Kaby Lake Overclocking Guide [With Statistics] - Overclock.net - An Overclocking Community
Forum Jump: 

Kaby Lake Overclocking Guide [With Statistics]

Thread Tools
post #1 of 5014 (permalink) Old 01-20-2017, 01:23 AM - Thread Starter
M'aiq the Liar
BoredErica's Avatar
Join Date: Jul 2012
Location: Bay Area, California, USA
Posts: 7,402
Rep: 418 (Unique: 272)

Kaby Lake Overclocking Thread




Haswell Overclocking Guide [With Statistics]

Skylake Overclocking Guide [With Statistics]


Welcome to the Kaby Lake Overclocking Thread. Nobody has made a thread to chart all of people's overclocks, so I decided it's time for me to make a thread. Hopefully you will find this thread useful. If you are new to OCN, know that you can open an image in a new tab to view it in full size. This guide is based on the Skylake guide but has been improved.


Disclaimer (Click to show)
I am a hobbyist with no electrical engineering background. While I've put thought into this thread, I am not responsible if your CPU blows up, your motherboard blows up, or your life blows up. Nobody really knows what voltage is 'safe' and what 'safe' means is vague. Everybody stresses their CPU in different ways and I'm writing this guide for free on my own free time. I can't afford to break multiple chips to even try to come up with a good answer. In a nutshell: I'm just some random dude on the internet.
Kaby Lake General Info (Click to show)
What is Kaby Lake?
Kaby Lake is the 7th generation line of consumer CPUs for the mainstream platform. The enthusiast platform can feature more connectivity (such as PCIE lanes and quad channel memory controller as opposed to dual channel) and more cores, but the best single-threaded performance remains at the mainstream platform. Of course, the mainstream platform is generally cheaper to boot. The two most interesting CPUs for overclocking remain the top i5 and i7 skus. They are the 7600k and the 7700k. However, the i3 7350k is new. It is an unlocked 2 core processor with hyperthreading. The price (as shown soon) is a bit overboard in my opinion.

How does performance differ?
When it comes to those two processors, in terms of performance the Kaby Lake processors are Skylake processors that are able to clock higher due to a refined manufacturing process (14nm vs 14nm+). This follows the current cycle of tick-tock-optimization cycle (process shrink, microarchitecture update, process refinement). This means IPC (instructions per clock) shouldn't change. Testing shows that 4.5GHz on Kaby Lake will be basically the same as 4.5GHz on Skylake. Whereas one can expect 4.7-4.9GHz on Skylake chips, the Kaby Lake counterparts can achieve 4.9-5.2GHz with similar voltages. Note that some testing by HardOCP suggests that power consumption is a bit less than Skylake with Kaby Lake. [SOURCE]

How are the 7600k and 7700k different?
Please reference this chart by PCPer in their review of Kaby Lake. You can find their review HERE.

Past experience shows that once overclocked the difference between the i5 and i7 chips are small. (Differences may be attributed to i7 owners having better cooling and deciding to overclock further, for example.) The usual 2MB increase in cache has minimal differences (probably 1% if that much at all). As usual the i7 part has hyperthreading, which is useful for most applications that could use more than 4 cores. Remember that this will cause the chip to be hotter. And finally, there is the difference in price.

How do the chipsets differ?
The release of Kaby Lake comes with the usual release of a new chipset. This new chipset is the z270 chipset. Kaby Lake processors will work on Skylake motherboards (which use the z170 chipset) with a BIOS update. In that situation, the new features of the z270 chipset which the CPU is capable of supporting will not work. They use the same socket (LGA 1151). This means cooling solutions that worked for Skylake/Haswell/etc will work here.

PCIE Lanes:
First let's be clear about what Skylake brought to the table. With Skylake there were 16+4 PCIE lanes. 16 of them are direct to the CPU and can get used for graphics cards. The extra 4 could be used by motherboard vendors however they want, but in practice allowed for a x4 PCIE SSD to run. Typically this was run with a PCIE NVMe SSD like the 950 Pro. Those 4 extra lanes are connected via the Direct Memory Interface (DMI) and could not be used for graphics cards. It runs at PCIE 3.0 speeds, allowing for up to 4 GB/s, but with overhead the figure might be around 3.5 GB/s. DMI lanes have higher latency than direct PCIE lanes but the differences should be minor.

With Kaby Lake we have an extra 4 lanes through the DMI, bringing us to 16+8 lanes. It is important to note that the extra 4 PCIE lanes through the DMI we get with Kaby Lake compared to Kaby Lake does not increase the amount of data that can pass through to the DMI. It is still ~3.5 GB/s. Those extra lanes simply allow for better connectivity (like having multiple 950 Pros). Esoteric setups involving strange RAID setups and such will not be covered in this thread.

With Skylake we saw the transition from DDR3 to DDR4 RAM on the mainstream platform, catching up to the enthusiast platform (x99 chipset). While with Skylake we had a few motherboards that supported both types of memory, it was not recommended. We've moved on past a year since then, so expect to run just DDR4. Fortunately, DDR4 prices have fallen a lot since it was only used on x99. Prices vary, but snatching DDR4 3200 RAM would involve a very small premium in the United States.

At this point Optane is not out yet. Whereas modern motherboards support Optane as a normal storage drive, z270 motherboards support using Optane as a cache drive to speed up slower drives. This seems to be similar to hybrid drives, which are sort of a SSD/HDD hybrid.

What about the chip Skylake PCBs?
To my knowledge the PCBs of Skylake and Kaby Lake chips are the same in thickness. A month or two after the release of Skylake there was a scare about a chip supposedly breaking under normal operation due to thin PCBs. This has not been proven and it has not been replicated since the incident. In other words, this is not a problem.

