Haswell Overclocking Guide [With Statistics]
Haswell Overclocking Thread
Why use this guide?
I get data from tests instead of taking claims from others on face value. I spend a long time testing each setting and charting the data to make sure my data is right. I have a long chart of OC results from OCN members with all of their settings noted. Average OC, etc, are derived from those values. I still update this guide and this guide is built and updated upon thousands upon thousands of replies.
What is overclocking?
Simply put, you up the performance of the CPU, typically by increasing its core clock speed. With 4670k/4770k we do so by upping the multiplier. So a x44 multiplier with 100 base clock is 44 x 100 or 4400 mhz or 4.4 ghz. Overclocking may potentially damage your cpu beyond repair. I am not responsible if your CPU dies. Just use common sense, watch the temperatures before you dash off to stress test for hours on end, read the entire guide before doing random changes, etc. There has been no confirmed Haswell CPU deaths I have seen due to sticking to my guide with my parameters for what is safe.
My recommendation is to not go into Haswell with delidding as your first goal. Only delid if you are thermally limited and you feel being able to crank voltage up a bit and thus generating more heat is vital to getting the next multiplier. You may find out going to the next multiplier is unstable no matter what voltage. First thing to do is to drop to x264 stress test and see if the temps are OK there. Delidding only makes sense when your overclock is being severely impeded by temperature AND you have voltage headroom. Often, this thermal issue can be bypassed by using a less hot stress test. I got approximately the same temperatures from 1.2v @ Linpack vs 1.5v @ Chess. Also, delidding won't help if your voltage headroom is very small. Delidding is to decrease temperatures to allow for higher voltages. If you can't push higher voltages despite being cooler out of fear of degradation, then delidding is mostly pointless as well.
What about Adaptive? I'll talk about this later in the guide but right now I recommend setting voltage mode to Override. Gigabyte boards don't even have adaptive options.
- Set Uncore (AKA Ring Bus) to core ratio to manual. Set it to stock multiplier manually. With ring bus running on stock and locked at stock for now, no need to fiddle with ring bus voltage. That goes to auto. Some motherboards MAY increase Vring to unsafe levels if you didn't manually set ring bus to stock because the motherboard will try to auto-overclock the ring bus if left on auto. Avoid future headaches by following step 1. If you have a Gigabyte motherboard, set uncore to x33. Reason stated in the next bolded section.
- Set any XMP profile OFF for ram. If your ram is above 1600, set it to 1600, no higher while we're testing overclocks. Heck, if your ram is XMP'ed for 1600, lower it to non XMP. Doesn't hurt.
- Start ramping up the core multiplier and voltage until you think you've found your sweet spot. You might need to increase input voltage from the motherboard-set setting once you hit higher voltages. More on that later. Do stress test and if you pass, go to step 4. NOTE: Yes, you up the core multiplier by 1 each time. You can probably get away with starting at 4ghz though. Because the core clock is 100, 100 x 40 = 4000mhz or 4ghz.
- Now we fiddle with ram, higher ring bus/ring bus voltage in effort to get marginally better performance. Stress test. Do not raise ring bus or ram if it means lowering core clock for stability.
- If you are hardcore or have a lot of time on your hands, you can try adding clockstrap to the mix but I'd stay away from it unless you're super duper picky. Nobody has reported very good results with this but you're welcome to experiment.
- Set Cstates to ON to C7. If you really care about lowering idle voltage, hit ctrl+f and search for "Power Saving Info".
- Have a glass of iced tea.
If x44 is stable, don't jump to x46, and then say it's not stable. And for the love of god, don't waste your time telling us x47 or x48 is unstable.
I recommend changing input voltage in 0.05v increments. Any less you need a zen-like patience to test everything. I recommend max 0.1v increment if you are lazy. Do not do the same with Vcore or other types of voltages obviously. The reason why input voltage becomes a larger factor at higher Vcore is because input voltage is typically automatically managed by the motherboard's own software. But when the Vcore goes high up, the motherboard almost never compensates the input voltage well enough to ensure stability. Depending on how good your motherboard is at making sure the CPU has enough input voltage for the Vcore, you may have to tweak the input voltage before you even hit 1.3v Vcore.
