Question How much should I undervolt?/Help for undervolting 13700K on Asus Board/low clock speed on multicore

[dashshad]

Hello guys, i want to undervolt my 13700k .

Short Question :How much should I undervolt? and how can i test my undervolt is stable

-------

mobo:tuf gaming z790 pro lastest update 0x12B Microcode

i see this guide of Buildzoid :

"Testing and Optimizing the 0x129 Microcode on the ASUS Maximus Z790 Apex Encore."

View: https://www.youtube.com/watch?v=XI2x2_skwSs


so here is my changes in Intel Default Settings(PL1/PL2=253W, 307 A)

IA VR Limit: 1400 for now( maybe 1300 later)

AC/DC Synch = Enabled

CPU LLC = LEVEL 6 (0.49 milliohms)

Tj Max = 85 C

Undervolting : Global Core SVID Voltage = adaptive mode = - 0.09000 (-90mV)

i run a prime95 Small FFTs test for a minute here is my HW info Values :

Vcore (under Load) =1.154

VID(under Load)= 1.162

Max package Temp : 88 C

Vcore max =1.252

VR VCC Current Max = 217.152 A

VID max = 1.270

Max Power Draw = 253W

and i run few loops of CB R23/R24/R20/R15 scores and temp is good (cuz CBR23 draw 220~230W) and no performance lose.(a little question here : my Core Clocks under load in CB23 Multicore go to 4750Mhz. why?! it must be 5300~5400Mhz)

so here is my main question /confusion:

#1 how much can i undervolt ?(Vcore (under Load) =1.154 is it safe/normal???!)

it seems i can undervolt till -110mV without problem but is Vcore (under Load) =1.154 considered as safe (tooo low maybe??)??????

#2 how can i stability test properly?

=== edited====

yes i know i must undervolt till edge of stability and add 5~10 mV to the value

=== edited #2====

I ran the Ycruncher pi 2.5 test. passed

SPs:

SP =88

P core SP = 98

E core SP= 68

 

[Roland Of Gilead]

Hey there,

So, why do you want to undervolt. What's the goal?

Prob easiest way, is to do a negative vcore offset. As you know, decrease in small increments and test at each setting. Cinebench R23 is a good one for an indication of stability. If you wanted to be absolutely sure you could use Prime95 on small FFts to really load up the cores with max temp. That will also test how good your cooler is.

There are many other benches. OCCT/AIDA 64 among others. Try them all.

 

[geofelt]

What is your objective?
Chips are binned so overclocking is not expected to yield much benefit.
Since you are at the latest motherboard bios level, your chip is safe when using the motherboard default.
13700K is plenty fast at stock default settings.

One runs the risk of screwing things up by messing with voltages.

 

[dashshad]

Hey there,

So, why do you want to undervolt. What's the goal?

Prob easiest way, is to do a negative vcore offset. As you know, decrease in small increments and test at each setting. Cinebench R23 is a good one for an indication of stability. If you wanted to be absolutely sure you could use Prime95 on small FFts to really load up the cores with max temp. That will also test how good your cooler is.

There are many other benches. OCCT/AIDA 64 among others. Try them all.

Hey, thx for the answer,
The goal is to have the lowest possible voltage(without instability) so that the processor does not degrade.
Did you read my text completely? Do you think the voltage I got under the load is good? Or is it too little?

Cinebench R23 is a good one for an indication of stability. If you wanted to be absolutely sure you could use Prime95 on small FFts to really load up the cores with max temp.

How long should the test be? 1 min ? 10min? 30 min?

There are many other benches. OCCT/AIDA 64 among others. Try them all.

My question is that with what settings should I run these programs to test the CPU?
.

 

[dashshad]

What is your objective?
Chips are binned so overclocking is not expected to yield much benefit.
Since you are at the latest motherboard bios level, your chip is safe when using the motherboard default.
13700K is plenty fast at stock default settings.

One runs the risk of screwing things up by messing with voltages.

Hey, thx for the answer,
The goal is to have the lowest possible voltage(without instability) so that the processor does not degrade.

One runs the risk of screwing things up by messing with voltages.

Well, do you think that the amount of voltages that I wrote in the original post is too low?

 

[Roland Of Gilead]

Did you read my text completely? Do you think the voltage I got under the load is good? Or is it too little?

