Higher Vcore & natural vdrop VS lower Vcore & higher LLC: which is safest/best?

[O.T.T.]

Higher Vcore & natural vdrop VS lower Vcore & higher LLC: which is safest/best?

Hi folks,
once found a target Vcore voltage that is sufficient to guarantee full system's stability, which is the best and safest choice to achieve it?

Priority is (for 24/7 OC):

1) CPU preservation (no degradation and/or risk of damage, both in the short and long run, due to too many/too high voltage spikes or too high starting voltage set in BIOS: see example below)
2) efficiency, as the ability to better maintain voltage at the desired value

That said, as per the title I'd like to know your opinions about which path is best:

A) set a higher Vcore on BIOS (compared to the target) that under full load drop naturally until the desired target voltage (LLC set as "flat")
B) set a Vcore close to the target on BIOS and then keep it stable by higher LLC level (a somehow aggressive level could be needed)

This interests me especially for fairly high voltages.


Example: TARGET Vcore 1.370, LLC levels 1-8 (1 is the highest, 4 is "flat")

if A): Vcore set to ~1.390-1.400 + LLC 4
if B): Vcore set to ~1.355-1.360 + LLC 2

NOTE: Vcore set to 1.360 + LLC 3 is ~1.368-1.376 under full load according to HWInfo, almost 100% stable but not quite yet - so a more aggressive LLC level (2) is required. Another option could be to rise a bit the Vcore in BIOS, i.e. 1.365, in order to keep LLC @ level 3, but for now I want to know which is better between option A) and B) listed above.

INFO: CPU 9600K, MB MSI Z390 A-PRO. TEMPS ARE NOT A PROBLEM

Any feedback would be greatly appreciated! Thanks!!!

 

[RJARRRPCGP]

I find that keeping the Vcore lower with LLC level high enough to keep the Vcore from lowering by much, is the best. IOW, don't use the highest level, a.k.a. "extreme" LLC option. (At least on Asus, because the Vcore apparently goes higher than where set to, with the highest level.)

Early Asus motherboards with CPU LLC, only have enabled and disabled. (only on and off)

 

[O.T.T.]

I find that keeping the Vcore lower with LLC level high enough to keep the Vcore from lowering by much, is the best. IOW, don't use the highest level, a.k.a. "extreme" LLC option. (At least on Asus, because the Vcore apparently goes higher than where set to, with the highest level.)

Early Asus motherboards with CPU LLC, only have enabled and disabled. (only on and off)

Thanks for your reply!

It's been a while since I wrote this post and I got to find out myself: I managed to fully stabilize my target CPU frequency with the lowest LLC setting that "does something" (lev.4 on my MB).

I tried literally every possible combination and plausible range of voltages: I could see that seemingly similar settings which led to the same target Vcore, of which all apparently without significant Vdroop, actually worked quite in an unexpected way...I mean, for ex., that 1.335 Vcore + LLC 3 (~1.350-1.355) was not stable (did not pass all tests from my suite) whereas 1.345 Vcore + LLC 4 (~1.344-1.352) is ultra rock-solid under any stress level and application (I ran a LOT of tests, many 24h+).

I experimented even with LLC 2 (higher setting on my board): 1.325 + LLC 2 gave even a little bit more Vcore (~1.355-1.360) and at first it *seemed* stable - since passed many tests - but ultimately failed the most critical one on the very long run (after about 15h).

Therefore, comparing the 3 settings described above (and many other I experimented with), I could find that lower LLC works best even if the final achieved Vcore is a bit lower than what is achieved with other settings + higher LLC.

So yeah, I agree with you! What you wrote seems to be absolutely true, or at least it is for my CPU/MB combo

 

[RJARRRPCGP]

Prime95 failing after a couple of hours, especially as late as 15 hours in, means usually not enough cooling, in my experience.

Especially that Prime95 error incident I had, in probably 2007, with my Barton 3000+ AQZFA 03xx, (at least week 39 and locked multi) where Prime95 failed at 23 hours, because of the sun in the window, IIRC, LOL.

 

[O.T.T.]

Prime95 failing after a couple of hours, especially as late as 15 hours in, means usually not enough cooling, in my experience.

Especially that Prime95 error incident I had, in probably 2007, with my Barton 3000+ AQZFA 03xx, (at least week 39 and locked multi) where Prime95 failed at 23 hours, because of the sun in the window, IIRC, LOL.

