Ivy Bridge/Z77 Overclocking Guide
Ivy Bridge Overclocking is almost identical to Sandy Bridge overclocking in that it is basically a CPU which is meant to be overclocked through the multiplier and not the base clock (BCLK). Sandy Bridge overclocking brought a whole new level of simplicity to the overclocking realm, a user only needed to change a few voltages, and change some ratios and they were easily granted a huge performance increase. With Ivy Bridge things get a lot easier as the CPU overclocks a lot further with better cooling and is more optimized towards higher memory and base clock speeds, thus making ambient overclocking much simpler and easier for the average overclocker. There is almost no need to increase the secondary CPU voltages, such as VTT, with Ivy Bridge on air/water cooling as the memory controller can already push the memory up to its limits without this. The same thing goes for base clock, while with Sandy Bridge the max base clocks we saw were pretty limited, around 105-107 on average, almost all Ivy Bridge CPUs will do 110mhz easily with LN2 cooling, and will scale way above that with the cold. With Sandy Bridge we same some very odd clock walls, as well as limitations with the IMC in which the memory controller couldn't readily handle the maximum memory multiplier and BCLK increase over a few MHz from stock, and this limited overall memory performance. However Ivy Bridge is more unlocked than Sandy Bridge, it offers many more memory multipliers and even adds in a second divider so that you can run memory at different speeds in more friendly increments (like 2000 MHz and 2133 MHz). Ivy Bridge also doesn't have the invisible clock walls which Sandy Bridge possessed, the CPU can overclock under the cold and scales very well in all aspect with cold temperature. However under air cooling Ivy Bridge exhibits much higher temperatures during full load due to its 22nm process, which will probably only get better though cooling optimizations and better contact between the HIS and the CPU Die. We will explore why Ivy Bridge has such high operating temperatures on air OC. This guide can be used for all "K" series Ivy Bridge SKUs, I used a GIGABYTE board and a lot of what I explain and show is on GIGABYTE boards, but I will help anyone with a question and I write the guide so the principles can transfer.
Ivy Bridge Basics
- Ivy Bridge Basics
- The Science Behind Ivy's Thermals
- Step #1 CPU Overclocking
- Step #2 Memory Overclocking
- Step #3 OC Optimizations and In-Windows Tuning
- LN2 OC Preperation
- LN2 OC Hints Tips and Tricks
So to begin this guide a primer on overclocking the platform will be given as well as Intel's recommended voltages and my recommended voltages:
Intel Rec. Max is Intel's absolute maximum rating for the Ivy Bridge lineup, many of the numbers provided are identical to those of Sandy Bridge, however while vcore should be lower because of a better processing technology (22nm vs 32nm) it is max 1.52v here because of the SVID max. When overclocking on air the only two voltages you should need to touch on an Ivy Bridge setup are the Vcore (which you increase) and the CPU PLL( which can be decreased to help temperatures). You should not proceed to just apply the maximum voltage for the vcore, vtt, or system agent as you will heavily increase the temperature so much so that the CPU will throttle and can be damaged. Also if you start off with a higher temperature it is very hard to test stability, as you will probably be more unstable than if you used a lower VCore.
For 5GHz for instance, it is possible to OC to 5GHz with 1.4v on air:
However the wPrime score isn't very good as the CPU's heat is causing it to throttle a little bit.
5.3GHz is my maximum validation on air:
Temperature is more important for high clocks than voltage is when it comes to Ivy Bridge. Also under LN2 higher vcore might not yield a higher clock, as it will add more heat which can have an opposite effect. So while at 1.84v I might do 6.6 GHz if I increase to 1.86 I can only do 6.55, but if I lower the vcore to 1.83v I can still only do 6.55, it is all about working the volts very carefully. I should take a second and note that Ivy Bridge is an extremely tough CPU, it is very hard to kill, however you can kill it if you go above 1.6v on air and ~2.0v on LN2. Ivy Bridge also seems to be more resilient to degradation than Sandy Bridge was, however the heat produced by the CPU can cause degradations when above what Intel recommends.
TJ Max for Ivy Bridge is 105C, however you shouldn't go above 85-90C load when overclocking.
