How do you overclock?

[Carfax83]

Just curious as to what type of overclocking people here do, if any.

Balls to the wall obviously means you try to get as much as you can get, given your cooling solution and setup.

Sweet spot means you find the best balance between the highest clock speed and the lowest voltage..

Personally, I used to be a balls to the wall kind of guy, but now I'm more of a sweet spot overclocker. I've learned that every chip has a sweet spot, a clock speed that is still significantly greater than default, but with a very small voltage increase.

On my current CPU which is a 4930K, the sweet spot is 4.3ghz, which I can get at 1.184v. For me to get 4.4ghz, I need 1.25v. And for 4.5ghz, 1.3v+ is necessary.. This is on a Noctua NH-D14.

So from 4.4ghz and up, the voltage requirements increase significantly, which to me makes targeting those clock speeds imprudent if I want to stay well within Intel's specified TDP boundary for 24/7 use.

A 200mhz increase will not be felt in anything other than sensitive benchmarks... And CPU degradation is a reality.

 

[Deders]

3800 at the lowest possible voltages is the sweet spot for me, any much more and it would be using a lot more power and probably wear components out faster.

 

[Burpo]

4.3Ghz still requires a bit of voltage with a Westmere Xeon, but I really don't care about degradation. Controlling temps is what's important, and as long as it doesn't throttle it should last, even @ higher volts. It's been a yr already & I hope to get 2 or 3 more out of it. Will let you know in 2017 or when it goes poof..

 

[Carfax83]

Yeah, chip degradation isn't a joke. My 3930K was slightly degraded as a consequence of running it at 4.5ghz 1.356v for over a year. We really have to be careful about targeting high clock speeds, especially if you are using air cooling.

Water cooling is much more forgiving of course. But even so, using high voltages will cause degradation no matter what.

 

[ZGR]

After 4.5 GHz, my chip just needs way too much voltage to remain stable. I look forward to see how Broadwell-k will fare with overclocking. Probably won't be any better.....

On the other hand, I haven't seen too many SB chips that have been running high clockspeeds/voltages for years. Maybe I should just oc to 4.8 GHz at ~ 1.4v and see how long my chip survives? Stability testing would be strictly for games and light benchmarks as temps will be toasty! jk

 

[BonzaiDuck]

Well, "Sweet Spot" has it, and I also bet my two-cents on that choice.

Balls to the wall for me would be a load voltage of 1.4V with an LLC giving a 20mV droop. In that scenario, the unloaded turbo voltage would be 1.42 before settling down to EIST idle and ~1.0V. I'd guess the load temperatures @ 80F ambient might be 85C average-of-maximums. I'd be inclined to choose the strongest stress test that took the least time.

You want the inflection point of voltage Y and speed X -- where the next highest target costs you noticeably more in mV than coming from below the current speed.

 

[Sable]

I used to have it at 4.8Ghz with 1.35v and it was pretty much rock solid but after the odd crash when returning from sleep, like once every couple of months, I decided it was too much. Dropped it back to 4.5Ghz, did a completely new install of 7 and it's perfect.

With serious tweaking I could probably get 4.8Ghz stable and 5Ghz balls to the wall but I don't really have the skills for that. I also use offset which I think limits what you can do with balls to the wall.

 

[hawtdawg]

balls out! Finding ways to squeeze every last bit of performance out of something, while staying within the "safe" zone is how I roll. I'm the type of person who de-lids an Ivy Bridge

 

[Spawne32]

balls out! Finding ways to squeeze every last bit of performance out of something, while staying within the "safe" zone is how I roll. I'm the type of person who de-lids an Ivy Bridge

delidding chips isnt the "safe zone" lmao

 

[Borealis7]

i do it naked with a tight belt around my neck. oh yeah, keep that mental image.

max stable frequency without voltage bump.

 

[BSim500]

Normally just a stock voltage OC "sweet spot". Got my i5-3570 running at 4.0-4.2GHz at barely 1.01v. Never exceeds 55c under Prime (for games its often 40-52c) even with 1x 700rpm silent fan on a 212 EVO. Total system power consumption (measured at the wall) is 37w idle / 88w Prime (both with dGPU but no GPU load, removing it and using iGPU results in 27w idle / 79w Prime). Best chip I ever had...

 

[VirtualLarry]

My Q9300 @ 3.0 chips are limited by RAM and mobo chipset. I'm sure that the CPU can go higher, as I'm on nearly stock voltage. Temps don't go over 70C on the CPU fully loaded.

I guess I would say "sweet spot". However, I sold a friend a Q6600 rig that was OCed to 3.6, which kind of was "balls to the wall", although it OCed to that speed at a low overall voltage, but it took a big step to go from 3.5 to 3.6.

 

[KaRLiToS]

Ballz to the wallz fo Life

 

[poohbear]

Overclocking is so simple these days with the overclocking focused mobos.... its a far cry from the athlon & pentium 4 days. U pop the cpu in, go into bios, and choose performance mode. BOOM. Im @ 4.6ghz on a 4790k.

