Having trouble switching to Adaptive voltage mode on Z390 Maximus XI Formula + 9900k

[Zammin]

Hi everyone

I've just recently finished dialing in my stable 5Ghz overclock in manual mode with the following settings:

- AI Overclock Tuner XMP I (clicked Yes at the prompt)
- Sync all cores
- AVX Offset 0
- Manual Voltage Override 1.33V
- LLC7
- IA/DC Loadlines 0.01
- Loadline Synch diabled
- CPU Current Capability 170%
- Long and Short Duration Package Power Limit 4095
- Package Power Time Window 127
- CPU Core/Cache Current Limite Max 255.75
- Cache Ratios Auto
- VCCIO 1.15V
- System Agent Voltage Auto (1.2V)
- BCLCK Aware Adaptive Voltage disabled
- MCE disabled
- SVID disabled
- Intel SpeedShift/Step disabled
- C-States disabled

Now that I'm stable in manual mode I'm currently trying to switch to adaptive voltage mode. I've done this before on Z370 and my 8700k with success, however with the 9900k and my ASUS Z390 motherboard it is exhibiting some strange behavior.

In manual mode it sits at 1.314V idle and drops to 1.27V under RealBench (or AIDA64) types of loads. When I re-enable C SpeedShift/Step, C-States and SVID, then switch to Adaptive voltage and set 1.33V as the additional turbo voltage it idles between 1.33-1.35V and sits around 1.33V under load, causing it to run much hotter than before. The only way I can get it to come down to 1.33V idle and sag to 1.27V under load again is to set LLC to level 6. This is all very strange to me as I would expect the LLC behavior as well as the idle voltage and load voltages to be the same as before until I set the windows power plan to balanced to allow the CPU to clock down when at idle.

This is not in line with my experience overclocking my 8700k on my ASUS Z370 Maximus X Code. In this case on the Z390 motherboard LLC6 in adaptive mode behaves closer to the way LLC7 behaves in manual mode, although the voltage level is not as stable under load, it fluctuates up and down a lot instead of sitting at 1.27V for the duration of the test. This appears to cause more heat and definitely isn't stable as I get BSOD's in RealBench with these settings in adaptive mode.

I'm not really sure what to do, I want to use adaptive mode to allow my CPU to clock down (and bring the voltage down) at idle but I just can't get it to work. I had no trouble doing it on Z370 following Raja's Z270 overclocking guide but it just doesn't want to work on Z390 for some reason.

If anyone has any advice for me I would greatly appreciate it. I've listed my settings above but if anyone requires BIOS screenshots I can get them for you as well.

[Silent_Scone]

Who knows. With your climate over there, the temperature will be a big factor. This and differences in UEFI config. For instance, are you running an AVX offset now but perhaps weren't before?

[Zammin]

Silent Scone@ASUS wrote:
Who knows. With your climate over there, the temperature will be a big factor. This and differences in UEFI config. For instance, are you running an AVX offset now but perhaps weren't before?

Even before the peak core temp was 92C. Today it was around 95C. I'm not using an AVX offset at the moment and I wasn't before either, no idea what is going on.

[Zammin]

Ok so I've started testing with Shamino's suggestion, I've set adaptive voltage mode with the additional turbo voltage on auto, set LLC6, set IA AC/DC LoadLines to 0.03, disabled SpeedStep and enabled SpeedShift. Although for the first test I've aimed a bit lower just to get a baseline so I've set the multiplier to 49. Seems to be working okay, ran a 30 min RealBench stress test and passed with a peak core temp of 89C but the cores were usually around 78-82C. Voltage readout on the LiveDash and iCUE shows 1.23V-1.24V during the test. Water Temp peaked around 36C, ambient 24C.

I will try 5Ghz but I'm pretty sure it will fail at this voltage. My question is, if I find it to be unstable, do I increase the number for IA AC/DC LoadLines? For example, go to 0.04, then 0.05 etc etc. Or do I go down to 0.02? I'm assuming I want to increase the number. Sorry for not fully understanding this feature.

Also my apologies JPMboy, I said that C-states was only an option to enable or disable them but I found what you were talking about when you mentioned "enable C-States up to C6". When I enable them rather than selecting auto, it lists all the individual C-states. Should I still disable Cstates above C6 and keep C1-C6 enabled? I've just got it on auto for now.

Thanks everyone I really appreciate the help and advice you've given me so far. I'm learning as I go but I clearly still have a long way to go..

