Originally Posted by Sleakcavi
Alright just got home and was able to run the test you mentioned. 1.20V set in BIOS. Idle Vcore 1.190, loaded it stays at 1.19 with a minimum of 1.181 and maximum of 1.208 (I am assuming overshoot/undershoot there) Seems to me that it is in fact reading the Vcore correctly which is a relief because I only have Best Buy around which means I'm limited to MSI or ASUS MOBO and they don't typically carry the enthusiast grade boards in large quantities so a replacement would have been hard to come by (None in a 200 mi range).
On a side note what LLC should I run to start using in trying to lock down a stable OC now that I know I can trust my Vcore reading that I'm seeing? And this is a very noobish question but I shouldn't go over 1.42V in Bios trying to chase higher clocks correct? My ultimate goal is to find a stable 5.0 GHz OC but I know my chip will do 5.1 GHz with extra voltage and heat which when not stressing to test it very manageable by my cooling setup. Thanks for the help I very much appreciate it. I'm tryin to learn and its always best to ask questions from those more experienced than trying to sift through numerous videos.
There's a 15mv resolution on the vcore sensor I think, so that's accurate.
No that's not overshoot or undershoot. Overshoot and undershoot happens in microseconds (units="us" in some strange squiggly letters) and can't be measured accurately unless it's sustained. Usually undershoot and overshoot is spec'd to be around 40 microseconds in duration (Intel calls this "Virus mode") and not to exceed 200mv. This however is based with a -standard- loadline of 1.6 mOhms (meaning loadline calibration is NOT used, or set to default), as then vdroop prevents overshoot and undershoot from affecting you.
Loadline calibration (LLC) helps reduce vdroop, however this comes at the cost of worse transient response. Transient response, for lack of a better explanation, is basically when the "missing" vdroop that you removed with loadline calibration, comes *BACK* due to sudden extremely fast load swings, except coming back in -both- directions. And naturally, the more vdroop you remove, the more will obviously "come back". This is obviously going to be worse at higher current (amps), because vdroop is by definition = Resistance * Current. Resistance is the loadline mOhms value (default for loadline calibration=Standard (Normal) is 1.6 mOhms, and amps is well, current. Gigabyte and some Asrock and MSI boards have VRM monitoring of current via VR IOUT, but I don't think Asus boards have current monitoring.
The problem with a 0 mOhm loadline is, since all vdroop is removed, when load changes happen, the vdroop actually winds up coming back for extremely short (microsecond) durations--the higher the current, the more droop comes back during a wild load swing. This vdroop is a "SPIKE" (not a droop) when load changes from heavy load to light (or no) load, and a droop (drop) when load changes from no load to heavy load. The spikes, if they exceed 1.4v, can slowly degrade your processor, and the drops can cause BSOD/crashes. There isn't really any way around this, but boards with better VRM's and tighter switching frequencies will have better transient response.
Here is the worst case scenario of a 0 mOhm loadline with an extremely heavy load (like prime95 AVX small FFT). And again remember sensors cannot pick this up. you need an oscilloscope.
(Pictured in attachment below this post).
That's why LLC8 should be avoided when overclocking at higher voltages. . The only point to LLC8 is when on LN2, or at low overclocks/low voltages, when using low loads. Otherwise you gain basically nothing compared to using LLC6 + a higher bios voltage (or VID, if using auto voltages + AC Loadline=0.01 mOhms (remember AC loadline is NOT LLC !!))
LLC7 is "Ok" to use for gaming at average overclocks as current load isn't going to put you in the danger zone for spikes, but LLC6 is the best compromise. LLC5 has even better transients than LLC6 (remember: the more vdroop you have, the better the transient response will be, because vdroop helps "cushion" transient drops (as they become "part" of the vdroop) while also limiting spikes as well--the less LLC you use, the less the spikes will go past original bios voltage!)
Here's an example of virus mode on my gigabyte board:
5 ghz, 4.7 ghz cache. SVID OFFSET: ENABLED (this greys out all voltage control, even Auto voltage), IA AC Loadline=0.9 mOhms, IA DC Loadline=1.6 mOhms (DC Loadline=1.6 mOhms matches VRM Loadline Calibration=Standard, which is the Intel reference value--this keeps CPU VID=CPU VR VOUT (VCC_SENSE) vcore!), CPU VCore Loadline Calibration=Standard (1.6 mOhms)
Prime95 small FFT (29.8 build 3) with AVX enabled:
1.230v VCC_Sense (VR VOUT)--CPU no-die sense voltage full load--182 amps (Gigabyte has Current IOUT monitoring via VR IOUT): Stable (temps: 94C)
Idle voltage (VR VOUT) is 1.332v.
Fun fact: At the default IA AC Loadline value of 1.6 mOhms, if SVID offset is enabled, IDLE voltage is still the same as SVID Offset: Disabled (see below), but VR VOUT with prime small FFT AVX is 1.330v (!!!!!). That's 212 amps (yes, 212 amps!! That exceeds Intel absolute max for 9900K which is 193 amps). Temps reach 105C in LESS than 10 seconds! SVID offset increases load voltage to compensate for weak Silicon Lottery losers, while keeping idle voltage the same. That's why setting IA AC Loadline to 0.9 mOhms (instead of 1.6 mOhms) helps reduce VR VOUT at full load from 1.330v to 1.230v.
ok example #2:
5 ghz, 4.7 ghz cache, SVID offset: Disabled, CPU Vcore: Auto. IA AC Loadline=1.6 mOhms, IA DC Loadline=1.6 mOhms, CPU Vcore Loadline Calibration=Standard (1.6 mOhms)
Prime95 29.8 build 3. small FFT with AVX:
VR VOUT: 1.240v, Current: 184.250 Amps, Temps: 97C,
Idle Voltage (VR VOUT) is 1.404v.
(tests are stable because AC Loadline actually boosts the internal VRM voltage to 1.518v at load (SVID Offset: disabled), then full vdroop (1.6 mOhms) drops it down to 1.230-1.240v at load, with excellent transient response. SVID Offset:Enabled with AC Loadline=0.9 mOhms boosts the internal VRM voltage up to about the same (1.510v) at load, then full vdroop drops it back down safely.
Ok the point of this wall of text?
Manual voltage: 1.30v
Loadline Calibration=Turbo (= LLC6 on your Asus). 0.4 mOhms of Loadline calibration.
Prime95 small FFT w/ AVX:
VR VOUT 1.230v. Amps: 185.
Thread crashes in seconds.
(why? Because transient response penalty. Prime95 does not do a 100% sustained load constantly (nothing does), so an oscilloscope reading would show the voltage dropping repeatedly below 1.230v and spiking higher than 1.3v also).