 

[Charlie98]

Because my OC machine is also my business machine, a reasonable OC with an undervolt... currently 4.2GHz at -.08v on air.

If I had the resources (i.e. a disposable machine) I would run it all out, just for the technical aspect of it.

 

[skipsneeky2]

Got a non k 3770, a few tweaks and all my cores under load hit 3.7GHz. Enough for me.

 

[biostud]

A good overclock that doesn't require too much tinkering to get it stable, and doesn't require too much voltage. Running my rig with 1.275V.

 

[fleshconsumed]

I don't do it anymore. I overclocked PIII's, I overclocked P4 celeron, I overclocked P4-3.0C to 3.4GHz, I overclocked the infamous fry's e4300+ECS combo from 1.8 to 2.4GHz using pinmod - that one was real fun project, I even sort of overclocked the q9450, but I removed overclock because I noticed more electrical whine coming from the motherboard when overclocked. That was the last CPU I tried to overclock. My current 4770K is fast enough that I just do not want to trade stability for some minimal increase in speed that I cannot even notice. I might get 4790K if Microcenter puts them on sale this year, but I won't overclock that one either. Used to think locked Skylake is a huge deal, but I don't think I care anymore as I do not overclock. I used to lap my CPUs but once again, I don't do it anymore because it's harder to sell lapped chips. The benefit is just not noticeable and I do not want to waste my time finding the "sweet spot" and then troubleshooting my computer should I ever have stability problems. It just isn't worth it.

 

[Cerb]

I'm finding the platform improvements mattering as much as clock-level performance, if not more. I've been OCing since the first PC I made, but I must say I'm quite content with my current stock rig, with a bit of negative offset on the CPU voltage. I've got a little fussing about to do to get the CPU and GPU cool under stress testing without the fans running so fast, but I haven't found a real application that can make things that bad, so it's not major (just one of those, "I know I can do this better, because I have done so in the past," things that bugs me).

 

[PhIlLy ChEeSe]

I don't do it anymore. I overclocked PIII's, I overclocked P4 celeron, I overclocked P4-3.0C to 3.4GHz, I overclocked the infamous fry's e4300+ECS combo from 1.8 to 2.4GHz using pinmod - that one was real fun project, I even sort of overclocked the q9450, but I removed overclock because I noticed more electrical whine coming from the motherboard when overclocked. That was the last CPU I tried to overclock. My current 4770K is fast enough that I just do not want to trade stability for some minimal increase in speed that I cannot even notice. I might get 4790K if Microcenter puts them on sale this year, but I won't overclock that one either. Used to think locked Skylake is a huge deal, but I don't think I care anymore as I do not overclock. I used to lap my CPUs but once again, I don't do it anymore because it's harder to sell lapped chips. The benefit is just not noticeable and I do not want to waste my time finding the "sweet spot" and then troubleshooting my computer should I ever have stability problems. It just isn't worth it.

You sound good n all that happy stuff, so why not buy a none K chip?

 

[fleshconsumed]

You sound good n all that happy stuff, so why not buy a none K chip?

Practical reasons. One, I have microcenter near me, so I can get K chip cheaper than non-K, and two, K chips are easier to sell if I ever decide to upgrade. Those are the only two reasons I got K chip last time around - cheaper and easier to sell. That might change if Intel decides to push locked chips en-masse, I'll just have to reevaluate my options when that happens.

 

[BonzaiDuck]

You sound good n all that happy stuff, so why not buy a none K chip?

Philly has a point. VirtualLarry's remarks also insightful. Biostud has me intrigued with his 5820K and H100 cooler with 1.275V(?!!?) I'm two generations behind with my sig-rig, with several years left of use because it overclocks very, very well. Someone mentioned a 4770K Haswell with which they were happy. I'm also happy in comparing my Cinebench scores (with a GTX 570 no less -- at the time of comparison!) -- to a 4770K OC'd to 4.4. I was barely a few points behind with 4.6 and DDR3-1866 with command-rate 2.

I can't speak for socket-1366 folks, 920's and Gulftown "X's". I certainly won't speak for folks I've known stuck in perpetual-noob gridlock with their overclock attempts -- who told me buying a Sandy Bridge with the built-in iGPU (their objection to it) was a bad idea.

The Sandy cores were immensely overclockable, and everyone here knows that all the 2500K, 2600K and 2700K folks are still happier than pigs in s***.

I just saw bad things coming with the new fabrication/TIM process for IB and Haswell. I also thought I saw bad things because of the die-shrink from 32 to 22: more electrical leakage at higher clocks, and "walls" of dead-ends beyond certain levels without something akin to LN2.

NOW I see the i7-5820K which seems to have good prospects for 4.4, maybe 4.5 -- maybe even 4.6.

The motherboards with their noob-friendly auto-overclocking features have really become souped-up with BIOS features that favor overclocking, but for some of these chips, the board features are more limited by the processor for how far they will take you.

