High voltage / temp issues with stock 4670k

Hi guys,

Built my current rig about this time last year, using a 4670k on an asus z87 hero motherboard and cooling it with a h80i with two noctua NF-F12s in push / pull.

I ran a small FFT test on Prime 95 today for about 30 mins and noticed a couple of strange things, basically I want to know if they're normal or not.

With my 4670k at its stock turbo of 3.8Ghz, CPUz was reporting it as pulling 1.23v. Is that not a little high for 3.8Ghz? I've checked my BIOS and the core voltage is set at 'auto' which is the default. Come to think of it, I think the only setting I have changed in the BIOS is to turn XMP on (I use 8GB of Corsair Vengeance 2133Mhz ram). Having said that, my BIOS also reports the DIMM voltage to be just shy of 1.7 volts despite showing that it should be 1.65v as per the XMP profile.

Is this normal?

As for my temperature issues (presumably coming from the rather high voltage that my cpu needs for stock) after running Prime 95 for 30 mins my cpu reached an average core temp of 80-84 degrees, with a max core temp of 85 degrees.

I booted up corsair link to check temps reported there and it shows my pump speed as 2180 rpm (which I believe is fine?) but the h80i temp as only 39.3 degrees.

I might be talking out of my backside here, but that suggests to me that I'm getting appalling thermal transfer between the cpu die and the heatsink. Obviously the first assumption would be that I've installed my h80i incorrectly. Over the past year I've re-seated it probably 4 or 5 times now, making absolutely sure that everything is installed as it should be and eliminating common mistakes. I'm using AS5 thermal paste and seem to be getting good coverage, but not using too much, everything is screwed down securely and the backplate is solid as a rock (many people report loose backplates).

If I am installing it incorrectly, I really don't know what else I can do differently to get better contact?

My Noctua fans are designed for low acoustics rather than blistering performance, but that said, they're in 'standard' mode in the BIOS as opposed to 'silent' mode and I really don't think I should be getting terrible performance from them, should I?

I'm kind of at my wits end with my temp issues, it would be really nice to at least get a moderate overclock on my cpu, considering the cost of my h80i + 2x NF-F12s + an unlocked processor.

If I was only going to run at stock, I might as well have just used the stock heatsink and bought the 4670 instead!

Help?!

It's prime doing it. Prime uses AVX coding which causes Haswell to draw a extra 0.1v when loaded. If you use IBT it should show the normal voltage.

Ignore the water temp which I presume is what Corsair link is telling you. Download Realtemp and see what the cpu is hitting.

pastymuncher said:

It's prime doing it. Prime uses AVX coding which causes Haswell to draw a extra 0.1v when loaded. If you use IBT it should show the normal voltage.

Ignore the water temp which I presume is what Corsair link is telling you. Download Realtemp and see what the cpu is hitting.

Click to expand...

So the standard voltage draw for a 4670k with turbo enabled is 1.22v at x38 multiplier? That still seems a little high.

Also, ignoring the water temp still leaves my core temps in the low 80s at stock under full load, and these were taken using HWmonitor. Surely something isn't quite right considering my setup and the fact I'm still at stock?

try a lower level loadline calibration

and use cpu offset to bring stock voltage down

usually around 1.2v for stock clocks

My stock load voltage is only 1.184v so if yours is still hitting 2.2v using IBT and not prime then it's too much. I think you need to manually set the voltage. Asus boards tend to overvolt if left on auto. If you are still in the 80's (also not running prime) at stock then there is something very wrong. Maybe you have a duff cpu. It wouldn't be the first we have seen recently.

OK,

I've just done a few tests in IBT using the maximum setting and using all threads etc, CPU-z is still showing 1.23v being used at 3.8Ghz core speed.

The funny thing is sometimes the voltage will reduce to ~1.15v while still operating at 3.8Ghz (and still under full load) and it runs fine - this might be a stupid question, but if the CPU is able to run just fine at 1.15v at its current clock, why is my motherboard feeding it the extra voltage?

Temps were pretty much the same as Prime95, low to mid 80s, not very good at stock I don't think. I basically have no room to overclock at all. Most people seem to be happy with overclock temps when stress testing yields values <80 degrees.

I'm new to all this (never overclocked before at all) so I can look into load line calibration and cpu offset etc, but I probably won't have much idea what I'm doing.