Yes. The voltage seems just fine. There is a bit of vdroop there. So you might need to increase LLC (Load Line Calibration) up a setting or perhaps 2 settings. Test in between to ensure stability.

How long should the test be? 1 min ? 10min? 30 min?

For Cinebench, doing just one run (min duration off) will at least show you your clock speeds at a given/repeatable load and basic stability. So, from run to run, if your scores increase, and your are stable without crashing, then you have a good place to start to test for stability. You can of course run a 10 min or 30 min test, which will give you a further indication of stability at your chosen settings. However, Cinebench whilst decent, does not really push your CPU to the max, where Prime and the other benches I highlighted does.

My question is that with what settings should I run these programs to test the CPU?

Prime95 on small FFTs only. This will test only the CPU and cache. This is where you find out for true stability.,Typically you might run Prime overnight for 8 hours,. But, you need to have good cooling to run these benchmarks.

As for the other questions about what is good or bad. Only you can determine this. Your CPU is unique. Settings that work on one CPU may not work at all on yours. So you have to do a little work with testing to get your desired results.

But, with that said your temps, voltages seem normal to me, and I don't see the reason to tinker further.

 

[dashshad]

Yes. The voltage seems just fine. There is a bit of vdroop there. So you might need to increase LLC (Load Line Calibration) up a setting or perhaps 2 settings. Test in between to ensure stability.

For Cinebench, doing just one run (min duration off) will at least show you your clock speeds at a given/repeatable load and basic stability. So, from run to run, if your scores increase, and your are stable without crashing, then you have a good place to start to test for stability. You can of course run a 10 min or 30 min test, which will give you a further indication of stability at your chosen settings. However, Cinebench whilst decent, does not really push your CPU to the max, where Prime and the other benches I highlighted does.

Prime95 on small FFTs only. This will test only the CPU and cache. This is where you find out for true stability.,Typically you might run Prime overnight for 8 hours,. But, you need to have good cooling to run these benchmarks.

As for the other questions about what is good or bad. Only you can determine this. Your CPU is unique. Settings that work on one CPU may not work at all on yours. So you have to do a little work with testing to get your desired results.

But, with that said your temps, voltages seem normal to me, and I don't see the reason to tinker further.

thx for the answer.
what about my frequency question.
a little question here : my Core Clocks under load in CB23 Multicore go to 4750Mhz. why?! it must be 5300~5400Mhz)

 

[fuuster]

thx for the answer.
what about my frequency question.
a little question here : my Core Clocks under load in CB23 Multicore go to 4750Mhz. why?! it must be 5300~5400Mhz)

When a single core is used, cpu will boost to its maximum speed. As more cores are used, the clock speed steps down (this is performed by the cpu itself). I think at 253W, all cores running at 4750Mhz is about right.

Please read this article for reference (Hit Ctrl-F and search "Firstly" if you wish to skip straight to the point mentioned above):
https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

If you wish all cores to run at higher clock speed, you will need to set higher PL2, and your cpu cooling must be powerful enough to keep core temperatures low enough to allow algorithm to set higher clock speeds. The performance gain is not worth it, because cpu will likely be drawing in excess of 300W. You will essentially be overclocking your cpu, and potentially shortening its little life.

 

[TheHerald]

1) You can undervolt as much as your CPU is capable off. There is no danger. Youll just end up being unstable

2) If your clockspeeds drop it's one of two things, either you have a power limit on it, or CEP kicks in. Check in your BIOS for the CEP option and disable it, although supposedly the latest BIOSes don't give you the option anymore. Also make sure you are looking at the Pcore clockspeeds, if you are looking at the averages then it makes sense that it's at 4750 since it takes Ecores into consideration as well.

3) There is no reason to have an IAVR limit when you are undervolting. If for whatever reason you want it to be there, set it to 1.45v

4) Your AC/DC LLs aren't configured properly, that's why there is a missmatch between VID and Vcore under load. That's not a big deal I guess but your power draw numbers will be off because of that.

5) For some fast stability tests, run CBR15 - CBR20 - Ycruncher. A couple of runs each. It won't tell you you are 100% stable, but it will tell you whether you are too far off from stability.


FYI, I've been running my 12900k at 1.06 volts under load, but of course my clockspeeds are lower than yours (4.9 P cores). Power draw was around 160w in CBR23. Sadly can't do that anymore since I majorly degraded my chip by running 1.64 volts for some benchmarking, now I need 1.12