Delidding is the process of removing the Integrated Heat Spreader from a CPU package and reapplying the thermal compound found underneath. A popular option is to use CoolLaboratory Ultra (CLU) or Kryonaut Conductonaut as the compound. It is not recommended to use normal thermal paste for this. We are past the age of delidding via razor blade or vice, so please do not use such unsafe methods. There exist tools to make it safer (Dr. Delid, Rockit 88, Delid Die Mate). Once the delid is done the user might want to glue the IHS back on, and for that there are multiple options such as super glue (Der8auer recommends UHU High Temperature Silicone glue). If you do not want to delid yourself you can always purchase a delidding service from Silicon Lottery, and that includes resealing of the package.

Expect 15C better temps with a proper delidding job. It is possible to run the processor without the IHS. This is called 'bare die' and must be done with caution. If the cooler is mounted overly tight on the processor then the die may crack.

Is there any insurance for my CPU?
Yes. The Intel Protection Plan still exists. For $25-$30 you can stop worrying about overclocking leading to death of your CPU. This is on top of the warranty you get with a CPU purchase. Visit here for details. In general people found that Intel has been pretty lenient in accepting replacements.
Pre-Overclocking Information (Click to show)
Here are some things to consider before overclocking:
  • Coolermaster 212 Hyper Evo < Noctua D14/D15 < x60/x61 Kraken/H100i < Custom Loop
  • Thermal transfer is pretty bad because the die and the IHS are small, and without delidding it's even worse.
  • Do you want to delid?
  • Lower temperatures improve stability, which means one needs less voltage to stabilize. With fine grain adjustment of bclk small headroom can be exploited.
  • Ripjaws 4 were around before Skylake. There were some reports of those kits not overclocking as far as Ripjaws 5 of similar specifications.
  • Like Skylake, Load Line Calibration affects core voltage and adaptive voltage mode is not dangerous under heavy synthetic loads like Prime95.
  • C-states decrease the voltage and in turn power usage during idle. This has a marginal effect on SSD performance.
  • The recommended utility for looking at your stats is HWInfo available here.
  • Terminology Check:

    Uncore = Cache Ratio = Ring Bus (Not technically 100% true, but when people say these things that's what they mean.)

    BLCK = Base Clock

    100 (Base Clock) x 45 (Multiplier) = 4500MHz or 4.5GHz

  • Changing the base clock affects ram speeds. You will have to readjust the ram setting accordingly if you change the base clock.

  • It is typical during a launch for there to be new BIOs updates. Keep your eyes peeled.

  • If you Bsod, you can look at some details from the crash log. BluescreenView will pull up the information for you. If you want, you can download it here.
  • VID is the voltage the processor requests. Generally it is not a useful reading in HWinfo. Vcore when read in real-time in a tool like HWinfo is a measurement of the voltage actually given. When you put in 1.3v into core voltage in the BIOS, maybe only 1.25v is given to the cores under load. This discrepancy is called Vdroop. To counteract that you can simply raise the voltage you entered or you can use Load Line Calibration or LLC. This setting impacts the real-time Vcore reading and increases it. Voltage delivered can have very quick drops, so quickly that specialized gear is required to detect it. LLC helps counteract that.
  • Can't find Vcore reading in HWInfo? Don't know where F-clock (Fclk) is on HWinfo?
  • LL
  • If too many overclocks are going on at once or you change too many settings at one time it would be unclear what caused the crash.
  • If you are nearing the limits of your processor it makes no sense to increase more than one multiplier at once. You could increase by less than that via BCLK tweaking.
  • It may be helpful to write down settings and whether it passed or not.
Overclocking Instructions (Click to show)

Core Overclocking

The instructions for Kaby Lake overclocking is basically the same as Skylake's.


0. Update your UEFI.
1. Manually set your cache ratio and ram to stock. Don't even use XMP profiles. Hell, turn your GPU's overclock off.
1a. Decide on your voltage mode. (More information at the end of this spoiler.)
2. Try 4.8GHz at ~1.35v. It should work and be stable. For those who are very new to overclocking, this means changing the core multiplier to 48. Core multipliers are always integers (meaning no decimals).
3. Just go up a multiplier. Increase voltage if you crash during stress testing with our x264 test. Refer to the recommended voltage section. Check the temperatures for the first minute or two to make sure everything is okay. Crashing from unstable settings will not harm your hardware, but may break your operating system if done excessively.
4. Eventually you will find the highest overclock you can hit without breaking 1.4v and this overclock will pass x264 test overnight. Another option is to try Prime95 v27.9, but v28.10 is overkill. To read more about different stress tests and to access quick download links to them (including our modified x264 test), check the "Stress Testing" spoiler.
5. You now know how far your core can go without changing Bclk. What happens next depends on the user's patience. Pick either the lazy method or the patient method.
Blck, or "base clock", affects multiple things including: Core, cache, ram frequency, and Fclk. The core frequency tends to have a larger impact on performance compared to cache or ram frequency. That in turn has a larger impact on performance compared to Fclk. In other words, put more time and energy on the things that matter more.
The Lazy Method


  • Since at this point you already know your maximum core clock without changing the bclk, just change the cache to something that will be stable for sure, like 4.4 or 4.5GHz. Don't sweat the last 200-300MHz, it really doesn't make a difference to performance.


The Patient Method


Essentially what you want with base clock changes is this:

  • Bclk that is not too far from 100 to cause instability.
  • Bclk when combined with a core multiplier, gives you the absolute highest core frequency that is stable.
  • Bclk when combined with a Fclk multiplier (4, 8, or 10) has to result in an overclocked, but stable Fclk.
  • Bclk when combined with a right memory divider, gives you a ram frequency that is overclocked somewhat near its maximum.

    How do we achieve this?