I also said that it seems a bit fruitless to OC a 4790K Canyon-Refresh, since some minor tweaks would get you to 4.4 "by all cores" right away.

And finally, if the prospects are limited, look to see folks going "Balls to the Wall" as they try harder with less "LebensRaum."

And one more -- with the Haswell "E" processors, I think we're at a crossroads between air and water. BioStud is using an H100, and I can beat him by a few degrees under controlled thermal power with my D14. I just think that E overclocking will do better with cooling more effective than an H110i.

I also think some of us have gotten skittish with stress-test choices, but I see us falling back to something like the OCCT:CPU tab. I can think of more subtopics to discuss about this, but my fingers are getting tired . . .

 

[exar333]

'Balls to the wall' here, but at the point of diminishing returns. It doesn't make sense for me to get an extra 100mhz if I need to bump-up voltage by 10-15%. BUT, I don't mind feeding some pretty high voltages to the CPU, provided it has been validated to be 100% stable. I am usually pretty happy with about 95% of the OC ceiling of my CPU for 24/7 use.

 

[biostud]

Philly has a point. VirtualLarry's remarks also insightful. Biostud has me intrigued with his 5820K and H100 cooler with 1.275V(?!!?) I'm two generations behind with my sig-rig, with several years left of use because it overclocks very, very well. Someone mentioned a 4770K Haswell with which they were happy. I'm also happy in comparing my Cinebench scores (with a GTX 570 no less -- at the time of comparison!) -- to a 4770K OC'd to 4.4. I was barely a few points behind with 4.6 and DDR3-1866 with command-rate 2.

I can't speak for socket-1366 folks, 920's and Gulftown "X's". I certainly won't speak for folks I've known stuck in perpetual-noob gridlock with their overclock attempts -- who told me buying a Sandy Bridge with the built-in iGPU (their objection to it) was a bad idea.

The Sandy cores were immensely overclockable, and everyone here knows that all the 2500K, 2600K and 2700K folks are still happier than pigs in s***.

I just saw bad things coming with the new fabrication/TIM process for IB and Haswell. I also thought I saw bad things because of the die-shrink from 32 to 22: more electrical leakage at higher clocks, and "walls" of dead-ends beyond certain levels without something akin to LN2.

NOW I see the i7-5820K which seems to have good prospects for 4.4, maybe 4.5 -- maybe even 4.6.

The motherboards with their noob-friendly auto-overclocking features have really become souped-up with BIOS features that favor overclocking, but for some of these chips, the board features are more limited by the processor for how far they will take you.

I also said that it seems a bit fruitless to OC a 4790K Canyon-Refresh, since some minor tweaks would get you to 4.4 "by all cores" right away.

And finally, if the prospects are limited, look to see folks going "Balls to the Wall" as they try harder with less "LebensRaum."

And one more -- with the Haswell "E" processors, I think we're at a crossroads between air and water. BioStud is using an H100, and I can beat him by a few degrees under controlled thermal power with my D14. I just think that E overclocking will do better with cooling more effective than an H110i.

I also think some of us have gotten skittish with stress-test choices, but I see us falling back to something like the OCCT:CPU tab. I can think of more subtopics to discuss about this, but my fingers are getting tired . . .

The choice of the H100i was two reasons: I was able to get one used for a review for half price and I prefer not to have a big air-cooler take up space inside my case.

We have a long discussion of overclocking of haswell-E in this thread: http://forums.anandtech.com/showthread.php?t=2394287

For normal stress (ASUS realbench) testing I can keep the CPU and GPU under 80C with combined load, but AVX torture tests will either crash or bring the CPU to 95C+.

I have yet to try OCCT, but I might give it a go.

 

[BonzaiDuck]

The choice of the H100i was two reasons: I was able to get one used for a review for half price and I prefer not to have a big air-cooler take up space inside my case.

We have a long discussion of overclocking of haswell-E in this thread: http://forums.anandtech.com/showthread.php?t=2394287

For normal stress (ASUS realbench) testing I can keep the CPU and GPU under 80C with combined load, but AVX torture tests will either crash or bring the CPU to 95C+.

I have yet to try OCCT, but I might give it a go.

Funny. I thought I'd been looking for such a thread in past weeks and months: it escaped me. That's the sort of information sharing I'd been hoping for, as I do EARLY planning for an E-build next year.

Lepton87 had a thread here and at Cases/Cooling on his 5820K build, and we discussed this.

The AVX2 features shoot the temps through the roof with standard and commonly used stress-tests.

For some reason, I stumbled back into using OCCT. I'd had it installed three years ago, but the free version has a 1-year use limit, and you have to download the new improved OCCT to get it back.

The sparse info offered by the OCCT author (who could be French or Russian but who knows? and what does it matter?) -- states that the "OCCT:CPU" tab will catch errors faster in OC settings than will the Linpack tab (which is essentially the same as IBT or LinX). He notes that it distinguishes thermal stress from "other stress" which would turn up those errors.

The OCCT:CPU test seems to give temperatures peaking at maybe 4+C below the LinPack option.