Rallycobra's Quick and Efficient Aircooled Haswell Overclocking Guide

[rallycobra]

Haswell doesn't make a lot of heat, but it has difficulty getting it away from it's components because the cap on the chip is such a good insulator. Because of this, the most voltage you can hope to run air cooled is 1.30V. Otherwise the internal temp is going to be 90C +

1) Set bios values to optimized default

2) Set cpu adaptive voltage to 1.30V

3) Set all cpu cores to the same multiplier and start working up 42x, 43x, 44x, 45x etc.

4) Stress test with prime95 and Aida64 simultaneously.

5) Once you crash or fail, drop the multiplier down and see what you can run overnight - This is your max stable overclock multiplier

6) Drop the voltage down .01V at a time while running your max stable overclock multiplier until you fail AIDA/Prime - Raise it back up to what was stable so You know now what your minimum adaptive voltage necessary is.

7) Use your minimum stable adaptive voltage and subtract 2 from your max stable multiplier so you have a nice stable system. Subtract 3 if you want extra extra safety margin.

All done with a minimum muss and fuss! We didn't throw away all of those cool power saving features of Haswell either with locked voltage, and the cpu and other components should last much longer with this strategy. Also, you don't spend a week testing every weird option in the bios for an insignificant performance boost.

I'm running a i5-4670k on a Asrock Z87 Extreme4 MB with CoolerMaster 212 Evo heatsink.
My max multiplier is 44x at 1.298V
I run the machine at 41x

 

[blackened23]

Is this the best idea? We know that adaptive voltage with the Haswell creates artificially high voltage loads when AVX instructions are in use (eg prime95) - from what I understand, AVX instructions will always add +.1V to Haswell with any type of program that uses AVX instructions. As well, I really feel like 1.3V is a bit on the high side for a starting point - for many people that would almost guarantee a BSOD for the 4770k/3770k. I mean, even with my IVB i'm leery about using over 1.25V. Technically, you CAN, but 1.3V is really pushing it with the Haswell or IVB, although I do see what you're saying about lowering the voltage down in small increments.

I think the wiser method is to use manual voltage with C3 and C6 disabled - this should prevent the voltage from spiking due to AVX instructions (this is something new with Haswell). Also, unless you have a pretty high end closed water cooler such as the H100i I do think that 1.3V will always produce excessively high temperatures. Heck, on my 3770k I easily hit 90C with 1.3V.....i'm not so sure that's the best starting point. Perhaps 1.25V?

Just my thoughts. Granted, you are using the 4670k without multi-threading. The i5's have lower temperatures across the board since HT is disabled, so you have a little more "room" to play around with in terms of voltage, I suppose, but I definitely wouldn't use 1.3V as my preferred starting point with HT turned on.

 

[rallycobra]

When I bench the machine with default bios settings (default clocks and voltage) I was seeing 1.1V on prime95. Bumping the adaptive to 1.3V is a .2 bump.

With the adaptive voltage, when you back off two or three multipliers, you drop down in voltage also, so the stress test under high voltage is a worst case scenario. And you should be below 1.3V

The whole advantage of Haswell is it's ability to downclock and undervolt on it's own, and it's adaptive algorithm to increase voltage when its needed. FWIW, I want the voltage spikes under AVX. If I was going to run manual voltages I would have puchased a Sandy or Ivy chip and cranked it up.

On all four cores, I can run 42x with stock voltages. 43x I have to go adaptive at 1.202V. 44x is 1.296V adaptive.
I have settled on cpu speeds 1,2,3,4 cores of : 41,41,40,40 and Adaptive Voltage 1.21
Running AIDA/Prime CPU-Z reports core voltages about 1.246V Core Temps are averaging under 60C Max of 73C

 

[blackened23]

First off, i'm not trying to take away from your guide - it is helpful, thanks for taking the time to post it. I do think the i5 and i7 chips differ in terms of tolerances, though, so let me point out:

Ivy isn't much different than Haswell in terms of temperatures. Also, another point is that the 4770k will be far more punishing in terms of voltage than your i5. HT will increase temperatures across the board, so with that being the case I believe 1.25V is a good starting point.

I've seen 4770k's stress tested at 1.25V and the temperatures can get really high even with an H100. Heck, even my 3770k gets hot at 1.25V with HT enabled (I also use offset, and not manual voltage). You do not, under any circumstance, want +.1V with an HT enabled part because your temperatures will quickly become untenable. Your experience may be different due to the lack of hyperthreading, but I can definitely state that 1.35V on a 4770K with HT enabled with not be tenable. Temperatures will be WAY too high - this is why manual voltage is preferable while trying to stress test IMO. Again -- you have a lot more "leeway" with temps/voltages with a non-HT i5 chip.

So I guess the main point i'm getting at is that the i5 and i7's differ in terms of voltage tolerances, so you may have to treat them a bit differently while stress testing. What works for an i5 may not necessarily be ideal for an HT-enabled i7. I do like adaptive voltage as well, but specifically with the i7 I think that doing manual during the stress testing phase to find a "sweet spot" is a pretty good strategy, and then you can use adaptive once you figure out what works for your chip.

Again, i'm not trying to take away from your guide! Thanks for the information.

 

[Dizon]

My CPU would fry going 1.3V. I get 96C load temps doing IBT max setting at 1.21V 96C 4.3GHz. I have a 4770k.

 

[blackened23]

My CPU would fry going 1.3V. I get 96C load temps doing IBT max setting at 1.21V 96C 4.3GHz. I have a 4770k.

Just out of curiosity, are you using auto/adaptive voltage or manual? Also are C3/6 enabled?

 

[Dizon]

Just out of curiosity, are you using auto/adaptive voltage or manual? Also are C3/6 enabled?

I turned off adaptive and put it on manual when doing tests. After reaching stability with manual, I switch back to adaptive for the C states.
Ambient has also been ~32-37C this past week. No AC

 

[rallycobra]

I have read that the i7's run a lot hotter than the i5 counterpart. I don't need HT, and the lower price were the three reasons I went with an i5. If someone out there is looking at this guide and using it with an i7, maybe you should set the adaptive voltage to 1.25V. You don't want core temps to break 90C.

I would rather stress test in adaptive mode, and back everything down once I have stability. Its one less variable than stressing in locked voltage and hoping the C states and lower voltages work. It's possible that you could be unstable somewhere under the adaptive curve if you are stressing on a locked voltage, but its unlikely.

 

[Dizon]

I have read that the i7's run a lot hotter than the i5 counterpart. I don't need HT, and the lower price were the three reasons I went with an i5. If someone out there is looking at this guide and using it with an i7, maybe you should set the adaptive voltage to 1.25V. You don't want core temps to break 90C.

I would rather stress test in adaptive mode, and back everything down once I have stability. Its one less variable than stressing in locked voltage and hoping the C states and lower voltages work. It's possible that you could be unstable somewhere under the adaptive curve if you are stressing on a locked voltage, but its unlikely.

Why would you have to worry about anything "under" the adaptive curve. Stress testing doesn't test the lower frequencies/voltages, it'll blast all the way up anyway.

The only good thing about testing on adaptive voltage settings is to account for the extra voltage influx OVER your manual settings and the heat it produces when running stress testing programs, not anything below your settings.