  • Since you know your maximum core clock without changing the bclk, now it's time to figure out what it is with bclk changes. If you are at 4.8GHz that means 4.9GHz is unstable. Try a setting that results in a frequency in between 4.8 and 4.9GHz. While doing this keep an eye on cache, ram, and Fclk to make sure they remain near stock speeds and stable.
  • Then, increase cache multiplier until it crashes, and back off one multiplier. (We don't fine tune cache with bclk because we did it for core and core matters more.) Unlike Skylake, there is a decent chance that your cache will not go as high as your core.
  • Overclock your ram.
  • Set Fclk multiplier to something that results in above 1GHz clock speed with your bclk.

Let's look at things more closely. This is important particularly for those who don't know how bclk works.


Fclk & You (Click to show)

Fclk is a setting that has to do with the way the GPU contacts the CPU. The default setting is 1000MHz. Fclk is supposed to be a GPU-oriented setting, so CPU benchmarks won't show a difference. The difference between various Fclk settings is relatively small. It varies depending on the configuration (especially from GPU to GPU), but for a 980ti the difference is within the margin of error. Here are Anandtech's results from the Skylake era:

Fclk Benchmarks (Click to show)

Notice how different the gains are depending on the GPU used.

From what I can gather, the performance gains of overclocking the Fclk is really quite small, even on graphical benchmarks like Unigine Valley. If the difference in performance past 1GHz Fclk doesn't even show under some graphical benchmarks, the difference is really quite small.


The exact settings will vary because motherboard vendors love assigning different names to the same thing. The instructions below are meant for Asus z170 Hero boards but it should be similar enough to whatever you have to make sense.


There should be a setting to adjust the Fclk directly in your BIOS, allowing you to set the Fclk to 400MHz, 800MHz, or 1000MHz. 


For example, in the Asus Hero z170 UEFI, under "Tweaker's Paradise", there is an option called "FCLK Frequency for Early Power On". What this setting actually does is set the Fclk multiplier to 10, and if the base clock is 100, 100 x 10 = 1000MHz Fclk. In HWinfo, under "System Agent Clock" (refer to Pre-Overclocking Information spoiler), it should now read 1000MHz.


So, if you choose to only overclock the Fclk through the dedicated Fclk setting, make sure your base clock makes sense. If you have it set to 1000MHz and you forget about it and you go back to changing the bclk to overclock your core clock, you won't understand why you won't POST at 170 bclk. The answer is that your Fclk has been overclocked to an insane value.


So let me state it again: Your Fclk frequency is affected by both your Bclk and the setting you've chosen in the dedicated Fclk setting.


Fclk setting at 1000MHz (Fclk multiplier = 10)

Bclk set to 100MHz


10 x 100 = 1000

Fclk is 1000MHz



Fclk setting at 800MHz (Fclk multiplier = 8)

Bclk set to 110MHz


8 x 110 = 880

Fclk is at 880MHz


If you didn't pick a Fclk setting and you left it at auto in the dedicated Fclk menu, then your motherboard will likely try to adjust the Fclk multiplier so that you will not crash.

Understanding Base Clock (Bclk) (Click to show)

Fine tuning Core Clock with Bclk Changes:

Base clock x Multiplier = frequency in MHz


Recall that multipliers can only be whole numbers. If we only tweak the multiplier, we can only do 4.5, 4.6, 4.7GHz etc. What if I can do 4.5GHz but I cannot do 4.6GHz? Maybe I can stabilize at a frequency above 4.5GHz but below 4.6GHz. To do that we change the Base Clock. The bclk can contain decimals (like 100.1MHz, etc).

Let's say we want to try 4.545GHz. Pick a number relatively close to 100 that when multiplied by something gives us 4545. 101 base clock with 45 core multiplier will do just that. Could I have done 181.8 * 25? Yes. But in general we want to keep bclk as close to 100 as possible to improve stability and decrease headaches.

Old OC:
100 x 45 = 4.5GHz Core clock
100 x 40 = 4.0GHz Cache clock
2133MHz Memory clock
100 x 10 = 1.0GHz Fclk
Base clock set to 101:
101 x 45 = 4.545GHz Core clock
101 x 40 = 4.04GHz Cache clock
2154MHz Memory clock
101 x 10 = 1.1GHz Fclk
If the above passes, we now have a stable 4.545GHz. Maybe we could aim for a higher frequency. Remember that core matters more than cache or ram which matters more than Fclk. This is why it's best to alter base clock to maximize core and simply raise the cache multiplier as far as it will go afterwards. Keep in mind that every time you adjust bclk you are also adjusting cache, Fclk, and ram. Adjust their multipliers accordingly. In my example the base clock change was so slight such adjustments were not necessary.
The higher you go from 100 base clock, the harder it is to stabilize. Generally the stability at 170 bclk and up will vary depending on the motherboard. You will sometimes fail to boot if the bclk is too high. There's usually no good reason to set the base clock above 170 though. With smart math, it should be possible to get very close to any frequency without exceeding 150 bclk. Don't forget that bclk can have decimals.
Ram Overclocking Basics (Click to show)
Ram Overclocking
Once both the core and cache ratio are set to stable and overclocked values you don't want to touch anymore, go ahead and overclock your ram. Don't forget that timings matter as well, and the "tighter" or the smaller the numbers are, the better. According to Asus, System Agent and VCCIO voltages can help stabilize a ram overclock, although more isn't always better. The ram itself could use some extra voltage. The default is 1.3v, with 1.35v being safe and 1.4v probably/possibly safe.
Here are rough guidelines for figuring out how your ram is doing for those too lazy to benchmark:
Ram can have lower latency or higher frequency. To figure out how good your latency is, simply divide your frequency by your CAS latency. This nets a number that Anandtech calls 'performance index'.
Obviously the higher your frequency, the better.
Generally you want lower latency while slightly favoring the higher frequency option when the performance index is similar. Run Memtest to see if your ram is stable. It's actually possible to simply tweak ram and see if the overclock holds up under use over time, but this requires over a week of computer usage doing a variety of tasks.
Final Step:
Go back and see if your overclocks still function perfectly with less voltage. How low can you go? This is just fine tuning of your voltages.
Safe Voltages (Always TENTATIVE):
Vcore: 1.45v/1.37v
VCCIO: 1.25v/1.2v
System Agent (SA): 1.3v/1.25v
Vdimm: 1.4v/1.35v
The first value shows voltages a pretty ballsy person can use. The voltage after the forward slash shows voltages for regular users who don't want to live on the edge. Refer to the disclaimer spoiler.
Voltage Modes (Click to show)