Ivy Bridge also shows a lot of power increase due to frequency alone as well. You can see this in the graph below which represents a 3770K with a fixed voltage of 1.4v. However at 5GHz it is still better off than Sandy Bridge-E and right around where Sandy Bridge is at for 5 GHz. What is increasing to increase the power is the current, you cannot control the current, but you can control the frequency and voltage.
(Power Scaling with Change in Frequency Alone)
The Science behind the 22nm 3D Transistor and how it can help us overclock!
So let me confuse you a bit and then unconfused you by simplifying everything. Now if you don't know this, decreasing the size of the transistor takes it down to a quantum level. Quantum mechanics is a scary word, but isn't very hard to understand, in very general terms it deals with all aspect of physics not covered by tradition physics, so it covers physics down to the molecular/atomic level. When we reach the 22nm size, we are dealing with quantum physics, and when we do this we can talk about the Hinesburg uncertainty principle, which basically states we cannot know where the electron will be at a certain point. That means that if the electron is outside of where it should be, then we have higher leakage. There is an equation where temperature and leakage are related, and while it is pretty complex, it does allow us to analyze certain points easily.
Sub threshold Leakage= A (W/L) (k^2/q^2) T^2 e^((-qV_t)/nkT) In more simplified terms this shows us that leakage increase exponentially with temperature, and that voltage also has a significant impact on increasing leakage. This has been true for almost all microprocessors, however on Ivy Bridge it is easy to see. So we can analyze Ivy Bridge's power properties in two ways, first we set a constant overclock and a constant voltage, and we take control of the temperature by decreasing the temperature at full load through the use of liquid nitrogen, and we measure the power input. The power input to the CPU will be reflective of the leakage, lower input power can be because of lower the wasted power and thus lower leakage, and the temperature we put on the CPU will be the temperature. We will then do this at another voltage with a higher frequency and see if the trend is affected.
We can see that not only is the temperature decrease having a great effect on the power consumption (representative of leakage), but also an exponential one, as at around -60C on both runs we see a leveling off of the power consumption. However as the temperature rises the increase in power is much more than it is when the temperature is lower. This confirms that the leakage on this CPU is very heavy, we can also see that the leakage is being decreased exponentially as we decrease the temperature.
So how can this help me OC? Well keep this in mind, for every degree you can reduce the temperature of Ivy you are decreasing the leakage at a faster rate than at the degree above it, when you do this you are increasing your opportunity for higher frequency at a much faster rate. So always keep pushing at better temperatures, with Ivy Bridge EVERY degree counts more than the degree above it. At around -60C this effect subsides, so phase change would be a point at which the power scaling starts to end.
Preface to OC:
Before you start overclocking it is important to know what type of memory and cooling you have, first you want to OC the CPU and then the memory separately as to not cause issues which are harder to pinpoint. After you change each setting you should use a stability test such as Prime95 or IBT to test for stability before going up another notch.
Step #1 Overclocking the CPU Frequency:
On Ivy Bridge overclocking is done through the CPU Multiplier on a "K" series SKU like the 3770K and 3570K and the multiplier is multiplied by the base clock. When you overclock the base clock you are overclocking the DMI and PCI-E busses as well, so you might damage or corrupt the devices hooked up to these busses such as your HDDs/SSDs and GPUs on the PCI-E bus.
CPU Frequency=CPU Multiplier X Base Clock
Memory Frequency= Memory Multiplier x Base Clock.
That means if you increase base clock you increase both memory and CPU frequencies, you also increase the iGPU's frequency as well. However with Ivy Bridge you shouldn't be increasing the Base Clock for Air/H2O overclocking as BCLK OC takes a toll on everything on the PCI-E bus including your GPUs and your SSD/HDDs, so it is pretty much reserved for benching with colder temperatures. Don't worry about increasing BCLK for memory speeds as there are enough memory multipliers with Ivy bridge/Z77 so you can always find the speed you want unlike with Sandy Bridge. However you can increase the BCLK slightly for high memory OCes where certain multipliers are better than others.
What I shoot for is a stable 4.6-5GHz OC with Ivy Bridge, such as something like this:
The temperatures are much better than that of the 5GHz shown earlier, topping out around 90C under full load, also the wPrime score, while the benchmark and OC aren't tweaked, shows it being faster at this speed as the temperatures are lower and the CPU isn't throttling.