[Zammin]

Ok so I tried to get 5Ghz a few times and it's just not working out.

I tried IA AC/DC LoadLines at 0.03 - crash within 10 mins, 0.04 - crash around 15mins, 0.05 crash after around 35mins. Voltage in the final test was 1.27-1.28 under load. Peak CPU core temp was 97C, average 88-92C. Water temp peaked at 38C. Ambient 25C. It's just getting too hot in RealBench now, I don't think I can keep pushing it further. If I hadn't used liquid metal between the die and IHS I may have hit 100C during that last test, since the liquid metal shaved between 4-6C off the CPU cores during heavy loads.

Unless there is some other tricks I can try I may just have to settle for 4.9Ghz or use an AVX offset of -1. It's a shame, many others seem to have no trouble reaching 5Ghz so I guess my CPU isn't a great one.

[Silent_Scone]

Please remember that ultimately, not all samples will achieve 5GHz on all cores. The statistics from SiliconLottery.com indicate less than half. You’re hitting the thermal capacity of your setup, too. At this point, my advice would be to tune the system for the tasks you intend to use it for.

[Zammin]

Silent Scone@ASUS wrote:
Please remember that ultimately, not all samples will achieve 5GHz on all cores. The statistics from SiliconLottery.com indicate less than half. You’re hitting the thermal capacity of your setup, too. At this point, my advice would be to tune the system for the tasks you intend to use it for.

Yeah I think I will try and tune it for 4.9Ghz as best as I can and leave it there, or use an AVX offset. Better luck next time I guess 😕

[Falkentyne]

Zammin wrote:
Yeah I think I will try and tune it for 4.9Ghz as best as I can and leave it there, or use an AVX offset. Better luck next time I guess 😕

When you said you first passed 5 ghz at the same exact settings you are now failing, were you using a different bios version or any different RAM settings at all?
Because while it is common for new CPU's to lose about 10mv off their minimum voltage after a short while even when not pushed to unsafe limits, they will usually remain quite stable, but yours seems to have gone for 1+ hour stable to insta-crashing, which is pretty unheard of without excessive voltage. And a 1 hour stable test means you were probably not at your bare VMIN.

What was the absolute highest load (or idle, if using manual) voltage you used?
And are you sure you were not on a different bios?

Do you have the ability to (if you were), downgrade to the same bios you did that stable test on and then test on that?

Because I found out something interesting, although it's still not proven directly but I can't be too far off...(still not jumping to conclusions yet until I can be 100% sure), but I'll just say possibly "Subtiming".

Did you test with non avx tests to verify stability or AVX tests that do NOT touch main memory, as well as your mixed AVX tests? (example: cinebench r15, prime95 29.6 build 2 with AVX/AVX2 disabled, smallest FFT's, etc?)

[Zammin]

Falkentyne wrote:
When you said you first passed 5 ghz at the same exact settings you are now failing, were you using a different bios version or any different RAM settings at all?
Because while it is common for new CPU's to lose about 10mv off their minimum voltage after a short while even when not pushed to unsafe limits, they will usually remain quite stable, but yours seems to have gone for 1+ hour stable to insta-crashing, which is pretty unheard of without excessive voltage. And a 1 hour stable test means you were probably not at your bare VMIN.

Nah I'm not using a different BIOS, I even had the profile saved so I could reload it exactly as it was. RAM has remained on XMP I the entire time. I've had this system for around 4 months now but didn't start overclocking the CPU until I got it under water because it was already at it's limit on air.

Falkentyne wrote:
What was the absolute highest load (or idle, if using manual) voltage you used?
And are you sure you were not on a different bios?

Highest on my previous manual mode 5Ghz OC? 1.314V highest idle, 1.270V steady under AVX load (Realbench).

Falkentyne wrote:
Do you have the ability to (if you were), downgrade to the same bios you did that stable test on and then test on that?

Still on the same BIOS.

Falkentyne wrote:
Because I found out something interesting, although it's still not proven directly but I can't be too far off...(still not jumping to conclusions yet until I can be 100% sure), but I'll just say possibly "Subtiming".

Do you mean RAM subtimings? I'm just using XMP I, so only the basic settings have been loaded (3200Mhz, 16-18-18-38, 2N, 1.35V) with VCCIO turned down to 1.15V because it was setting it to 1.314V for some reason, and from what I understand you don't want it above 1.3V.

