Not enough VCore when idle?

I have a really nagging weird problem running Linux on my overclocked machine. I can beat the living daylights out of it with Prime95, LINPACK, etc., but it seems to crash at the most random times under very little load.

I'm at 4.8GHz and using a VCORE offset of 0.30V. Everything seems stable, I can pound this machine for all it's worth without a single issue, but like I said, it'll crash at idle in Linux due to what I think might be too little VCORE.

You can see many of the issues I've been working through on unix.stackexchange: http://unix.stackexchange.com/quest...ause-of-linux-kernel-panic#comment85760_60574

Is there any other setting besides the VCORE offset to up the idle voltage to prevent this from happening? Also, is there any quick test to validate that it's stable enough?

I would go to constant voltage, but I've had very little success with that. If I set my voltage to 1.40V and I load the machine with Prime95 or with Linpack, the voltage drops to about 1.38V and crashes basically immediately, despite having ultra-high LLC at 75%. Any ideas as to settings to boost in order to prevent this issue? I'd be happy to give any logs I can in order to help debug.

i have been seeing this problem quite often across different posts.

problem is of course the voltage adjustment over dropping. and there are a few things that will do that. Speedstep, LLC, and offset voltage adjustments.

If you use Fixed Vcore, setting LLC on HIGH from medium usually solves this.
for OffSet Vcore, you might need to keep LLC on ultra-high. If already so, you will need to + a bit more on the offset.

LLC only adjusts on load though. And can cause one to overcompensate with the base voltage... Personally, I would LOWER LLC, and raise the base voltage so you have more at an 'idle' state and around the same upon load.

What doesnt make sense to me is the experience you are sharing about offset voltage vs fixed. It shouldnt make a bit of difference. With offset, what is your load voltage?

In the meantime think about this...An offset of .30+ is .3v over the stock voltage which can be as high as 1.25v. Say its 1.20v, add .3v and that is 1.5v. So if you are only giving it a fixed amount of 1.4v that is LESS than your offset. So the bad luck was simply a misunderstanding of the actual voltage I would imagine.

EarthDog said:

LLC only adjusts on load though. And can cause one to overcompensate with the base voltage... Personally, I would LOWER LLC, and raise the base voltage so you have more at an 'idle' state and around the same upon load.

What doesnt make sense to me is the experience you are sharing about offset voltage vs fixed. It shouldnt make a bit of difference. With offset, what is your load voltage?

In the meantime think about this...An offset of .30+ is .3v over the stock voltage which can be as high as 1.25v. Say its 1.20v, add .3v and that is 1.5v. So if you are only giving it a fixed amount of 1.4v that is LESS than your offset. So the bad luck was simply a misunderstanding of the actual voltage I would imagine.

Click to expand...

Interesting. I finally got it stable by increasing my offset to 0.030V.

With offset 0.030V, my load voltage is 1.421V. It was stable for benchmarks and burn tests/Prime95 at offset 0.025V, but would crash at 25% or less load.

I'm fine being on offset voltage, and I'd like the benefit of less voltage on the CPU when idle, but yeah. So lowering LLC and upping the static VCORE would get me a more consistent voltage? I might play around with it later, but for now I'm enjoying my 4.8GHz clock These issues with idle crashes just need to be made known to newbies like me who immediately suspected it was the video card crashing or X server or something else, anything but my stable OC which can survive ridiculous burn tests and P95 runs. Lost a lot of time on this, but it was a learning experience.

I might experiment and see what it takes for me to get to 5.0 and beyond with an offset voltage, but I'll have to hear back from the other members in the forums who have a daily driver at >=5.0GHz about processor degradation and how long they've had that clock for.

After a while of seeming stability, it crashed again and, well, that means I've left offset voltage for now.

The only way I've found to keep voltages stable so far is to completely disable LLC (0%) and to turn off Intel SpeedStep, which has the unfortunate side-effect of running all cores at 4.8GHz all the time, generating a lot more heat and thus a lot more noise. I feel like there's something I'm definitely doing wrong here.

I set manual VCORE to 1.420V, which is TOTALLY stable for my OC, and it crashed very soon after starting IBT. The voltage dropped to 1.4V when I started IBT, then suddenly dropped to 1.32V, which BSOD'd, needless to say. WTF is wrong with my configuration? Do other people have to do this in over to achieve what would appear to be pretty simple OCs?

So the bad luck was simply a misunderstanding of the actual voltage I would imagine.

And still is vic, yes.