With Ivy Bridge, you want to slowly
increase the VCore as temperatures will hurt your max OC much more than voltage can stabilize it. I would go one multiplier at a time sticking to my voltage ranges in the graph below. If you end up with too much heat then the logical thing would be to decrease the voltage, however at this point you can try to decrease the CPU PLL, and if that doesn't help much you can always decrease the VTT and System Agent (IMC) to levels where they are lower but still remain stable. When I was messing around with LN2 I could validate 5 GHz with less volts than my CPU needed at stock frequency, that is how much heat has an impact on frequency. However I do not want to show that shot as people might not always read stuff, but 5 GHz at 1.2v isn't impossible at -190C.
Below is a chart that shows the optimal voltage settings which you should aim to better:
You should try to fall under these voltage ranges or slightly above to stabilize your OC, these are the recommended voltages per frequency, however the CPU I used is very good it seems, so you might need more voltage than I did.
Above you can see where it says CPU/PCIe Base Clock as well as the CPU Clock Ratio which is the CPU Multiplier and the System Memory Multiplier. With just increasing the multiplier you can increase the clock speeds of the CPU up to about 4.2 GHz with 42x100.00. If you want a set 100 MHz even base clock it is best to set the base clock to 100.00. SVID will stabilize the CPU to about 4.2 GHz but not beyond that, so you will need voltage increase above 4.2 GHz.
If you want the best results you should disable power saving options like I have below, however if you want the CPU frequency to drop under idle conditions, you should leave them enabled. You should also leave them enabled if you will use DVID Voltage offset instead of fixed voltage. If you decide to leave on power saving options, make sure that you increase the turbo current limits for the CPU within the turbo settings list to 200A and 300W to totally maximize Intel Turbo limits; however this might not be needed.
Any overclocks above 42x will probably require VCore increase, and this can be done through the CPU Voltage menu:
You will also want to set LLC which is under the 3D power menu, the LLC should be set to Turbo for a slight droop, or Extreme for no droop at all.
The LLC on these boards is rock solid, what you set is what you get, and nothing other than that. If you want you can also mess with the other PWM settings, but that shouldn't be needed as Ivy Bridge doesn't pull enough power to warrant those changes under air cooling. I recommend a slight drop of voltage under load, this might help with temperatures.
Step #2 Overclocking the Memory:
Memory Overclocking on Ivy Bridge is extremely easy to do. You have to overclock with the memory multiplier along with the base clock, best set at 100.00 MHz for air OC. If you want more fine gradual increases you might try increasing the base clock.
The Z77 Chipset provides the memory multipliers of: 10.66x, 13.33x, 14.00x, 16.00x, 18.00x, 18.66x, 20.00x, 21.33x, 22.00x, 24.00x, 26.00x, 26.66x, 28.00x, 29.33x, 30.00, and 32.00x
. All of which this board provides. If you want high frequency it is best to use the 28.00x multiplier for speeds above 2800 MHz with slight BCLK increase.
Below is some very crappy Hynix, but it can do very high speeds on this platform:
The reason I show you such an uneventful score is because of the fact that the memory I used is rated DDR3 2000MHz Cas 9 T2 and this was done on a very easy BIOS and i took less than a few minutes. it just goes to show how easy Ivy Memory OC is.
The voltages you should change for high memory overclocking on Z77 on air is the DDR Voltage, and if you like you can try increasing the VCCIO(VTT) and VCCSA(IMC) the VCCIO (VTT) can help with memory OC, however you will also need to increase VCCSA along with it on these GIGABYTE Z77 boards (except on the Sniper M3). If you want to increase VTT you need to increase IMC voltage to within 0.005v below it, so 1.1v VTT would be 1.095v IMC on these GIGABYTE boards. However I didn't really need to change it much at all.
Memory timings are a bit trickier; you should use XMP and then loosen or tighten timings from there. However for Z77 GIGABYTE has tightened up most of the latencies involved to improve 2D efficiency, however this means that the max memory OC might not be as high as it can be, so below I am showing you how to loosen up all your memory timings for high clocks. The second timings are pretty much maximized, and the third timings start with TREFI, and the 3rd timings are what provide that increased efficiency here, and they are changed to 8, but at stock they are 3.