Falkentyne wrote:
Did you test with non avx tests to verify stability or AVX tests that do NOT touch main memory, as well as your mixed AVX tests? (example: cinebench r15, prime95 29.6 build 2 with AVX/AVX2 disabled, smallest FFT's, etc?)

It was only the other day so I ran a couple of 1 hour RealBench passes, a little bit of AIDA64 and an hour of OCCT Large Data Set as well as a few Cinebench runs and some gaming.

[Falkentyne]

Zammin wrote:
Nah I'm not using a different BIOS, I even had the profile saved so I could reload it exactly as it was. RAM has remained on XMP I the entire time. I've had this system for around 4 months now but didn't start overclocking the CPU until I got it under water because it was already at it's limit on air.



Highest on my previous manual mode 5Ghz OC? 1.314V highest idle, 1.270V steady under AVX load (Realbench).



Still on the same BIOS.



Do you mean RAM subtimings? I'm just using XMP I, so only the basic settings have been loaded (3200Mhz, 16-18-18-38, 2N, 1.35V) with VCCIO turned down to 1.15V because it was setting it to 1.314V for some reason, and from what I understand you don't want it above 1.3V.



It was only the other day so I ran a couple of 1 hour RealBench passes, a little bit of AIDA64 and an hour of OCCT Large Data Set as well as a few Cinebench runs and some gaming.

Yes I meant RAM subtimings, because I just flashed to beta bios F8H on the Aorus Master, and I noticed my 1344K in place fixed FFT (AVX) at my 2 hour rock stable settigs, were now not only crashing a thread (usually the first thread to crash, thread 6 or 7), BUT were also causing CPU L0 errors (WHEA correctable) too!
The 1.275v, 5 ghz + LLC Turbo test which had been stable 1-2 hours was now crashing/generating L0 errors in 5-15 minutes. So I decided to test my stable 5.1 ghz + 1.345v + LLC Turbo with 1344K AVX also, and it crashed in less than 6 minutes! Ran two tests at 5 ghz and one at 5.1 ghz to make sure.

So I flashed back to beta F8G and used the exact same settings precisely.
and guess what?
1.275v 5 ghz LLC Turbo passed 1 hour 10min at 1344K in place fixed AVX FFT's till I stopped it.
Then 1.345v 5.1 ghz LLC turbo 1344K AVX passsed 1hour 15 min (stopped it)
Then went back to 5 ghz 1.275v again and again 1 hour 5 min pass and stopped it.

Exact same VR VOUT (cpu on die sense) voltage, watts, power draw as before. The only change was the bios between the unstable and stable tests.
Oddly enough, on both bioses, there was no problem with 15K in place fixed AVX FFT at 5 ghz, 1.285v (LLC turbo) with hyperthreading disabled (so, 8 threads only) for 1 hour+, which I long tested as my VMIN for this. Since 15K runs entirely in the cache (i assume), whatever was affecting the above didn't affect this.

On one of the comprehensive memory guides on overclock.net, someone mentioned that one of the low subtimings affected AVX speed greatly.

Again I have no absolute proof as I would still need to have "VMINS" for 1344K AVX with hyperthreading disabled (I don't), 1344K Non-AVX with hyperthreading enabled, to determine if it's RAM access with AVX or hyperthreading itself with accessing RAM. And all of that crap takes WAY too long to test, since you need to find the min stable voltage for each test.

And then on top of that I would have to then test if increasing VCCIO or SA or something would help 😕

But anyway you said you are on the exact same bios as before, so I just wasted my time writing all of this.
(tl;dr: in my case, new beta bios must have changed a RAM subtiming that may have affected AVX with RAM access, hurting stability)

[Flokke]

So, this seemed to switch from the topic of trying to get Adaptive voltage to work properly more so to purposes of getting a stable overclock at 5ghz. I'm also having the same issues with getting my CPU to downvolt when at idle running like an 800mhz clock speed. I have it steady at 4.9ghz all core at 1.279v but no matter what I change with adaptive voltage I can't get it to reduce bellow that 1.279v it just sits there. Granted this I suppose is first world problems at this point.

Maybe I will switch back to 605 as I'm almost certain that was the bios i had before upgrading just now and everything was working like it should then. I just figured the bios updates would be good to have as there seems to have been a lot of improvement since then. If someone has this current bios 0805 and has their CPU dropping the Vcore down like it's should at idle multipliers please post your bios I'd love to take a look.