Overclock.net banner
Cancel
Create thread New Forums

"never push your voltage beyond 1.4V"

Jump to Latest
81 - 100 of 124 Posts
#81 ·
bhav said:
And now I'm kind of furious at my MSI board pumping 1.375v through my lovely new 13600KF at 'stock' settings.
Click to expand...
Recent MSI motherboards with stock settings (no change whatsoever to CPU clocks) configure AC Loadline to 0.80 mOhm using a default LLC (having an apparent impedance, using VCC sense, of 0.96 mOhm). The "Lite Load" setting indirectly configures AC Loadline through a simpler user-facing number.

An AC Loadline of 0.8 mOhm would seemingly be within specifications, even though it does tend to use relatively high voltages. Setting it to the maximum specification value (same as DC Loadline) can easily bring voltages well above 1.4 volts under load, even with stock clock settings.
 
#82 ·
SSBrain said:
Recent MSI motherboards with stock settings (no change whatsoever to CPU clocks) configure AC Loadline to 0.80 mOhm using a default LLC (having an apparent impedance, using VCC sense, of 0.96 mOhm). The "Lite Load" setting indirectly configures AC Loadline through a simpler user-facing number.

An AC Loadline of 0.8 mOhm would seemingly be within specifications, even though it does tend to use relatively high voltages. Setting it to the maximum specification value (same as DC Loadline) can easily bring voltages well above 1.4 volts under load, even with stock clock settings.
Click to expand...
Its not really stock though is it if dropping the lite load to 1 reduces the CPU voltage to 1.25 which is much more correct for stock voltage.
 
Save
Reply Quote Rep+ Rep+
#83 ·
Decreasing CPU Lite Load to 1 (or more directly, AC Loadline to 0.01 mOhm) could be undervolting your CPU under load, relatively to its built-in voltage-frequency curve.

Skatterbencher described a way to retrieve what would be the "true" stock voltage (i.e. the VF curve) at various frequencies on MSI motherboards, you could try it, and then check if during actual tests in Windows you're below these values:


Font Parallel Number Terrestrial plant Rectangle
 
#84 ·
I tried this the other day with my 13600K, following all instructions, but I'm not sure why he suggested adaptive instead of auto. All other things being equal, auto provided lower voltages than adaptive, so I'm not sure which one to trust.
 
#85 · (Edited)
Possibly Auto voltages could be using other voltage modes or changing other settings under the hood. The aim here is obtaining BIOS voltage readings as close as possible to the processor's built-in ("fused") VF curve, which is intended to be used with Adaptive voltages.

EDIT: it could be that recent microcode or BIOS updates have changed the behavior slightly because I'm getting somewhat different values than when I tested the same procedure a couple months ago. It seems also more dependent on the LLC setting than before.
 
#86 ·
Having now used a 12600 non k, 12600K and 13600KF this past year, I can say that lite load set to 1 is applying correct voltage and not causing undervolting under load (tested in Cinebench).

Both 12600K and 13600KF use 1.25v at stock, 12600 non k uses 1.2v. This is far more correct than the default voltages set by lite load 12 at stock settings.

Also 13600KF stock frequency of 5.1 all core at 1.25v, my 12600K needs 1.38v for that.
 
Save
Reply Quote Rep+ Rep+
#87 · (Edited)
Not all CPUs are created equal. Also, the effective load voltage with the minimum AC Loadline depends on the configured LLC.

With the above procedure I previously got about 1.22V in BIOS at 4.7 GHz (stock all-core frequency for my 12700K), but with LLC8 (the minimal setting with MSI motherboards) and AC Loadline of 0.01 mOhm with Windows stress-testing at 4.7 GHz I get load voltages below 1.10V and the CPU is not stable.

(It's never been stable with these settings, I doubt it's due to degradation ;))
 
#88 ·
Look at this
 
Save
Reply Quote Rep+ Rep+
#89 ·
Bit of an oc newb here. I measured cinebench voltage and total cpu power draw in hwinfo at stock with an 13900k and z790-e gaming wifi. Adaptive boost and Asus multicore enhancement are both disabled. Bios is 0806.

At stock:
Cinebench single core voltage and power - 1.46v / 47w
Cinebench multicore voltage and power - 1.24v / 253w

Next, I turned on 'sync by core' in bios and set p cores to 61 60 59 59 58 57 55 55.

Cinebench single core draw increased to 1.534v / 54-56w at 6100mhz. Temp of the single core is 72-74c. Is this increased voltage/current safe for every day usage? It did pass cinebench. Thanks in advance.
 
#90 ·
In this thread three different views were/are being debated:

  • 1.52V+ are unsafe at any current greater than 0
  • 1.52V+ are always safe
  • 1.52V+ are safe as long as all thermal and electrical specifications are respected

Would you really trust an opinion on the subject with no agreement on it yet?
 
#91 ·
Ocfailfish said:
Bit of an oc newb here. I measured cinebench voltage and total cpu power draw in hwinfo at stock with an 13900k and z790-e gaming wifi. Adaptive boost and Asus multicore enhancement are both disabled. Bios is 0806.