Before talking about voltage modes and power saving modes it's useful to think about what 'auto' voltage mode means. Auto means the CPU or the motherboard will try to assign what it thinks is the correct voltage for a given situation. It is down to multiple factors, and some of those factors are out of our control or are opaque to us. When a person only changes the core multiplier in a sloppy overclock they are letting the CPU decide with auto voltage mode. Offset and adaptive voltage modes seek to build on that behavior whereas manual voltage mode stops that behavior. The reason why it is not recommended to leave voltage up to auto mode is because auto mode is not perfect. In some cases it's very possible to achieve stability but the auto rules are too lax with the voltage, or vice versa.

Offset Mode
With this mode, you can add an offset to all of the voltages the CPU/motherboard would have used. If at a given frequency the auto mode would have used 1.3v, adding an offset of +0.01v would result in 1.31v used in that situation. You can set positive or negative offsets. This offset applies to all situations all the time.

Adaptive Voltage
Whereas offset mode simply shifts the entire voltage curve up or down, adaptive increases voltage to a set amount when core frequency goes above stock (meaning heavier load). Otherwise, the voltage curve will be default behavior unless the motherboard allows for an offset in adaptive mode.

Manual Mode
Manual mode means a fixed voltage is delivered to the CPU at all times regardless of load. If frequency and voltage are held constant then the best possible performance is possible (although the difference is not very big).



On a 7600k at 5.1/4.8ghz with 1.4v Vcore, idle power decreased from 72w to 54w. The more background tasks on idle, the less the CPU is able to downclock and downvolt, meaning less savings. My testing doesn't even include having Chrome open while testing. The core temperature also decreased 3C under 24C ambient temperature. It's also worth noting that downclocking/downvolting settings in UEFI should be paired with balanced power plan settings in Windows. Also, on Asus motherboards the minimum cache ratio must be set to a low number in order for voltage and cache frequency to downclock.


Other Power Saving Info (Click to show)

The following settings are only relevant for those not using manual override voltage modes to keep frequencies and voltages at maximum.


CPU SVID Support
This setting allows the CPU to talk to parts of the motherboard when it comes to voltages. This must be on for adaptive or offset voltage modes. Some software require SVID to be on for some readings, like CPU power draw in HWInfo. If neither points apply to you then set this off.

The CPU can enter various low power states. There are multiple C states, varying from C0 to C6. C0 is the default state and the higher the number, the lower power state in general. When parts of the processor powers down it will take time for it to ramp back up to full speed again. Note: I have been unable to detect an appreciable difference with or without C-states on Kaby Lake platform since adaptive voltage mode with balanced power setting seem to do the same thing.


Speed Shift

If the operating system and CPU supports it, the downclocking process can be handled by the CPU itself. This allows it to more rapidly downclock or upclock as situations change. This is more relevant for mobile, since burst workloads can be completely more smoothly from idle. Normally this should be on if the user is using adaptive or offset voltage modes.

Other Settings (Click to show)


Spread Spectrum
There are regulations about how much ElectroMagentic Interfererence consumer electronics can cause. To pass these regulations 'spread spectrum' was introduced. Practically speaking this should be disabled when overclocking. It will cause the core frequency to be unstable at a micro level. The frequency changes are way smaller than a multiplier's worth of changes, so much less than 100hz. However, these small changes are not ideal in an overclocking environment. CPU Spread Spectrum affects the core clock while BCLK Spread Spectrum affects base clocks.

Asus Multicore Enhancement, etc.
These settings are on multiple motherboards and will have different names depending on the motherboard vendor. Multicore Enhancement is the turbo boost settings you see on processors. A CPU may clock at a base frequency (say, 3.5GHz) by default, but turbo up to 3.8GHz under load. How much the turbo is could depend on things like temperature or the amount of cores used. While Intel has guidelines for the behavior, motherboard vendors are free to put their own spin on it. Often what these company-specific settings do is enable a more aggressive form of multicore enhancement. This lets motherboards show up higher in motherboard benchmarks. But because we are overclocking in Overclock.net in my overclocking guide, this setting has no benefit. To my knowledge there are no positives or negatives associated with these types of settings for our purposes.


AC/DC Load Line
This setting seems to be related to LLC (discussed later). It only affects users with adaptive voltage mode, improving spikes and drops. Asus recommends setting these settings to 0.01.


Load Line Calibration (LLC)
By default voltage when the CPU is under load decreases, making it lower than what the user sets in the bios. On idle it would be the same on manual voltage mode. Voltage drops under load on purpose and is called Vdroop. The Voltage Regulator Modules (VRMs) cannot react instantly to a load starting or ending. This causes voltage under shoot (voltage falls too low) when a load begins and over shoot (voltage spikes too high) when a load ends. The purpose of Vdroop is to lower voltage under load to avoid this spike, since spikes in voltage can be harmful to the processor even if it is brief. The changes in voltage can occur way too quickly for software to measure, and instead would require an oscilloscope. All software can detect are the idle and load parts of the graph which I've labeled.  Please refer to the picture below.




Say I want 1.25v under load. There are two ways of achieving that goal. I can raise Vcore such that Vcore under load is 1.25v, meaning idle voltage is higher than that. That simply shifts the curve upwards.