That may be the case sometimes, but not mine: my temps are generally very good - like ~75C MAX under Prime 95 AVX. Also everything else stay quite cool: DRAM <45C, MOS ~75-77 MAX, MB <33C.
With the settings I chose at the end of my testing period, the rig easily passed all tests - included Prime 95 for >24h. In addition, I was able to raise the cache frequency to 4.6Ghz, whereas with the other settings I described I got issues with anything above 4.4 Ghz (another spy that the system was not 100% stable).

 

[EarthDog]

I prefer to set LLC to match what I set in the bios under load in windows... but only for my sanity.

That said, if you're stable at 1.35V, you're stable at 1.35V....it typically doesnt matter how you get there...at least in my experience. I also don't touch the cache as there are little performance gains from it.

I chose "B". As a (former) extreme overclocker and overclocking most everything still, this is what worked well for me.

 

[RJARRRPCGP]

If you go for a higher CPU OC, then it's possible that more than 55C, will result in an error or a system crash.

 

[EarthDog]

Prime95 failing after a couple of hours, especially as late as 15 hours in, means usually not enough cooling, in my experience.

If you go for a higher CPU OC, then it's possible that more than 55C, will result in an error or a system crash.

These CPUs are good to 100C before they start to thermally throttle (shutdown is several C higher). We tell everyone to keep these CPUs at 90C during stress testing. I hit 55C when I open up Chrome, bud! Temps are clearly not an issue for this guy and temps over 55C are PLENTY fine. It's not like he's rocking a barton from 13 years ago (relevance?)... he's got a modern Intel.

 

[RJARRRPCGP]

These CPUs are good to 100C before they start to thermally throttle (shutdown is several C higher). We tell everyone to keep these CPUs at 90C during stress testing. I hit 55C when I open up Chrome, bud! Temps are clearly not an issue for this guy and temps over 55C are PLENTY fine. It's not like he's rocking a barton from 13 years ago (relevance?)... he's got a modern Intel.

I'm talking about pushing a chip hard with extreme Vcore on modern chips, where the phenomenon is possibly back. And I wasn't talking about shutdown, I was talking about arithmetic-corruption-syndrome, where Prime95 errors out or Windows errors out, usually WHEA_UNCORRECTABLE_ERROR (0x00000124) or IRQL_NOT_LESS_OR_EQUAL. (0x0000000a) Even though the first BSOD code mentioned (0x00000124) can mean an ODT issue and more termination voltage being required. (on-die-termination)

Actually, my Q6600 needed very low temps, like less than 55C or I would get 0x00000124 for an apparent ODT issue. Sort of acts like it's BioNTech, LOL! BTW, I got a report from another person who also has a Q6600 G0 (most likely) with data code of "L804", IIRC. I think that means assembled in 2008, week 4 and it also had the dreaded bus/interconnect WHEA issue. For my Q6600 G0, it was with the FSB at only 367 Mhz, for only 3.3 Ghz core speed. And that was with a P45, FFS!

 

[EarthDog]

I'm talking about pushing a chip hard with extreme Vcore on modern chips, where the phenomenon is possibly back.

I don't buy it under that guise either... and the OP isn't using 'extreme' voltages. These chips are good to 100C. We say to stay under 90C for stress testing as most everything else is 10-30C cooler. I don't get these errors when overclocking modern intel and pushing it to 10C below its throttle point.

Not sure what Q6600 has to do with things in this thread (or the barton reference). Let's focus on helping the OP with his hardware and not anecdotes about irrelevant CPUs.

P95 failing after 15 hours is typically an instability, not temps. In most cases, you'll hit your highest temperatures within the first hour of P95 IIRC. It takes ~ 15 mins for an air heatsink to saturate and 30-60 minutes for an AIO/custom loop to saturate. P95 switches FFT length and such multiple times so it's likely something there. Not temperatures after time as the peak happened hours earlier.

 

[wagex]

I don't buy it under that guise either... and the OP isn't using 'extreme' voltages. These chips are good to 100C. We say to stay under 90C for stress testing as most everything else is 10-30C cooler. I don't get these errors when overclocking modern intel and pushing it to 10C below its throttle point.