Is it worth increasing the turbo clock in the bios a little (say to 4Ghz) and leave the voltage settings as they are, and see where that gets me? I don't mind my current thermals for 24/7 use really (around mid 50s while under normal load, low 30s at idle), I just wouldn't be happy if the only way I'm going to get better single core performance out of my cpu is to pump even more volts through it - It doesn't look like I have any thermal headroom whatsoever.

pastymuncher said:

My stock load voltage is only 1.184v so if yours is still hitting 2.2v using IBT and not prime then it's too much. I think you need to manually set the voltage. Asus boards tend to overvolt if left on auto. If you are still in the 80's (also not running prime) at stock then there is something very wrong. Maybe you have a duff cpu. It wouldn't be the first we have seen recently.

Click to expand...

Just out of curiosity, considering we have the same cpu and motherboard combination, what voltage is your cpu set to for your 4.6Ghz overclock?

I realise every chip is different and yours seems to be fairly good if your stock load voltage is only 1.184, but I'm curious to see just how far off the average my cpu seems to be.

Also, what sort of temps do you get out of it while stress testing, if you don't mind me asking?

My voltages are applied as a offset so that the voltage drops at idle but I need a total of 1.265v for 4.4Ghz, 1.285v for 4.5Ghz, 1.325v for 4.6Ghz and silly volts for 4.8Ghz which got me the top i5 in the cinebench thread. I also have the cpu cache overclocked as well. To get the best out of Haswell you not only need to play with vcore but cpu input voltage and cpu cache voltage as well.

My temps are not comparable to other set ups as i have a pair of top end 120.3 rads mounted in a box on a windowsill sucking cooler air in from outside. Temps when running IBT are low-mid 50's.

Your 1.23v should get you to 4.2ghz or so. Personally I think the only way to sort your issues out is to manually input the voltages.

I've been doing some testing this afternoon with basic over clocking.

I reduced my vcore to 1.2v (manual) and was able to post at 4.0, 4.2, 4.4 and 4.5Ghz. I bsod'd when attempting to boot at 4.6.

at 4.5 and below I was stable for small gameplay sessions, cinebench, superpi, and intel burn test (up to and including the maximum stress test option). For some reason though at 4.5, 4.4 and 4.3Ghz I would either bsod or my computer would just restart while running prime95 small FFT.

I would imagine that these are actually pretty positive results, considering that perhaps a small increase in voltage would render the over clock stable.

Problem is that my current cooling probably isn't up to the task. At the 4.4 and 4.5 over clock my temps were hitting a core average of ~85 during IBT very high or maximum setting.

I don't really know whether to see if I can get 4.2 completely stable at 1.2v, or stick with stock clocks and just lower the voltage, I reckon I can probably get it stable at 1.5-1.75v at stock, that would reduce my temps quite a bit.

Another thing that has thrown me a bit while doing my testing this afternoon is that while sticking at 1.2v vcore I thought that my temps would always be pretty much the same regardless of clock speed?

Maybe I'm stupid but I thought that when using the same program (so, for example the AVX instruction set used by prime95) and the same cooling solution / fan speed that temps should be reasonably similar whether the core is clocked at 4.0, 4.2, 4.4, when using the same voltage.

My results clearly show an increase in temps when increasing core speed but keeping voltage the same - how does that work?

The voltage going through the die is the power, and heat is a byproduct of the power, is it not? o.0

Shouldn't the only thing that affects temperature be the voltage?

(Probably showing my woefully inadequate knowledge of electronic principles here lol)

its down to a constant fixed voltage of 1.2v,doesnt matter what load is on the cpu

you need to learn to use cpu offset,this lets the cpu enter idle mode and reduce voltage at idle

it will use around 0.800v then,and switch to full speed and 1.2v if under any load

wazza300 said:

its down to a constant fixed voltage of 1.2v,doesnt matter what load is on the cpu

you need to learn to use cpu offset,this lets the cpu enter idle mode and reduce voltage at idle

it will use around 0.800v then,and switch to full speed and 1.2v if under any load

Click to expand...

Obviously for testing my overclock I used a manual vcore setting to prevent Prime95 and other AVX instruction based stress testing applications drawing more voltage than my desired setting.

After determining a stable overclock I have switched back to adaptive vcore in order to utilize the power saving settings available.

Also the load on the CPU does increase temperatures at a set vcore value. Using a manual value of 1.2v I was idling at roughly 40 degrees, with load temps of ~80+ - unless I've misunderstood what you were trying to say?

if its drawing more voltage than you want just use few clicks less cpu offset and or a lower level loadline calibration

that's what I do,i never bother with fixed voltage

Don't use adaptive mode, use offset mode.

pastymuncher said:

Don't use adaptive mode, use offset mode.