At stock:
Cinebench single core voltage and power - 1.46v / 47w
Cinebench multicore voltage and power - 1.24v / 253w

Next, I turned on 'sync by core' in bios and set p cores to 61 60 59 59 58 57 55 55.

Cinebench single core draw increased to 1.534v / 54-56w at 6100mhz. Temp of the single core is 72-74c. Is this increased voltage/current safe for every day usage? It did pass cinebench. Thanks in advance.
Click to expand...
What was the total watts with the overclock?

13900Ks are degrading fast when pushed past 253w so that should be your limit, ideally you overclock at however much voltage can stay withing 253w max.

This is inadvertently confirmed with the 13900KS specs, they are also 5.4 all core boost, same as 13900K because it looks like even Intel can't push them further without issues. 13900K and KS are pushed to the max out of the box, they have very little headroom and heavy OCs are quite risky on them.

Much happier with my cheapo 13600KF with gold IMC now.
 
  • Rep+
Reactions: Ocfailfish
Save
Reply Quote Rep+ Rep+
#92 ·
Falkentyne said:
I don't remember. Back in those days, people thought that 'transient spikes' degraded CPU's, not sustained current.
Click to expand...
Right, I saw your older posts on XS. That was around the time IBT and Linx were the rage IIRC. Yes, very little mention of "current" and power estimations made through VID so hard to correlate true power and true current.

On a side note, it would be interesting to have some extreme cooling reports as to how long those LN2 CPU's lasted under extreme voltage.

bhav said:
And now I'm kind of furious at my MSI board pumping 1.375v through my lovely new 13600KF at 'stock' settings.
Click to expand...
My MSI board was happy to "pump" 12V at stock through the CPU. Not sure why some OVP didn't kick in such as turning on the low side FET's. Maybe they were worried about blowing the crap out of the faulty FET and making a mess rather than killing my CPU's. Or maybe there is some protection and it didn't work.
 
#93 ·
SSBrain said:
In this thread three different views were/are being debated:

  • 1.52V+ are unsafe at any current greater than 0
  • 1.52V+ are always safe
  • 1.52V+ are safe as long as all thermal and electrical specifications are respected

Would you really trust an opinion on the subject with no agreement on it yet?
Click to expand...
You're looking at this the wrong way. all we can do is analyse this based on individual user results.

If you want a 100% concrete ironclad proof conclusion. then you would only run CPUs at stock and abide by Intel / AMD's specification.

There is no way to guarantee a 100% 'safe setting' for any CPU, one 13900K can be completely different to another, low bin vs high bin.

Now if someone has a high bin and is comfortable running 1.5+v for ages without issue, then their conclusion is '1.5v is safe'.

If another user gets a crap bin and degrades at 1.4v, their conclusion is '1.4v is unsafe'.

Now in general, anyone can get the crap bin, in which case reading and following the first opinion is going to cause disaster. Even if some chips degrade like the second example, then that would be my general conclusion to anyone buying a 13900K and saying 'Hi, this is my first time overclocking, what do I do?'.

This also explains why motherboards doing 'auto enhancements' like LOL Gigabyte's instant 6.0 is universally terrible, so what happens if you turn that feature on with a crap bin CPU and it degrades or even fails?

You cannot reach a 100% conclusion on safe settings because every chip is going to have different tolerances.
 
Save
Reply Quote Rep+ Rep+
#94 ·
I can only reiterate that Intel appeared to be comfortable with stock voltages in the order of 1.52-1.56V for single-core frequencies with the 12900KS (not always due to the highly varying silicon quality, but some samples definitely did).

Even taking into account TVB voltage optimizations (automatic downward voltage adjustment with temperature), that means that in idle or light loads (definitely non-zero current, especially at short time scales) all CPU cores could potentially be in the 1.45-1.50V range and yet be within specifications.

Font Line Red Technology Screenshot


If stock CPUs with all limits in place and respecting all operating parameters by specifications are fine with these voltages, I don't think the assertion that 1.52V is (basically) never safe will be correct. And as I mentioned elsewhere, the DC loadline range starting from 0 mOhm and the AC Loadline parameter used in modern CPUs would throw out of the window the formula used for arriving that 1.52V is at 0 amps with 1.1 mOhm anyway.

Of course, the safest answer to the previous question is that high voltages will slowly but certainly degrade the CPU over time (i.e. eventually render it unable to maintain its default frequencies at the pre-programmed VID) and that degradation will have an exponential dependence with voltage, so for very long-term usage it will be better to keep it low. It could be assumed that at the edge of Intel specifications the CPU will degrade within the warranty period, but this might possibly occur faster depending on CPU settings and/or luck; nothing certain in this regard.

Unfortunately no definitive answer can be given yet other than the 'safest' one.
 
#95 ·
I really can't trust voltages shown on an Asus board, those are not stock, they are whatever additional enhancements mobo manufacturers use.
 
Save
Reply Quote Rep+ Rep+
#96 ·
Those voltages (in detail, the average voltage between the worst and the best core) correlate with the "SP" rating that these motherboard give. People have been using that for processor binning.