@ rfk - you disabled/lowered LLC and set the same voltage in the bios. Due to vdroop (which LLC eliminates) the drop is below your stable voltage. He needs to enable LLC again and make sure his stable ACTUAL LOAD voltage is the same. Read the first line of my post again as I mentioned that point already. All I stated was to lower LLC, and raise the base voltage to compensate. This should resolve your problem when done properly.

I finally got it sorted out (I think with the offset voltage). Though I was totally stable at only 0.025V offset when pounding the CPU, I found that it would crash if I quickly cycled the CPU from 0% to 100% load. My test was to simply start and stop IBT over and over again until it crashed. I couldn't get it stable until 0.040V offset, which is acceptable, as this boosts the idle voltage too. The sad thing is that at that offset, I can run 10xIBT@Standard at 4.9GHz, but not bigger, so I might just add 0.005V to the offset and get up to 4.9GHz.

As far as the static voltage is concerned, I am completely lost. When I set my static VCORE to 1.420V, I assume that it'll stay at that voltage when LLC is off. I would assume that if LLC is ON, I'll see my voltage spike higher, not lower. However, this is not happening; the voltage is spiking in the wrong direction and crashes happen.

Is it reasonable for me to expect that if I give it 1.420V that it will only go higher than that voltage with LLC? And if LLC is disabled, that it won't go too much of anywhere?

It seems silly to me to have to set something ridiculous like 1.440V in order to get my CPU to have 1.415V at load. Why does the voltage ramp down under load? I would expect the opposite.

Again, its called vrdoop. The WHY that exists Im not sure, but its there and LLC is what takes care of it.

With LLC disabled, the intel spec (includes vdroop) will be what you see. LLC prevents that vdroop from happening.

Ah. I see now. I had my best success with static voltages and LLC set to >=75%. At 100%, it was awesome, but threw much more (unneeded) voltage at the CPU in exchange for preventing vdroop. At least this is starting to make sense now, I've been losing my mind on this.

I really wish that offset voltage allowed another field for minimum CPU voltage so as to be able to have a lower offset while still maintaining stability at lower clocks. In these forums, has it come up a lot with offset voltages to have idle BSODs?

Furthermore, does anyone know of a tool which will oscillate the CPU from ~0% to 100% in order to test stability of offset voltage? It seems like something like this is necessary when OC'ing with offset voltage.

The stress tests do that... but its a manual process. Why do you have to 'oscillate' to test for stability? I dont know of anyone that needs to do that...

I have to oscillate because that's when I get a crash. I don't crash under full load or (necessarily) at zero load, but somewhere inbetween. As the frequencies change, so do the voltages, and at some point along the way, not enough voltage is supplied to the CPU and I get a BSOD. By repeatedly starting and stopping IBT, I cause the CPU to oscillate between idle and full load, so as to reproduce the crash. I was able to test this last night by setting my offset voltage at around 0.030V and while it could survive a huge IBT, it couldn't handle starting and stopping IBT repeatedly and would crash. Upping it to 0.040V seemed to fix the problem, as I can now start and stop Prime95 and IBT repeatedly and will not experience the crash.

Get off offset and use fixed if voltage increases do not help.

Otherwise, that isnt a realistic test as how often is that really happening?

How often is the crash really happening? Sporadically and often at the worst times. How often is my computer oscillating between full load and light load? Often. Starting programs, opening things, etc. It just happens. Running something like IBT and turning it on and off repeatedly takes me through the whole spectrum of CPU frequencies and voltages, so I can reproduce the problem faster.

Is it just me or does offset voltage seem like a dark art? Black magic? Secrets upon secrets?

I might move back onto static voltages if this turns out to not be stable, but now that I've determined more or less a way of testing it, I might be lured into living on the mysterious edge that is offset voltage.

Its just odd. Never have I seen anyone test like that for any reason, none the less a good one.

Another work around is to leave the speed/voltages at your overclocked speeds. That way there isn't a fluctuation on anything.

By turning off SpeedStep, I was able to keep things pretty stable, but that sacrificed my nice idle temps, voltage, and sound level. In any case, I've been working on this machine all day without any crashes, so it seems pretty stable.

low CPU load crash? turn off the the auto-voltage dropping protocols. like speedstep, and use Positive Offset. also I believe setting LLC to ultra-high on offset will eliminate most problems there too.

If you disable LLC, then prepare for cold boot problems with a reasonable core OC.

I ran into a problem back in 2011, the core OC passed every test I ran, but failed to POST after I turned it on when it was off. After I power-cycled, it POSTed.