Please note I can only test LLC for MSI G45 Gaming Board. For this mobo, the setting is under the "DigitAll Power" section. Also note that LLC is for Vrin, NOT Vcore, Load tested with Prime95 28.3 with HWinfo. The Vrin as set by BIOS is 2.15v.
This means that upping that LLC can potentially help you when your Vrin is the offender.
What does this mean? Your Vcore could be above your VID. If you set 1.3v in the BIOS that's 1.3v VID. If you are also under adaptive voltage and you're running Prime95, your Vcore could be a whopping 1.5v, way above your set 1.3v.
And here are the most recent tests for uncore that I have done:
The 4.2 vs 3.4 is the uncore 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.
- Which test are you running on what setting? Double check with the chart down below. For example, Aida64 full suite is NOT the same as Aida64 FPU test only. IBT is hotter than Prime 27.9 but not hotter than Prime 28.3 on small. Prime 27.9 is cooler than Prime 28.3.
- Ensure you are not getting Vcore spikes due to adaptive voltage while stressing on synthetics.
- If your temp from one core is 10C+ hotter or cooler than some other core in a load that loads all CPU to 100%, it might be a thermal paste application error. Pea method, folks. 10C degree difference and lower is NORMAL on stress tests.
- Does your cooler just suck? Are you trying to apply Sandy Bridge voltages onto Haswell?
- It is hypothesized that some temperature variance between CPUs is normal because the way the gunk inside the CPU under the IHS is applied is spotty, better in some and worse than others. While this is NOT backed up by actual scientific testing, it makes sense and accounts for the temperature variance.
- Hyperthreading makes the CPU hotter. This means 4770k might run hotter than 4670k.
This game is known for being easily unstable under an overclock. Some people consider this a better test of stability than even Linpack! For Battlefield 4, be careful because crashes may be due to immature drivers for GPU or software issues!
...Is currently still unstable as a game. So in terms of stability testing, it's of limited use. But if you BSod, well, that's still probably CPU overclock, not a crashy game. If the game just crashes, blame the game.
'Prime95 is not Certified for Haswell'
Right, another unsubstantiated rumor floating around. No software is ever certified with a CPU. There isn't a committee that goes around with official badges handing out official certification after a 10 step process, ok? There have been zero proven reports of CPU death due to Prime95. You can say, better safe than sorry, in which I reply with, don't even overclock then! Nothing is absolute, but the data has come in and it shows the scare being unsubstantiated. Look at the chart, there are piles of people who stress with Prime95. Half of the misconceptions stirred from a claim that Prime raised voltages way above what is normal. YES, if you run it on adaptive. But the same thing happens for every other synthetic, Prime isn't special in any way. It just so happens that Prime is often the first stress test people run, and if they don't know about adaptive stressing, they will freak out at the voltages before they know the full picture.
Input Voltage and Uncore increasing temps: The Verdict
No significant or easily measurable increase in temps were present. The temperatures were within margin of error. Tested with Prime95 28.3 with HWinfo.
Testing Uncore increase in temps:
Testing Uncore + Input Voltage:
Haswell is still a new launch. That means programs associated with it will constantly get updates. Older versions of Prime are easier to pass than newer ones. Many programs have glitches. There are idiosyncrasies. The only way to iron out what's what is with your cooperation, and lots of communication.
If you want to find out when your computer Bsoded for whatever reason, you can look at Windows/Minidump. It is time stamped.
You can find out how stable your setting is by running a stress test and seeing how long it takes to crash. I tested what VCCIN is optimal by running x264 runs, 1.85, 1.95, 2.05v, 2.15v VCCIN and I recorded how long it took to crash after five tests. This is how I figured out I needed 2.15v VCCIN.