Step #3 Optimizing the OC/GIGABYTE OC Profiles
The GIGABYTE Z77 boards allow a user to set an offset instead of a fixed voltage, so instead of picking 1.4v, I could instead tell the CPU to drop its voltage in scenarios with less load, and increase during load. This is done with DVID offsets, which is an amount of voltage you select which is added to the CPU VID(the CPU's stock Vcore). However you need to also have C1E, EIST, and C3/C6 states enabled, as well as Turbo Mode enabled to properly drop and raise the CPU frequency.
Newer boards such as those from GIGABYTE and ASUS, use digital PWM technology, will have a special page where users can mess with and optimize PWM settings. To determine whether or not your motherboard has a digital PWM you should look in the power menu and look for settings such as those below, one the staples of digital PWM technology is a user configurable VRM.
Please note that you do not need to set LLC to Extreme, nor the others to extreme for only 4.8GHz, most of these options I change by habit for extreme overclocking with liquid nitrogen. For 99% of Overclocking there is no need to change any setting other than LLC.
However many of you want to know what the other settings do, and I will show be below.
PWM Phase Control:
This setting determines how to balance temperature with performance to provide either the best VRM performance or the best temperature for the VRM itself.
This is a setting which directly correlates with the transient response of the VRM, turning this to fast will increase the temperature of the VRM.
Load Line Calibration:
This setting can be increased in intensity which will decrease the standard Vdroop setting for the voltages, the CPU VCore LLC is the most important, and if you are OCing on air you should set Turbo and if on LN2 you should set Extreme.
Over Voltage Protection:
This setting determines the upper limit on how much voltage can be supplied over the maximum setting.
Over Current Protection:
This setting determines the upper limit on how much current can be supplied over the standard setting, I set to extreme always, just because the CPU uses as much current as it needs and no more.
This setting determines the upper limit on the MOSFET temperature of each phase. Each phases uses a tiny thermsistor to gauge its temperature, and that information as well as current are fed into the PWM to balance out the output across all phases evenly. This setting just allow you to not have VRM OTP shut down. There are upper limit protections which are not visible nor modifiable by the end user, and they should ensure a shutdown if the VRM overheats.
PWM Switch Rate:
This is the switching frequency, or more simply put the amount of time each phase can switch, increase this and you increase the amount of ON time of each phase, thus increasing overall heat, and increasing the rate at which the current is supplied to the inductor, however the VRM on this board is already optimized for auto. Even under LN2 with the UD5H I do not change this setting.
GIGABYTE Profile Sharing:
For Z77 many manufacturers have brought out all the tricks, as Ivy/Z77 is the first time in a year that anything exciting for the extreme LN2 crowd has come out, and manufacturers love the extreme crowd because it gives them a chance to test out certain aspects of their boards to the max. To facilitate a much more interesting and easy user experience GIGABYTE finally recently incorporated named profiles within their UEFI. However they also added in the ability to save your profile to a USB or SSD/HDD, this means that users with the same model board (example= Z77X-UD5H Rev 1.0) can swap profiles and help each other or share their memory settings and so on.
You can save a profile name for a total of 10 user profiles in the BIOS.
Or you can save a file and set its name to a USB stick and send it to a friend or keep it safe.
GIGABYTE's Tweak Launcher:
Known as GTL this program is where it is at for LN2 and even air OCers. It is a simple executable like RealTemp, in that it has all its drivers in its folder and takes no installation. You just have to have Intel ME drivers installed just like for EasyTune6 for BLCK and multiplier change. Make sure to run the program as admin if you want BCLK change, but this program makes EasyTune6 not needed for LN2 OC.
BTW if you like EasyTune6, it is fully functioning for Ivy Bridge
For memory timings you need to use GTL.
Ivy Bridge LN2 Overclocker's Guide:
Insulation is done very easily for those of you who have done this before; you are insulting for basically a full pot scenario. However everything that deals with LN2 OCing or OCing in general can and will void the warranty of your products, including but not limited to the motherboard, CPU, and memory. I take no responsibility for your actions.
First there are two methods I like to use, the first one is meant for use if you want to use the board after you have done LN2 OC, the second is for if you don't have time, and want to be extra safe with the board, however it isn't very reversible.