I can instead use LLC, which won't touch idle voltage and instead jacks up voltage under load. LLC improves under shoot (which can potentially render an overclock unstable) and worsens over shoot (which can harm a CPU). The improvements in under shoot and severity of the over shoot depend on the switching speed of the VRMs (which can be altered in bios on some motherboards). As the VRMs switch faster they will be less efficient (requiring more power and emitting more heat), although on the mainstream platform that is usually not a problem with normal case airflow.


Of course, it's possible to use a combination of both methods. LLC settings typically come in different 'levels', with minor LLC boosting Vcore under load only slightly, and extreme LLC drastically decreasing under shoot and increasing over shoot, with Vcore under load higher than at idle. My recommendation is to set LLC such that Vcore under load is a little bit under idle. I would not set LLC higher than what is needed for Vcore under load to equal Vcore at idle. Most people have a natural tendency to set Vcore under load as equal to idle, but there is nothing magically better about that setup.

Finally, it's worth noting that LLC levels are not a standard. "Level 2" can mean whatever the motherboard vendor wants it to mean. But on Asus z170 Hero, LLC level 4/5 is generally enough to set load voltage around equal to bios voltage (which, again, is not automatically the best setting).



CPU Current Capability
This and the settings after this are generally not relevant, but let's discuss them anyways. This setting tells the motherboard to stop interfering even if the CPU is drawing far more current than normal. A notable case of this setting making a difference is with Der8auer and Tiny Tom Logan's testing with x299 Asus motherboards. Running Prime 95 would suck up so much power, it causes things to shut down. Setting it to 140% prevented that, although it then allowed the motherboard VRMs to get overwhelmed and throttle the CPU. This setting should simply remove limits, so any decreases in longevity of the CPU should be indirect instead of direct.


CPU VRM Switching Frequency
As explained in the LLC section, a faster VRM switching speed allows the motherboard to adjust voltage more quickly, preventing nasty dips or spikes in voltage. Faster switching speed with lots of Vcore and LLC is harder on the CPU VRMs, so monitor VRM temperatures in HWInfo via the T1/T2/Temp2/etc sensors. Passive airflow from a case beats an open test bench.


CPU Power Phase Control
By default it seems phases are put in a lower power state when idle. You can set it to be engaged all the time, or to cause the motherboard to spring back to life more quickly. Note that this setting has not been tested.


PLL Bandwidth

This is also known as CPU PLLs OC in HWinfo. Some Asus motherboards have weird auto rules that involve increasing this setting (found in Tweaker's Paradise) far too much. The UEFI description suggests setting this to level 6 to 8 when overclocking CPU or blck heavily. All that was observed was CPU temperatures went up, with instability creeping up. On z170 Asus Hero with 7600k this auto rule problem can be reliably encountered as soon as CPU frequency hits 5.3ghz or goes above it. Setting PLL Bandwidth to level 0 allowed the overclock to continue without crashing despite the UEFI description.

Stress Testing (Temperature Chart, Quick Download Links, etc) (Click to show)
Quick Word About "24/7 Stability and Safety"
24/7 stability is a request people often make without thinking about what it actually means. '24/7' refers to the amount of time a CPU is spent under load. It says nothing about the type of load. Playing video games a couple of hours a day or week is not the same as hammering your CPU at 100% load for hundreds of hours in a row. If you're really that concerned about CPU longevity you shouldn't be using Prime95 to stress.




As the settings chart notes, I detected temperature fluctuations in Linpack, IBT, and XTU stress even though the load on the CPU still read 100%.


Note how XTU stress is cooler than XTU bench, and AIDA64 varies in temperature wildly based on the settings checked. Without a way to loop the test, applications like XTU bench and Cinebench are not viable stress tests. As expected, custom x264 at 16 threads is hotter than the 4 thread setting, and using more memory for Linpack causes a hotter test. Tests that allow the usage of more ram tend to be more difficult to pass the more memory your let it use. I had 16GB of ram.

Please note that some stress tests hammer different parts of the CPU harder than others. Exactly which is toughest on what is not totally proven.


My temperatures are lower than what most people will observe because I am not running hyperthreading and my chip has been delidded. My case as good airflow.


Below is a hierarchy of stress tests, listed in order from hardest to pass to easiest to pass. Anything that is counted as easy to pass or even easier are not recommended and will not be enough to be entered into the main overclocking settings chart. More details in the charting form spoiler.




Linpack (Max) (From Intel's website, not from OCCT or any other place or XTU.)

P95 28.10



P95 27.9

IBT (Max)



x264 16T

ROG Realbench



Stockfish (Chess, BMI2 version)


Aida64 (Full Suite)


Walk in the Park:



Booting into Windows



x264 is the recommended and the default go-to stress test for this thread. If you feel the need to use a hotter test that is your right, but know that your overclock may be hampered by that choice. You could forego delidding in many cases simply by switching to x264. The downside to this method is that the overclocking process will take longer because we are replacing a very stressful program and a short test duration with a less stressful one and a longer duration.


I recommend running x264 looped all night as you sleep, and if it passes, it's stable. Run it, sleep, wake to see the test still running, pass, smile. Angelotti and JackCY have tweaked the x264 Bench utility and turned it into a stress testing tool. You no longer need to download other programs to get it to work; just download, unzip and run. Simply put, our version of x264 test is better in every way to the original x264 benchmark. There is no reason to use the original utility. There is a readme inside to tell you what options to pick but I will also summarize it here: By default, try the 16 thread setting (yes, even if your CPU is an i5) with normal priority.

Regardless of which test you choose, you will probably crash multiple times when finding the right settings. Every time your computer shuts down unexpectedly there is a chance the operating system will be corrupted. The chance is relatively low. You may run /sfc scannow in run prompt in Windows to have the operating system try to repair itself, although it's not proven that it does much.