I gave the link earlier already, but here are a couple hundred results made on the same motherboard with one BIOS version, compiled by 4 different people:

docs.google.com

ADL Binning Mew 2022 Update

docs.google.com docs.google.com

Slope Rectangle Plot Font Parallel
 
#97 ·
Perhaps a better way could be as follows.

  • Set minimum/default LLC (1.1 mOhm on 125W CPUs).
  • Set AC/DC Loadline to 1.1 mOhm. This is the default, unchangeable AC/DC LL configuration on B660 motherboards.
  • Set the current limit to specification and clocks to default values.
  • Disable TVB voltage optimizations, set power limits to a high value and system fans to the maximum.
  • Boot the OS, observe the voltage after loading cores in a stress test at a progressively higher count (1, 2, 3, ... etc).

With my CPU, the 'default' settings use voltages generally above 1.35V, often in the 1.40V range. They do appear to get close at times to the previously suggested loadline having 0A at 1.52V (which I'm doubtful about), but never exceed it. Testing was performed in Blender (equivalent to Cinebench R23).

Previously I had an all-core 52x overclock with Adaptive voltage of 1.495V and 0.05/0.80 AC/DC LL. To make the CPU be within the same range of voltage-currents as with 'defaults', I only had to turn this into a 1-3c 52x, 4-8c 51x overclock. All-core heavy loads get more heavily current limited, although frequencies are still higher than with default settings.

It would be interesting to see data with true default settings (except power limits which aren't a specification), having 1.1 mOhm LLC, ACLL=DCLL=1.1 mOhm and a 12900KS or other CPUs with a VID >1.4V for the highest frequencies. Since with a high ACLL the CPU appears to get overvolted, I think it would easily get close to 1.5V in few-thread loads.


Rectangle Slope Plot Parallel Font


Data for 'defaults':

Font Pattern Number Parallel Screenshot



Data for the overclock:

Font Pattern Number Parallel Screenshot
 
#98 ·
AC Loadline should never, under any circumstances, be set higher than 0.6 mohms.
Setting AC Loadline to 1.1 mohms on a 13900K will give you overtemp alarms booting windows (80C or higher!) and at least 1.65v IDLE, if you are foolish enough to set this at the max P-core turbo ratio.
I posted a chart showing how AC Loadline changes native CPU VID at a much lower P-core frequency (and I was using fixed vcore to prevent ACLL from influencing vcore). At 1.1 mohms, CPU VID is boosted above native by over 300mv (!).
So please do not use AC/DC loadline interchangably like this. I know Asus did this in the past. This was fine on Z390 and prior, when ACLL was the inverse of loadline calibration, but not on Z490 and newer.
 
Save
Reply Quote Rep+ Rep+
#99 ·
Falkentyne said:
AC Loadline should never, under any circumstances, be set higher than 0.6 mohms.
Setting AC Loadline to 1.1 mohms on a 13900K will give you overtemp alarms booting windows (80C or higher!) and at least 1.65v IDLE, if you are foolish enough to set this at the max P-core turbo ratio.
I posted a chart showing how AC Loadline changes native CPU VID at a much lower P-core frequency (and I was using fixed vcore to prevent ACLL from influencing vcore). At 1.1 mohms, CPU VID is boosted above native by over 300mv (!).
Click to expand...
I don't see this on my MSI Z690 motherboard using adaptive voltages. I even tried disabling C-states with both AC and DC loadlines to 1.1 mOhm and minimum LLC since I recalled you mentioned something along these lines in the past (although all C-states are intended to be left enabled, not disabled), but haven't observed the behavior you're mentioning.

Voltages are indeed on the high side relatively to the built-in VF curve, but they definitely do not reach 1.65V in idle. In the images above I posted the voltages observed under load with varying numbers of threads; you can see they are still reasonable and below the loadline slope at 1.1 mOhm with 0A @ 1.52V.

Falkentyne said:
So please do not use AC/DC loadline interchangably like this. I know Asus did this in the past. This was fine on Z390 and prior, when ACLL was the inverse of loadline calibration, but not on Z490 and newer.
Click to expand...
I don't understand what you mean by using them interchangeably.

Recent-generation Intel CPU datasheets directly state that AC_LL can be the the maximum the "same as DC_LL". This should be expected to work without issues.
 
#100 ·
Falkentyne said:
AC Loadline should never, under any circumstances, be set higher than 0.6 mohms.
Click to expand...
I also have an i7-10750H laptop, and according to HWInfo the AC/DC loadline is seemingly set to 1.1/1.1 mOhm there. Could it be that it isn't supposed to work without C-states, if disabling them is the reason for the high voltages you observed? Intel does point out in the datasheets that all C-states must be enabled, after all.

Rectangle Slope Font Parallel Number



From the 13th gen datasheet you linked a while back (which I still cannot find in the Intel website):

Rectangle Azure Font Parallel Screenshot
 
81 - 100 of 124 Posts
About this Discussion
123 Replies
16 Participants
Last post:
Ocfailfish
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling