What does 13900k instability look like

[smegle]

Hi all,
I’m currently tuning my 13900k on a rog strix z790e mobo.

I’m using the guide attached below:

Asus Maximus Z790 and Intel i9-13900k/14900k - An...

www.overclock.net

I initially got my ac_ll down to 0.17 however when it came to test with geekbench it just wouldn’t finish the test. R23 and Firestrike were fine.

I’ve settled on 0.24 now and will pass geekbench and have just increased OCTVB to +1Boost.

It’s not crashing, bsod’ing etc but my cores do move around a great deal on the clocks. Temps are 90ish degrees max after r23 completes. I’ve not had any crashes in games at all. 3D mark firestrike seems to vary between 49950 and 54000!?

Is this completely normal with these new gen chips? - after coming from an overclocked 9900k at 5ghz locked it’s just something I ain’t used too and wondered if it was lowering clock speeds due to instability?

Ta in advance!

 

[Freddie1980]

Doesn't geekbench put more stress on the ram then R23? You might want to loosen your timings or set the ram back to default speeds work from there.

 

[Deleted member 258511]

Hi all,
I’m currently tuning my 13900k on a rog strix z790e mobo.

I’m using the guide attached below:

Asus Maximus Z790 and Intel i9-13900k/14900k - An...

www.overclock.net

I initially got my ac_ll down to 0.17 however when it came to test with geekbench it just wouldn’t finish the test. R23 and Firestrike were fine.

I’ve settled on 0.24 now and will pass geekbench and have just increased OCTVB to +1Boost.

It’s not crashing, bsod’ing etc but my cores do move around a great deal on the clocks. Temps are 90ish degrees max after r23 completes. I’ve not had any crashes in games at all. 3D mark firestrike seems to vary between 49950 and 54000!?

Is this completely normal with these new gen chips? - after coming from an overclocked 9900k at 5ghz locked it’s just something I ain’t used too and wondered if it was lowering clock speeds due to instability?

Ta in advance!

Are you checking HWInfo for WHEA errors?

 

[Nickolp1974]

all depends how you have it set up, what are your P/e/c cores set too? are you using adaptive voltage? Running Geekbench checks the transient response of the PC, the change in clock frequency. Sometimes this can be remedied through setting an offset to the particular frequency in the V/F curve.
1st thing i always do is set to all core say X55, Set your mem to XMP or whatever, LLC4, X44 E cores 48/50 Cache and now find the lowest voltage stable for all your tests and write it down( you want the load voltage while the test is running)
Now keep all cores at X55 and switch to by core usage, D/C 1.02, A/C 0.20, LLC4, change to adaptive voltage, adjust A/C as before to get same stable voltage under load.
Then you can adjust core table something like 55,55,56,56,57,57,57,58, then run tests, if geekbench crashes and or 3dmark a positive offset to V/F points 8,9,10 may be required, try 0.040 and test, increasing by 10mv at a time till it passes.
If you find the frequency is downclocking at full load and you have cooling headroom you can introduce a temp offset in the TVB menu section, if you do this and a bench fails increase A/C till it passes as your allowing a higher temp threshold allowing the frequency to sustain higher and longer thus requiring more voltage. All about balance

 

[smegle]

Are you checking HWInfo for WHEA errors?

Yes mate - zero errors

 

[smegle]

Doesn't geekbench put more stress on the ram then R23? You might want to loosen your timings or set the ram back to default speeds work from there.

Ram was at stock when I’ve been testing

 

[smegle]

all depends how you have it set up, what are your P/e/c cores set too? are you using adaptive voltage? Running Geekbench checks the transient response of the PC, the change in clock frequency. Sometimes this can be remedied through setting an offset to the particular frequency in the V/F curve.
1st thing i always do is set to all core say X55, Set your mem to XMP or whatever, LLC4, X44 E cores 48/50 Cache and now find the lowest voltage stable for all your tests and write it down( you want the load voltage while the test is running)
Now keep all cores at X55 and switch to by core usage, D/C 1.02, A/C 0.20, LLC4, change to adaptive voltage, adjust A/C as before to get same stable voltage under load.
Then you can adjust core table something like 55,55,56,56,57,57,57,58, then run tests, if geekbench crashes and or 3dmark a positive offset to V/F points 8,9,10 may be required, try 0.040 and test, increasing by 10mv at a time till it passes.
If you find the frequency is downclocking at full load and you have cooling headroom you can introduce a temp offset in the TVB menu section, if you do this and a bench fails increase A/C till it passes as your allowing a higher temp threshold allowing the frequency to sustain higher and longer thus requiring more voltage. All about balance

Thanks for this. If I run r23 they are definitely down clocking the p cores to 5300/5400 at full load so I’m guessing something isn’t stable when I start out. I’ll try your method you mention above.
When you are reducing the initial voltage what setting do you lower? Is it the ac 0.2 as in the guide?

 

[Nickolp1974]

Thanks for this. If I run r23 they are definitely down clocking the p cores to 5300/5400 at full load so I’m guessing something isn’t stable when I start out. I’ll try your method you mention above.

Check temps while running R23 and LOAD voltage

When you are reducing the initial voltage what setting do you lower? Is it the ac 0.2 as in the guide?

yes lower A/C 0.02 at a time till it crashes then bump it back up 0.02 and you should be good to go. LLC4 must be set and adaptive voltage, the one below Global core voltage

what cooler are you using?