Prime95 Specifics (Click to show)
Prime95/OCCT Specifics:
When you are closer to stability, Prime95 may stop with an error. This is a rounding error, meaning the crash was minor enough so that your computer itself does not crash. There is some data to suggest that Prime95 gives out rounding errors very frequently, even in overclocks considered functionally stable. With Skylake, unlike Haswell, version 28.10 is not significantly hotter than version 27.9. Still, v28.10 has been shown to crash unstable overclocks much faster than 27.9, so consider it a harder test.
There isn't conclusive evidence so far about which setting in Prime is the most stressful and prone to crashing unstable overclocks. It is known that smaller FFT sizes tend to cause higher temperatures. (Small data set is a similar story for OCCT.) 8 is the smallest size (in K, but that's a technicality). Here is a picture showing how to set your own FFT size:
OCCT 4.5 is tougher to pass than previous versions according to its change log, so keep that in mind.
IBT/Linpack Specifics (Click to show)
Linpack/IBT Specifics
Linpack is a newer version of IBT. These tests have a "warm up period" where CPU load is relatively light, and an extremely thermally intensive period. When running these tests check termperature for a good 5 minutes because the temperatures in the first 10 seconds do not represent peak temperatures at all.

Below among the list of stress test download links there is a link for "Linpack Package". I have taken Linpack and added Linx GUI to it. You can now use Linpack as Intel originally intended or run the GUI to easily change the test settings.



Stress Test Download Links


Custom x264 with Loop Functionality and Other Improvements v2.06
Custom x264 v2.07

Aida64 5.80.4000

IBT v2.54

Linpack Package v1

OCCT v4.5.0

Prime95 v27.9

Prime95 v28.10

ROG Realbench v2.43

XTU v6.2.0.19

y-Cruncher 0.7.1

Latest Version of HWinfo (Monitors temps, voltages, etc.)

Memtest v6.2.0 (For testing ram overclocks.)





'I must pass all stress tests!'
So if I made a program that crashes you at stock clocks, you would feel compelled to underclock your CPU, even if that application in no way represents real-world usage? Passing "all stress tests" really means passing "all stress tests that people happen to have made". If nobody decided to make ultra-mega-Prime95-on-steroids, you would think your overclock is stable. That seems like a random, haphazard way to figuring out if your overclock is stable or not. Computers are built for using, and what really matters is whether you crash often enough while using it normally. Forcing yourself to pass a stress test "just in case you use it to its limits" makes no sense either. No point in going down "what ifs" which have no signs of ever happening. And if it does, work it out when it does.
Run 2 different types of stressing programs, and then use your computer normally. If you crash, then it's not stable. What's stable for you might not be stable enough for me. Some people need 100% reliability because of their jobs. 
Let's not get into a semantic debate about the word 'stability'. If you define stability as 'never crashes on anything, ever', then I don't care about your notion of stability. That criteria makes no sense either because the only way to be sure you are stable forever is to test your CPU forever. The world doesn't end if your CPU crashes on you. Run a stress test overnight, then go play video games to test things out. If you ever end up crashing in the heat of the moment, lower the multiplier by one and you should be perfectly stable.
You know your own use case and tolerance to problems better than anyone else.
Cache Frequency Doesn't Matter Much. (Click to show)

I have redone benchmarks to test the performance difference between a high and low cache frequency.

As you can see, a decrease of 100MHz in core clock has a larger impact on performance than a 1,000MHz decrease in cache frequency. Therefore, my position on cache frequency remains unchanged: It is a secondary setting that you should only overclock and worry about once everything else is done. Cache frequency seems to affect ram benchmarks as well.



If you'd like to see Haswell cache frequency testing, open this spoiler.

Haswell Cache Testing (Click to show)
Credits to Maxforces for the second part of the benchmarks. From my personal benchmarks, I found the drop of 0.7Ghz for the cache to be an equal performance hit of 0.05Ghz decrease in core clock and this difference shows in a very CPU reliant benchmark like chess.




And here are the most recent tests for cache frequency that I have done:

The 4.2 vs 3.4 is the cache setting. The core multiplier for this test was x45.


Testing methodoloy in this test is much more well documented by me.

Chess: Houdini 3, 9mb hash, starting position, 5 minutes.

BF3 Multiplayer: 64 player server in a closed map (Canals). Regular gameplay for entire round.

BF3 Campaign: Second misson, following scripted NPC movement.

Enemy Territory: 30 vs 30, Fueldump.

Runescape: GE, World 3. Capturing FPS while stationary. Max detail, non HTML5. x4 AA Bloom enabled. (It seems to use CPU to do AA)

Oblivion 1: Walk out in the wild, through Oblivion gate, to town gate.

Oblivion 2: NPC combat in Imperial City. Several guards/NPC vs Umbra. Spawn 50 player copies and begin combat once Umbra dies.


These were done on tests, as you can see, that vary from CPU benchmarks to CPU reliant games.

Maxforces Says:
Test setup












but if you play 3dmark you will gain some points


Troubleshooting (Click to show)
Some of my cores are hotter than others. Is this normal?
If the variance from hottest to coldest core is 10C or under, I would call that normal. If it's greater, consider doing a re-paste.
Ram XMP profile doesn't work.
Make sure the motherboard bios has been updated to the latest version. If that doesn't work, try adding a bit more ram voltage, SA voltage, and VCCIO.
Monitoring software shows incorrect data.
Make sure you are using HWinfo and make sure it is the latest beta version.
Prime95 stopped and says there's an error.
Most likely it is a rounding error. This means you've failed the test, but in a more minor way such that the computer doesn't crash. There is some data to suggest that Prime95 gives out rounding errors very frequently, even in overclocks considered functionally stable.
My temperatures are through the roof!
  • Stop using IBT or Linpack or Prime95. Use our custom x264 test or use something similar like ROG Realbench,
  • If your temperatures from core to core vary over 10C, consider a re-paste of your thermal paste.
  • What ambient temperatures are we talking about here? Are you sitting in an oven?
  • Hyperthreading makes your CPU hotter.
  • CPU delidding service is $50 in the United States from SiliconLottery.
  • I recommend using D14 or better in terms of cooling.
  • Ensure the cooling solution is mounted properly.