- Did you read the entire guide? No, seriously. Did you?
- Did you ask a question in this thread? Listing more info than is needed is typically better than listing too little. What is your Core and uncore multiplier, their respective voltages, input voltage, ram speed and timings, CPU, base block, and used stress test? What is your Bsod code if you have one? It crashed? How did it crash? What temperatures were you getting?
- Did you remember that higher Vcore means a higher Vrin is required to maintain stability?
- Did you set uncore to stock? Is the voltage for that uncore ridiculously low?
- Did you overclock your ram or GPU?
- Did you try to run Prime95 for 1,000,000 hours?
The Haswell Terminology Listing
Possible Haswell CPU models:
4670k (without Hyperthreading), 4770k (with Hyperthreading).
Base Clock: Typically 100. Multiplied with the core multiplier to get core speed or with the uncore multiplier to get uncore speed, etc. Screwing with this can screw with multiple components.
Core Multiplier: The primary speed setting you change to change the CPU speed. x34 or x35 is typically the default. 100 base clock x 34 = 3400 mhz or 3.4 ghz.
Core Voltage: The voltage for core. Higher multipliers require higher core. aka Vcore.
Uncore Multiplier: The secondary speed setting. Always of secondary important to the core multiplier. AKA Cache Ratio or Ring Bus.
Uncore Voltage: The voltage for uncore. AKA Vring or Cache Voltage.
Input Voltage: The all-around voltage sucked in by the CPU for all the components. The higher the Vcore, the higher this should be. AKA Vrin, VCCIN, Eventual Input Voltage.
Last Ditch Effort to Find Stability:
This is typically best attempted when you are trying to hit that last multiplier and you're unsure what the instability is from. Run x264 multiple times (I did 5). Each time the computer Bsods, that count as 1 run. Time how long it takes to Bsod. Then change one setting. Note any change in average time until Bsod. This is how you figure out whether a setting is increasing, decreasing, or not affecting stability without relying on personal feelings but hard data. Specifically, this will tell you if you need higher Vrin or Vcore or both.
The reason why this test is effective is because it pinpoints the effect of a setting. Does increasing Vcore change stability? Yes or no? It can tell you the answer without you having the pass all of the runs. Just having a noticeable increase in average time until bsod demonstrates an increase in stability. Or maybe it's input voltage that's problematic. You can test that too. Just change the input voltage and see if the stability changes. For this test I would change input voltage in 0.1v increments.
Can't see the chart? Log out and back into Google Docs!
Want to be in the chart above? Click this spoiler!
In order to be charted you need to fill out this form:
|AVERAGE OC||45.39||MEDIAN OC||45.00|
|AVERAGE VID||1.29||MEDIAN VID||1.28|
Amount of overclock submissions: 181
-Chart now shows last update time at upper left hand corner.
-Added new VCCIN column in the chart.
-Vdimm values now moved into the ram speed column, now renamed 'ram settings'.
-Charting form now includes a pointer to note if delidded under 'cooling solution'.
-Chart columns have lower width to help those with smaller monitors read it more effectively.
-Now specifically noted what the x264 OCN link leads to.
-Removed an old portion of the guide with wrong info.
-Removed old, irrelevant info on x264 that now only confuses people.
-Now includes Prime95 download links.
-Charting form now includes me asking whether CPU is Malay or Costa chip.
-Input voltage section has some revisions.
-Edited delidding info.
-Removed the line repeated where I said, set xmp off even if it's 1600.
-Added info on Prime95 stopping while stressing.
-Added 'ram voltage' column to charting form.
-x264 download now includeds AVsynth and batch file in one download
-x264 loop exe updated.
-Override = Manual voltage noted.
-Chart average voltage values now rounds to 3rd decimal place.
-Added info about Gigabyte's ridiculous uncore ramping to x40 when set to stock.
-Added more in-depth info on power states.
-Added a quick note on Asus bios version.