 

[OC2000]

A Lower AL_LL shouldn't affect geek bench, only all core stress like CB23. So get the AC_LL as low as you can 10 min CB23 stable then set an adaptive Global SVID to 1.48 and and work your way down until geek bench is unstable. If you are using an OCTV boost +1 or +2 profile or single boost higher than 58, you may need to alter the V/F curve 10 +0.040or higher as Nickolp says, which will then increase your target die sense for CB23, so work you AC_LL down more. If you hit 0.01 and your target die sense voltage for CB23 is still too high, use the octool which is linked on that guide to generate a multi point V/F curve adjustment which will lower the target die sense voltage further while providing high enough voltage for low load to keep geekbench 5 stable.

I ended up with the following:

LLC4 - 1.04/ 0.01 (1.04 because the VIDs then matched the Vcore. At 1.02 there was a 0.01 variance between the two)
Adaptive voltage 1.45
Cores (62,62,61,61,59,59,58,58) with OCTV boost +2
E Cores - 47x4, 46x8,45x16
This provided enough Low load Voltage to pass Geek bench, however running CB23 I got a die sense voltage of 1.24V I know 5.6 all core will do it at 1.22V so created a V/F curve profile using the OCTool to maintain the higher V for low loads and reduce the target die sense to 1.22V
V/F 8 +0.079
V/F 9 +0.049
V/F 10 +0.049

 

[smegle]

Check temps while running R23 and LOAD voltage

yes lower A/C 0.02 at a time till it crashes then bump it back up 0.02 and you should be good to go. LLC4 must be set and adaptive voltage, the one below Global core voltage

what cooler are you using?

Is a h115i capillex 280mm

 

[smegle]

A Lower AL_LL shouldn't affect geek bench, only all core stress like CB23. So get the AC_LL as low as you can 10 min CB23 stable then set an adaptive Global SVID to 1.48 and and work your way down until geek bench is unstable. If you are using an OCTV boost +1 or +2 profile or single boost higher than 58, you may need to alter the V/F curve 10 +0.040or higher as Nickolp says, which will then increase your target die sense for CB23, so work you AC_LL down more. If you hit 0.01 and your target die sense voltage for CB23 is still too high, use the octool which is linked on that guide to generate a multi point V/F curve adjustment which will lower the target die sense voltage further while providing high enough voltage for low load to keep geekbench 5 stable.

I ended up with the following:

LLC4 - 1.04/ 0.01 (1.04 because the VIDs then matched the Vcore. At 1.02 there was a 0.01 variance between the two)
Adaptive voltage 1.45
Cores (62,62,61,61,59,59,58,58) with OCTV boost +2
E Cores - 47x4, 46x8,45x16
This provided enough Low load Voltage to pass Geek bench, however running CB23 I got a die sense voltage of 1.24V I know 5.6 all core will do it at 1.22V so created a V/F curve profile using the OCTool to maintain the higher V for low loads and reduce the target die sense to 1.22V
V/F 8 +0.079
V/F 9 +0.049
V/F 10 +0.049

Cheers.
So just for my understanding, when under 100 percent load in cb23 for example, I should not see the core frequencies not under clock themselves?
So in your case above, under full load you never see a peformance core go under 5800mhz?

 

[OC2000]

Cheers.
So just for my understanding, when under 100 percent load in cb23 for example, I should not see the core frequencies not under clock themselves?
So in your case above, under full load you never see a peformance core go under 5800mhz?

Under load with OCTV boost +2, my core clocks drop to 5.6 which needs 1.22v die sense to work which is 270w with CB23. I don’t use prime 95 as it’s pointless for me

 

[Cyber-Mav]

You doing 6ghz on the P cores?

 

[smegle]

1st thing i always do is set to all core say X55, Set your mem to XMP or whatever, LLC4, X44 E cores 48/50 Cache and now find the lowest voltage stable for all your tests and write it down( you want the load voltage while the test is running)

Sorry for another question, so it’s the ac 0.2 you reduce for this first stage or another setting?

 

[smegle]

You doing 6ghz on the P cores?

It tries and on low load but immediately clocks them down

 

[Nickolp1974]

Sorry for another question, so it’s the ac 0.2 you reduce for this first stage or another setting?

once your set up to a by core setting and using adaptive voltage, then use A/C to tune the voltage, use 0.2 as a start point and work down, 1.18, 1.16 and so on till you lose stability and then just knock back up 0.02.

 

[Nickolp1974]

Here is a great guide, says for beginners but it goes into too much detail and can confuse, best to read and test as you go.

Asus Maximus Z790 and Intel i9-13900k/14900k - An...

www.overclock.net

 

[OC2000]

It tries and on low load but immediately clocks them down

Have you removed power limits?

In internal thermal management system (where you change the AC_LL and DC_LL) have you changed the values that say 253W to 4095W? You maybe limited to 253W which will down clock to keep you at that. Also make sure MCE is Disabled and set to Keep within Intel limits

 

[Nickolp1974]

Have you removed power limits?

In internal thermal management system (where you change the AC_LL and DC_LL) have you changed the values that say 253W to 4095W? You maybe limited to 253W which will down clock to keep you at that. Also make sure MCE is Disabled and set to Keep within Intel limits

Keep within limits!! Never!!!

@smegle it would help if you could post a screen of HWinfo after a run of R23

 

[smegle]

Think I found the issue. If I disable all limits, the p cores are rock solid at 5.5.
However my temps are huge and up to 100 degrees.
Could it just be I have a bad chip and my h115i 280mm just ain’t cutting the cooling?
This is with all stock, other than llc4 acll at 0.24 (it’s unstable below this) and dc at 1.02