!

Final Words

Well, that about wraps it up.  Believe me, there is so much more to overclocking.  There are SO MANY settings you can continue to fiddle with, you may have a million questions at this point about all of the settings in the BIOS that we never touched.  They’re valid questions, but not meant for this guide.  My goal was for this guide to get you 95% of the way in 5% off the time.  Hopefully you’re there!  The other settings in your BIOS will be needed to get you to 100%….but if I included those things in this guide, it would be 3 times as long, and much more complicated.

Please feel free to comment, and post any questions in the Intel section of the techreation forums.  The community is the best resource to continue pushing your system beyond this guide.  Post a new thread with a very detailed description of which part of the guide you are having problems with, how far your progressed through the steps, and screenshots to help us understand the problem your encountering if necessary.  We will do our best to help you out with any problems you may encounter!

Thanks to Brolloks, dg170775, and others for their help with this guide.

A comparison of BIOS terminology used by four top manufacturers

Pages: 1 2 3 4