Not sure what Q6600 has to do with things in this thread (or the barton reference). Let's focus on helping the OP with his hardware and not anecdotes about irrelevant CPUs.

P95 failing after 15 hours is typically an instability, not temps. In most cases, you'll hit your highest temperatures within the first hour of P95 IIRC. It takes ~ 15 mins for an air heatsink to saturate and 30-60 minutes for an AIO/custom loop to saturate. P95 switches FFT length and such multiple times so it's likely something there. Not temperatures after time as the peak happened hours earlier.

Agreed, I let p95 tickle 100c because nothing realistic else ever even gets it close, and I've never had an issue temp related with stability.

 

[O.T.T.]

I prefer to set LLC to match what I set in the bios under load in windows... but only for my sanity.

That said, if you're stable at 1.35V, you're stable at 1.35V....it typically doesnt matter how you get there...at least in my experience. I also don't touch the cache as there are little performance gains from it.

I chose "B". As a (former) extreme overclocker and overclocking most everything still, this is what worked well for me.

Thank you for sharing your experience and for your advice! As I told to RJARRRPCGP (difficult name ), I managed to achieve what I was looking for.

Initially, I was hoping to reach 5 Ghz fully stable with less voltage but sadly it turned out that my chip is in the lower end of average pool...

Yes I know that cache give little to no gain but since I'm quite methodical with my OC and I found out that there was no harm in increasing cache frequency after tuning my BIOS settings...why not?
Anyway, in my experience with this chip how I got the "right" Vcore mattered: we all know that every and each single chip is a different beast
Though I suspect that, even more that the CPU, my MB has a major role in whether settings with LLC 4 worked better than settings with LLC 3 or 2

 

[EarthDog]

Though I suspect that, even more that the CPU, my MB has a major role in whether settings with LLC 4 worked better than settings with LLC 3 or 2

Well, yes... it isn't the CPU that determines vdroop...its the motherboard and buck controller/MOSFETs/how they work. Some motherboards are able to 'hold' voltage better than others. Just depends on the power delivery setup and what you're trying to get out of the board with the CPU.1

That said, vdroop is actually part of the intel specification and 'normal'. The problem comes in with overclocking and the droop...hence LLC was born.

 

[RJARRRPCGP]

For the cases where the temp peak is reached sooner, I would be wondering about MOSFETs and the like. That's why I made sure I got a motherboard where they didn't knowingly go cheap on the MOSFET cooling.

 

[EarthDog]

For the cases where the temp peak is reached sooner, I would be wondering about MOSFETs and the like. That's why I made sure I got a motherboard where they didn't knowingly go cheap on the MOSFET cooling.

So long as the board isn't throttling, the MOSFETs have little to do with CPU temperature. Less efficient/capable MOSFETs will raise VRM temps, however. During motherboard testing using Hwinfo (to read onboard VRM sensor) and an Extech temp meter with probes attached to the VRMs, max temperature is easily reached within an hour there (assuming the same loads of course). Typically after around 30 mins (AIDA64 stress test, CPU/FPU/Cache checked).

 

[O.T.T.]

Well, yes... it isn't the CPU that determines vdroop...its the motherboard and buck controller/MOSFETs/how they work. Some motherboards are able to 'hold' voltage better than others. Just depends on the power delivery setup and what you're trying to get out of the board with the CPU.1

That said, vdroop is actually part of the intel specification and 'normal'. The problem comes in with overclocking and the droop...hence LLC was born.

Yes, I know: what I mean is that some CPUs have a better tolerance against Vdroop. Of course MB/MOSFETs play the main role in keeping a CPU stable, managing Vdroop.

 

[EarthDog]

I wouldn't call it a tolerance so much as it is normal variance between CPUs. Each CPU will need a different voltage to be stable at a given clockspeed. It's stable or it isn't at a given voltage... but this varies by chip due to silicon quality.

 

[RJARRRPCGP]

So long as the board isn't throttling, the MOSFETs have little to do with CPU temperature. Less efficient/capable MOSFETs will raise VRM temps, however. During motherboard testing using Hwinfo (to read onboard VRM sensor) and an Extech temp meter with probes attached to the VRMs, max temperature is easily reached within an hour there (assuming the same loads of course). Typically after around 30 mins (AIDA64 stress test, CPU/FPU/Cache checked).

I apologize, I was meaning the VRM temps.