-Ring bus benchmark spoiler title changed to "Ring Bus Doesn't Matter [Evidence]
-Greatly revised the old chart of stress test info and temps.
-Added info on Cstates vs Adaptive mode. Any suggestion for using adaptive has been deleted.
-Vrin/Uncore and its effects on temperature have been listed.
-LLC investigation done and a new section included.
-Removed extra unneeded data from Stress Testing section.
-Revised total number of OC submissions.
-Better numbered list implementation.
-CPU voltage parameters cleaned up.
-Input Voltage info listed twice, now fixed.
-Added 'Why use this guide" part.
-Added some new titled at the stress testing section for easier reading.
-Last updated changed to guide last updated.
-Included picture showing where Vcore reading is on Hwinfo.
-Enlarged font at start saying not to stress on adaptive.
-Started a change log.
-Charting form now includes question for LLC.
-Added point to check another settings for Asus mobos if looking for input voltage reading from HWinfo, courtesy of Veerk.
-Moved the input voltage info into its own seperate section and moved it up physically. That way more people will read it as it is more important than some other settings. Also clarified Input voltage info.
-Added the 'Why are my temps so high?' troubleshooting tips under 'Stressing' section. Added tidbits on why a core is cooler than some other cores by up to 10C.
-Added this new 'changes from last edit' section.
-Revised total number of OC submissions, from 110 to 115.
-Included extra note to use HWinfo is readings seem off.
is the Ring Voltage you refer to what ASUS calls CPU Input Voltage? because that has turned out to be important too.
Great start to a thread by the way
My OC is 4670K with a multiplier of 42, 44 is on its way though, just need to tweak to perfection
p.s. information on the different naming conventions between motherboard manufacturers should be on here imho
Hmm Asus mobos call it 'cache ratio' for ring bus. I don't see where they call it CPU Input Voltage. Video I am referring to is here:
I'm going to try to grab mobo terminology from other vendors. The more data I can get from others the better, right?
I'll add something on LLC but what I heard is it's not really important anymore due to vdroop almost being gone.
CPU Input Voltage
link to my bios screenshot
this is the voltage people are putting up as far as 2V maximum in range is 2.7v, AUTO sets it at around 1.8 for mine
to flesh out my entry in your sheet
its a 4670K with multi at 42 vcore at 1.22 LLC at level 5 everything else auto Cooler BEQuiet Dark Rock Advanced C2
Cache Voltage on Asus is Vring on Gigabyte.
Not sure about the others. And LLC only affects the voltage going into the CPU (the CPU Input Votlage/VRIN) not the actual Vcore, so it is less important than it used to be.
LLC seems to be a valid tweak for increasing stability, however so does setting the CPU Input Voltage so if they are more or less doing the same job then that answers that question
Yeah, LLC still has an impact, it is jut more indirect than it used to be. Before LLC directly affected the Vcore, now it affects the VRIN, but if VRIN drops too low (like under heavy load) then the on-board voltage regulators don't get enough voltage and so can't provide a stable Vcore, which can cause crashes. Same effect but less immediate, in a manner of speaking.
Folks with Haswell....DO NOT GET DISCOURAGED! Most are stuck at 4.2-4.3ghz because they never touched their UNCORE. Once you lower your uncore down to 34x or 35x, you can easily increase your multiplier headroom. Sometimes by even 4.6ghz....thats an increase of 400mhz over an initial 4.2ghz....which is a HUGE bump in performance..... compared to the 1% loss in performance by lowering uncore down to 34x or 35x. CORE IS STILL KING. Oh yeah and please....if you have a 2400mhz memory kit...DO NOT run 2400mhz initially. Lower it to 1600mhz and try to see the max potential of the CPU, before raising the memory back to 2400mhz. Most CPUs cannot hold a high overclock (4.7ghz) with a high memory speed (2400mhz)
I have no doubt in my mind that most Haswell samples can run approx 4.5ghz, by playing with these new settings we weren't familiar with on Sandy/Ivy. Good luck