Click to expand...

Is there a reason why adaptive shouldn't be used? Surely it accomplishes the same goal as offset?

I am having doubts about this myself now after re-reading the guide below. At Haswells launch all the guides said to avoid using adaptive voltage as it can allow the cpu to draw more volts than you want it to. More volts equals more heat and that's exactly what you don't want with Haswell. I found from my own testing that AVX coding still draws a extra 0.1v even if you use offset voltage. What I don't understand in the guide is that it says using a + offset also gives a higher idle voltage yet my idle voltage is 0.768v. The guide here is for ROG boards and it explains what each setting actually does. It helped me get my head around the options on this board.

pastymuncher said:

I am having doubts about this myself now after re-reading the guide below. At Haswells launch all the guides said to avoid using adaptive voltage as it can allow the cpu to draw more volts than you want it to. More volts equals more heat and that's exactly what you don't want with Haswell. I found from my own testing that AVX coding still draws a extra 0.1v even if you use offset voltage. What I don't understand in the guide is that it says using a + offset also gives a higher idle voltage yet my idle voltage is 0.768v. The guide here is for ROG boards and it explains what each setting actually does. It helped me get my head around the options on this board.

Click to expand...

Yeah, the guides I'd read stated that it is best to use a manual vcore setting while testing an overclock (using stress testing applications) specifically because the auto and adaptive settings will draw more voltage than you specify, but that it was fine to switch back to adaptive after you've confirmed the clock stable.

With my user inputted value of 1.2v even with a manual setting cpuz would report a draw of 1.216v while using anything with an AVX instruction set. Now that I've switched to adaptive though, it seems to cap at 1.2v when using any normal program and obviously undervolts to 0.704v at idle.

I had a interesting experience this afternoon. I saw that there is a new bios available (1505) that is supposed to improve performance so I flashed to it only to bsod every time I tried booting into Windows. I cleared cmos and re-entered all of my settings, still no joy. Switched back to manual voltage (bsod were 0124 errors so vcore) and even giving it a extra bit of voltage still couldn't boot into Windows. Tried setting to optimal defaults and it got into Windows. Only thing now was ther was no overclock. Goes onto the ROG forums and it would seem that 1505 is rather buggy. Someone started a thread on there saying that some of the default voltages were higher than the previous bios. Anyway, flashed the board back to 1402 and all is well again. I don't suppose you are on 1505 by any chance?

Another thing I noticed while I was on their forum is that the latest Realtek audio driver (7218) is also very buggy. I have suffered from very quiet sounds for a while now and have been trying to get it working properly. After reading on there I uninstalled that driver and installed the one before (7023) and instantly jumped out of my seat after the pc rebooted as the windows boot sound was so loud. All is good again with my mobo.

pastymuncher said:

I had a interesting experience this afternoon. I saw that there is a new bios available (1505) that is supposed to improve performance so I flashed to it only to bsod every time I tried booting into Windows. I cleared cmos and re-entered all of my settings, still no joy. Switched back to manual voltage (bsod were 0124 errors so vcore) and even giving it a extra bit of voltage still couldn't boot into Windows. Tried setting to optimal defaults and it got into Windows. Only thing now was ther was no overclock. Goes onto the ROG forums and it would seem that 1505 is rather buggy. Someone started a thread on there saying that some of the default voltages were higher than the previous bios. Anyway, flashed the board back to 1402 and all is well again. I don't suppose you are on 1505 by any chance?

Another thing I noticed while I was on their forum is that the latest Realtek audio driver (7218) is also very buggy. I have suffered from very quiet sounds for a while now and have been trying to get it working properly. After reading on there I uninstalled that driver and installed the one before (7023) and instantly jumped out of my seat after the pc rebooted as the windows boot sound was so loud. All is good again with my mobo.

Click to expand...

No, I'm on 1402 as well. I generally don't update my bios unless there's a problem I want to fix. I had been getting 124 errors with default settings for about a year on the release bios. Updating fixed that problem completely. I didn't realise 124 errors are vcore related - perhaps as optimized defaults the mb wasn't feeding my cpu enough voltage. 1402 has seemed pretty stable for me so far, probably going to stay on this one for a while I think.

As for the audio driver, I still have the release version, but don't think it matters because I use an essence stx.