My CPU is downclocking under load!

Check your motherboard's power settings. Set them to max.


The Vcore is far higher than what I've set in the BIOS under load!

Your LLC is probably being overly aggressive. If possible, please set it to a lower amount manually.



Sample Size 61  
Average OC 5.03 Median OC 5.00
Average Vcore 1.36 Median Vcore 1.36
Extra Statistics (From the Bottom of the Chart) (Click to show)
None yet!
Charting Form to Submit Your Overclock to the Chart (Click to show)
CPU Model:
Base Clock: Bclk.
Core Multiplier:
Core Frequency:
Cache Frequency:
Vcore in UEFI: This is the CPU core voltage value you input into BIOS/UEFI.
Vcore: This is the *average* CPU Vcore reading from Hwinfo or HWMonitor under load. "Load" depends on what you're stressing.
FCLK: Reminder: In HWinfo, it is "System Agent Clock".
Cooling Solution: If you are delidded, note it here. Please explicitly state if you are doing bare die.
Stability Test: Please list the version of Prime95 and what FFT preset/size it is if you are using Prime95. Please list the number of threads used if using custom x264 test. In other words, please provide as many details as you can. Acceptable stress tests will be listed at the bottom.
Batch Number: What country? Please list the entire batch number if you can. You can find it on the box.
Ram Speed: State the frequency and timings (3200 16-16-16-35, etc etc)
Ram Voltage:
VCCIO: If not tweaked by the user, this may be left blank.
VCCSA: If not tweaked by the user, this may be left blank.
Motherboard: If possible please list chipset first.
LLC Setting:
Misc Comments: If you are submitting with AXV offset note it here. 5.0 w/ 1 offset = charted as 4.9.
To be charted in the main chart, you must fulfill one of the following requirements:
Prime v28.7 1 hour
OCCT v4.5 1 hour
Linpack from the Linpack Package download run at max settings (not recommended) 2 hours
Prime v27.9 3 hours
IBT 3 hours
x264 16T 8 hours
Realbench 8 hours
Aida64 and XTU do not count no matter the length of the test. When running tests it is assumed you are using the most amount of ram you can when prompted to choose. It is also assumed that you did not touch AVX voltage settings in the BIOS.
You must submit a picture showing that the stress test has been completed as you claimed (both the test and the duration). You also must have HWinfo open, showing both the frequency and Vcore. (Many people forget to make sure the Vcore reading is showing.) Failure to comply with every step results in being charted into the secondary chart at the bottom of the spreadsheet and your data will not be counted in statistics.
Changelog & To-Do List (Click to show)
Improved information on power saving settings.
Added info on PLL Bandwidth.
A while back some extra motherboard settings were added to the guide and explained.
Added info on delidding kits.
Improved info on LLC.
Improved info on SVID.
Added info on AC/DC Load Line.
Made a note on thermal transfer of mainstream CPUs in general.
Specifying small data set for Prime and OCCT in stress test difficulty hierarchy has been removed.
Renamed cache frequency spoiler.
More motherboard settings have been added.
Some typos fixed.
Removed 'to do list' since it was always empty.
Chart's bottom link got fixed.
Added post reference to chart.
Fixed references to an old Prime version. Fixed "10" bclk typo, meant 101.

Edited charting form again...
Fixed an error in chart.
Fixed charting form.
Fixed errors in chart.
Charting requirements revised.
Added HardOCP power draw data.
Removed some data that was too Skylake-specific.

Thread is created.


Thank you for checking out my guide!

Feel free to ask questions or provide suggestions!

Please read the guide before asking questions though!

Please do not PM me unless you think I've missed your post!
I love using exclamation marks!

BoredErica is offline  
Sponsored Links
post #2 of 5014 (permalink) Old 01-20-2017, 01:32 AM
New to Overclock.net
GroinShooter's Avatar
Join Date: Mar 2012
Location: Finland
Posts: 75
Rep: 4 (Unique: 4)
Nice to see this thread up, very interested in what people can achieve with the Kaby Lake chips thumb.gif

GroinShooter is offline  
post #3 of 5014 (permalink) Old 01-20-2017, 07:00 PM
New to Overclock.net
spddmn24's Avatar
Join Date: Oct 2015
Posts: 287
Rep: 7 (Unique: 7)
Guess I'll go first

Username: spddmn24
CPU Model: 7700k
Base Clock: 4200
Core Multiplier: 51
Core Frequency: 5100
Cache Frequency: 4600
Vcore in UEFI: 1.320
Vcore: 1.344
FCLK: Reminder: 1000
Cooling Solution: Delidded, H110i
Stability Test: 8 hours realbench

Batch Number: Malaysia L640F751 (Label is torn can't tell if F or E)
Ram Speed: (3866 18-19-19-39)
VCCIO: 1.208
VCCSA: 1.288
Ram Voltage: 1.376
Motherboard: Asus Strix Z270E Gaming
LLC Setting: 6
Misc Comments: Set both VCCSA and VCCIO to 1.25 after this to get memory stable in HCI memtest.