The boards this time from GIGABYTE have a lot of interesting OC features:
Method #1 Conformal and Eraser:
This method is the eraser method, in which the user will first coat the board with a thin film of conformal coat such as liquid electrical tape or some other silicone based conformal coat. Liquid electrical tape can be removed afterwards as can dragon skin. Then eraser is put in critical areas to displace the air so that condensation cannot form. In this method you will also either put grease in the memory and PCI-E sockets or you won't, if you do not then you should pay special attention to those areas. As you can see in the picture below, the board seems to have frozen solid, even the heatsinks still have ice on them, and it literally looks as if it snowed on the motherboard. However I was being lazy and didn't have proper airflow so I could show you guys how it looks. All the white on the PCB is a thick layer of liquid electrical tape, providing a barrier to water and condensation, of course condensation will form on the coating so you still have to dry the system. I was being lazy and I actually wanted good shots of condensation forming, so I didn't use a fan blowing air upwards away from the board, but you will want to do this. Just one fan blowing air up, sucking air from below the top of the POT is all you need to move the air away and help reduce condensation.
You still want to use paper towels in this method to form an air tight seal as well as catch ice from the POT. I did a bad thing, in that I didn't have a high RPM fan blowing upwards so that much less ice would have formed, instead I let the vapor from the LN2 condense and form ice to show you some of the worst case scenarios. You however shouldn't do this, you should always have a fan blowing upwards, and preferably not connected to the board or PSU so that it always stays on. LN2 doesn't stop boiling because your system is off.
Method #2 Grease and Paper Towels:
Grease and release method is when you use silicon grease, and basically cover everything but the socket in it, personally I don't grease the DIMMs or the PCI-E slots unless I am doing GPU or memory on LN2. Then you lay paper towels in a square layout such as that shown below, and then you mount the CPU pot on the paper towels. The paper towels will act like the eraser and help form an air tight seal between the POT's insulation and the board, but the paper towels also provide minimal insulating properties. However the grease when in all the small crevices will prevent water from hitting the electronics, which is the goal of this. So if you just use grease, you must be sure to cover everything. Except do not grease the CPU socket.
Now this picture is very interesting, you can see this is a UD3H, and notice how the frost as formed on top of the grease used. There is no conformal coat on this board, just grease on top of all the electrics. Grease is hydrophobic which means its molecules are scared of water, and thus they repel water. So you can literally wipe away the water and some grease as the same time, and then reapply more grease. Just as in the previous method, you have to let the board dry before using it again. A heat gun can hurry this process.
You can see that there is water on some of the chokes as this is after drying out. The greased system is easier to dry out than the eraser, because the paper towels are all that need to be removed. However there is no water in the socket, thus I don't use any grease in the socket.
Memory LN2 OC
With the Z77 platform it may be necessary to LN2 cool the GPU or the memory. This might be one of the first platforms where LN2 cooling the memory might be necessary to get WRs, and you might have to do it not on the newer memory such as from Samsung or Hynix, but on older kits such as PCS which can do tight timings up to 2800mhz where they are superior. Speed while very important for this platform, also needs to be complimented by nice timings which can occur when under LN2.
My LN2 OC procedures:
I tend to boot into Windows at -80C with 110.00BLCK and 55x and 1.7v, I will then get into Windows and boot up CPU-Z and GTL:
Then I will increase the multiplier and until I am at a point where I can slightly raise other volts and BCLK until where I want them. The GIGABYET Tuning utility and EasyTune6 both change settings very easily, but GTL is easier to use and is lighter than ET6, its whole purpose is Extreme Tuning. I can then proceed to tweak the timings, run a benchmark, then increase a few more MHz and run it again until I am happy.:banana:
I will also load up some profiles for different boards, like one with those memory timings in place. Always load optimized defaults after a BIOS flash and before OCing. :banana:
Tips for Memory LN2 insulation:
- Use Vaseline or little bit of silicone grease in the DIMM sockets, but be careful as too much grease can cause lower memory OC because of the fact that it gets in between the contact pins to the memory. A light coating should be enough to not cause issues. If it does cause issues Vaseline can be melted away.
- Stuff paper towels in between DIMM slots, and you need to put paper or grease into un-used slots as well.
- Only use a TINY amount of grease in-between the slots!!!!
- Don't forget to put in your memory sticks.
- Use the slots pictured below first as they are the first slots per channel.
- Post Code 15 or 51 could be memory OC related.
- LN2 helps the IMC
The memory gets cold even without any pot on top of them. If you have a fan blowing air up away from the board, then you shouldn't get all this frost you see.
