Troubleshooting instability at idle while overclocking with adaptive voltage

[Queek]

A problem occasionally reported by overclockers of recent Intel processors (Haswell, Haswell-E, Devil's Canyon) is instability at idle.  The usual problem can be generalized into the statement:

"my processor was stable for X hours on a stress test at Y voltage, but my computer crashes when it is at idle" 

I have also had this issue with my Z97 4690k overclock, and wanted to summarize and explain my solution to this problem in a thread.  

 

I am not an electrical engineer, and I'm just an enthusiast who superficially understands what's going on (I hope).  I assume no responsibility for any damage resulting from recommendations made in this thread or overclocking in general (overclocking is a risk, you assume the responsibility for that risk when you start overclocking).  If I make any errors, please correct me, as I want this thread to be a resource for folks having this issue.  

 

Why there is instability at lower clock speeds:

When you validate an overclock with a stress test, you are testing a single scenario: is your processor stable at this clock speed with this voltage.  Stress tests are designed to simulate worst-case conditions and loads, so that you have a pretty good idea of whether your processor can take that specific load.  A stress test does not test the stability of your system at lower clock speeds.  This wasn't a problem until recent years.  In the past, overclocking meant that your processor would run at X GHz and Y volts at all times, so a stress test actually validated your system for all scenarios.  

 

Over the last few years, Intel and motherboard manufacturers have been moving towards optimizing power efficiency and usage.  One of the things that they've done is implement a feature that lowers the multiplier and voltage while the system is not being used (when the processor is "idle").  This a fantastic feature for power saving, and temperature mitigation, but can be problematic for overclockers.  It also means that a stress test doesn't fully validate your overclock anymore if you use this feature.  

 

The reason this can be problematic is that the motherboard manufacturer sets a standard voltage:multiplier curve that will work for most processors (author's note: I do not have a citation for this).  The key word in that statement is "most", so it will probably work for a given processor, but not necessarily.  To illustrate what I'm trying to say, I made a few charts.  In the figure below, the black dotted line is the amount of voltage supplied to your cpu at various ratios (as determined by your manufacturer), and the blue solid line is the voltage that your cpu needs to be stable at any given ratio (another term for multiplier).  The purpose of this figure is to illustrate that at some ratios, the voltage required by your cpu exceeds what is supplied by the motherboard manufacturer.  If your cpu doesn't have enough voltage, it will destabilize and your system will crash.  This is generally regarded as bad.  

 

How to fix it:

This voltage discrepancy can be fixed by using the "Offset" parameter.  This will shift (or offset) your motherboard's voltage curve up or down in voltage across the entire ratio range.  This is illustrated in the figure below, where the red dotted line has been increased across the entire range slightly so that it meets or exceeds the voltage required by your cpu.  It's important to note that this isn't the whole picture, and that these figures are only an illustration of what's generally happening.  Also, you won't have that blue line (unless you painstakingly determine it), so you'll have to guess at the offset you require.  

 

As an example, I found that my 4690k required 1.310V to be stable at 4.6 GHz, but at lower clock speeds, my system was destabilizing.  To fix this, I changed my "OC voltage" (the voltage that will be supplied at the maximum multiplier, in this case 46) from 1.310V to 1.280V, and added an offset voltage of +0.030V.  Overall, that should still result in 1.310V at my maximum multiplier (1.280V + 0.030V = 1.310V), and will stabilize my system at lower multiplier values.  

 

If you are encountering instability at idle/lower multiplier values, try decreasing your OC voltage by 0.010V and increasing your offset by that same amount, and try to use your system.  You may have to do this several times to get a fully stable system (I had to do this three times).  

 

Why can't I just overclock by offset voltage?

Very high offset values do wonky things.  I can't validate this statement personally, but from what I've read from various forums is that offset overclocking is a bad idea.  At a minimum, if you use offset overclocking, you will undoubtedly overvolt your cpu at many ratios.  

The figures above are only illustrations, and you can see some wonky things happen at higher offset values in ASUS' software outputs (see the spoiler).  

With no offset, the voltage curve stays identical to the motherboard's set curve until it reaches above the cpu's stock frequency, and then it starts a linear ramp up until it reaches your overclock multiplier.  

 

With higher offset values, it appears that the line deviates from linear, which suggests that offset is doing more than just shifting that curve up or down.  

 

Adaptive voltage overclocking is a fantastic feature, keeping temperatures down while still maintaining a high overclock at load.  But sometimes that overclock requires a bit of offset tweaking to get the system stable 24/7.  

I hope this helps folks having idle instability issues.  If this helps you, please like the thread so that it might show up more prominently in searches.  

[CoolaxGaming]

Naa, I would just disable the power saving features and run the processor at all times at a fixed clock

 

But really loving your articles, keep it up! +1

[Queek]

Naa, I would just disable the power saving features and run the processor at all times at a fixed clock

 

But really loving your articles, keep it up! +1

That is definitely another option.  

But for those who want to use the adaptive settings and overclock, they'll probably have to do this.