(13 items)
Z370 AORUS Gaming 7
Hard Drive
750 evo
Hard Drive
850 evo
Hard Drive
XPG SX8200
Power Supply
evga 750 g2
SilverStone Primera
Operating System
windows 10 pro
g502 lightspeed
▲ hide details ▲
spddmn24 is offline  
Sponsored Links
post #4 of 5014 (permalink) Old 01-21-2017, 08:21 AM - Thread Starter
M'aiq the Liar
BoredErica's Avatar
Join Date: Jul 2012
Location: Bay Area, California, USA
Posts: 7,402
Rep: 418 (Unique: 272)
Originally Posted by spddmn24 View Post

Guess I'll go first

Username: spddmn24
CPU Model: 7700k
Base Clock: 4200
Core Multiplier: 51
Core Frequency: 5100
Cache Frequency: 4600
Vcore in UEFI: 1.320
Vcore: 1.344
FCLK: Reminder: 1000
Cooling Solution: Delidded, H110i
Stability Test: 8 hours realbench

Batch Number: Malaysia L640F751 (Label is torn can't tell if F or E)
Ram Speed: (3866 18-19-19-39)
VCCIO: 1.208
VCCSA: 1.288
Ram Voltage: 1.376
Motherboard: Asus Strix Z270E Gaming
LLC Setting: 6
Misc Comments: Set both VCCSA and VCCIO to 1.25 after this to get memory stable in HCI memtest.

Somebody's got to be first. :o

Charted. As for my overclock, I'm still working on mine.

BoredErica is offline  
post #5 of 5014 (permalink) Old 01-21-2017, 03:32 PM
New to Overclock.net
ParanoidZoid's Avatar
Join Date: Dec 2016
Posts: 36
Rep: 1 (Unique: 1)
Would it be possible to add another column in the GDoc to address the ability to set an AVX Multiplier offset for these Kaby Lake CPUs?

ParanoidZoid is offline  
post #6 of 5014 (permalink) Old 01-21-2017, 10:12 PM - Thread Starter
M'aiq the Liar
BoredErica's Avatar
Join Date: Jul 2012
Location: Bay Area, California, USA
Posts: 7,402
Rep: 418 (Unique: 272)
Originally Posted by ParanoidZoid View Post

Would it be possible to add another column in the GDoc to address the ability to set an AVX Multiplier offset for these Kaby Lake CPUs?

It would be, yes. But I'm not sure if I want to do that. The requirements were set with no avx offset in mind. If somebody made an offset it would allow them to pass an overclock that I do not consider to be stable.

BoredErica is offline  
post #7 of 5014 (permalink) Old 01-21-2017, 10:27 PM
New to Overclock.net
PatRaceTin's Avatar
Join Date: Apr 2010
Location: Bangkok, Thailand
Posts: 25
Rep: 3 (Unique: 2)
Nice thread. thumb.gif

I saw some users (another webboard) said,
their i7-7700K running at 5.0core/4.5cache passed quick stable test
(Cinebench R15) at 1.1XX - 1.2XX ish V Core.

i7-8700K + Asus Maximus XI Gene + G.Skill RGB 4400C18 (8x2) Antec 1200W OC Version
EK Supreme HF + Swiftech 655 + XSPC ResTop + BlackIce GTX360 + Thermaltake 12CM x3

[My history systems] 80286 80386 80486 Cyrix K5 Pentium133 Pentium II Duron1G Athlon1G E2180 E3300 E5300 E7200 E8200 E8400 E8500 E8600 Q9550 QX6800 X3-720BE i7-920 i3-530 i5-750 [email protected] 955BE X4-B55 Q6600 i5-2500K i7-2600K X4-B60 X6-1055T FX-8120 i7-4790K i5-6600K G4400 i3-6100
PatRaceTin is offline  
post #8 of 5014 (permalink) Old 01-21-2017, 10:38 PM
New to Overclock.net
MooMoo's Avatar
Join Date: Dec 2009
Posts: 1,611
Rep: 65 (Unique: 62)
Could you add some explanation what does some things mean, example: AVX, cache frequency etc. It would be nice to learn what they actually do/mean, like back in the sandy bridge days I learned a lot by reading these explanations.
MooMoo is offline  
post #9 of 5014 (permalink) Old 01-21-2017, 10:56 PM - Thread Starter
M'aiq the Liar
BoredErica's Avatar
Join Date: Jul 2012
Location: Bay Area, California, USA
Posts: 7,402
Rep: 418 (Unique: 272)
Originally Posted by MooMoo View Post

Could you add some explanation what does some things mean, example: AVX, cache frequency etc. It would be nice to learn what they actually do/mean, like back in the sandy bridge days I learned a lot by reading these explanations.

To my knowledge:


Avx is an extra part of the x86 instruction set. When you run avx/avx2, things get hotter but they get done faster. The uses are typically image processing (so like in the x264 test) and complex algorithms (like Prime95).

Cache frequency helps determine the speed in which the CPU accesses its own cache. Practically speaking the performance impact of faster cache is rather small, and that's what I care about the most.

Some reading material if you are interested:



I also recall reading that faster cache can have better results for ram latency performance.

BoredErica is offline  
post #10 of 5014 (permalink) Old 01-21-2017, 11:30 PM
New to Overclock.net
MooMoo's Avatar
Join Date: Dec 2009
Posts: 1,611
Rep: 65 (Unique: 62)
Originally Posted by Darkwizzie View Post

To my knowledge:

Avx is an extra part of the x86 instruction set. When you run avx/avx2, things get hotter but they get done faster. The uses are typically image processing (so like in the x264 test) and complex algorithms (like Prime95).

Cache frequency helps determine the speed in which the CPU accesses its own cache. Practically speaking the performance impact of faster cache is rather small, and that's what I care about the most.

Some reading material if you are interested:

I also recall reading that faster cache can have better results for ram latency performance.

Nice, that cleared a lot for me thumb.gif I'll take a look on those reading material smile.gif
MooMoo is offline  

Quick Reply

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Please enter a password for your user account. Note that passwords are case-sensitive.
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page

Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off