Ivy BCLK OC:
Tips for BCLK OCing:
How to get rid of the Cold Bug:
- Increase VTT Voltage for BCLK OC over 110 in some cases
- Drop temp to -60 to -80C, and you can start with 105 or 110.
- If you are using the iGPU you will get stuck at around 111MHz
- If you are using a PCI-E Gen3 card, the 7000 series have some issue with PCI-E 3.0 and BCLK and cold, so you might want to change them to PCI-E 2.0 .
- BCLK change will require restarts, however in Windows it is okay to do a 3-5% raise in BCLK without restart.
- Post code 72 is usually BCLK related if you are pushing the BCLK.
- Your initial BCLK change will require a full reset of the PCH to properly initialize a BCLK change, however there is a 3-5% margin where there isn't a needed restart after that. With some BIOSes changing from 108-111 BCLK might not require a restart if you go in 1-2MHz increments.
Most CPUs do not have cold bugs; if they do then they also have a cold boot bug. The best way to proceed to maximize the cold boot bug is by first increasing the BCLK over 102-104mhz. When you OC with LN2 you have to increase the BCLK, and when you increase the BCLK you will also increase the CPU frequency which allows users to get above 6.3 GHz as the maximum Ivy Bridge multiplier is 63X which is readily exhausted. With my 3570K I increase BLCK to as little as 4 MHz and was able to change my cold bug from -120c to -150C and my cold boot bug from -70C to -140C. I would start with 106mhz BLCk and work up from there!
This voltage can be changed, now I have tested from 1.4v to 2.2v in 0.05v increments and found that CPU PLL had no positive effect on CB or CBB from 1.4v to 1.8v, and had a slight negative effect above 1.8v. The fact that it had a negative effect above 1.8v means that it has some effect, and that perhaps most CPUs are optimized for 1.8v, however if your isn't it could be nice to find out by testing every voltage in 0.05v increments. Through deductive reasoning we can think that this voltage might help.
A note on odd CB/CBB Experience
I will find my best settings, and then I will go for something and then restart. If I crash in windows, then I hit restart button on the board very quickly, and then I can proceed to restart without CBB. However some CPUs that have no CB, might have a CBB, and for some reason they will only trigger the CBB if the board has powered the CPU down for longer than a few seconds, so if you restart quickly enough you should be fine. This means that I can go into BIOS and change BCLK to 112 from 110, go through a restart (if needed as some BIOSes won't restart for only 2 MHz) and then I can have no CBB! But if I restart and wait like 30 seconds I might then have a CBB. With a 3570K I had a bad CBB or -70C and a bad CB of -150 but that was with 110BCLK, now if I hit the CB and crashed because of CB then I would trigger the -70C CB, however if I crashed because of OC or instability then I would not trigger the -70C CBB, instead my CBB would be -140C.
LN2 OCing Voltages:
Voltages to increase:
Many users just increase vcore, however very high vcore like over 1.9v might not be always better, as the vcore will increase the die temperature, and many cases the die temperature is more pertinent to the CPU frequency. Some users also like to increase the VTT and System Agent (IMC) to help increase the BLCK and memory stability, it is hard to find clear cut benefits to doing this, however it can help with certain clocks and CPU. You should use high Vcore for max clocks, but many CPUs do not like over 1.83-1.85v for 2D and 3D benchmarks because the high voltage increases the temperature of the CPU more. Ivy cares more for temperature than it does for voltage.
Voltages to decrease:
You should always try to decrease the CPU PLL and perhaps the VCCIO(VTT) and VCCSA(IMC) if not only to help decrease temperatures even more if you don't need them for BCLK and Memory stability. CPU PLL Might help CB by affecting the temperature of the die so that the CPU can run at a colder temperature.
Power Consumption under LN2:
If you take a look here we see the power draw (12v current monitor on the 8 pin connector) when running wprime 1024 on a 3570K on a G1 Sniper M3, what is amazing about this shot is that 234W is being provided to the CPU VRM through a 4-pin power connector powering a 6 phase VRM (identical CPU phase quality as UD3H and UD5H and Sniper 3). The CPU is at 6.1 GHz. This just proves there is no need for huge power delivery, like there is no need for 24 phases on these boards.
If you have any questions please feel free to ask.