Optimal settings and issue for PRO Z690-A DDR4 + 13600KF

[alex20079x154d02d6]

Hello friends, long story short I've saw the instability and all the Intel issues in the last few weeks and decided to act.
First of all apparently from the get go I had "Tower air cooler" which made my PL1/2 on 288W and 512A IccMax, also had lite load 7.
I never experienced any issues but decided to run Cinebench R23, I got around 24k on multicore with all Pcores being on 5.1Ghz, but I saw that my core temps are hovering 100c,
So I decided to follow the guidelines posted here and First of all change from "Tower air cooler" ---> "Boxed air cooler" which lowered my PL1/2 to 181W and IccMax to 194A,
I also sat lite load = 5 and I disabled IA CEP/Turbo Enhance and ran Cinebench again.

This time my temps were completely fine (hovering around 80c) BUT my score was much lower for multicore, around 22k and I've noticed that my Pcores no longer reach 5.1Ghz, and my power draw never reaches 180W (added pictures of HWInfo64 while running R23 multicore).

What is happening? Why did I get such a decrease in performance? Would appreciate some help!

 

[citay]

So I decided to follow the guidelines posted here and First of all change from "Tower air cooler" ---> "Boxed air cooler" which lowered my PL1/2 to 181W and IccMax to 194A,

This is not really in the guide (at least not in mine).
I highly recommend to determine good values yourself, which apply to your specific situation. This is covered by the whole step 1) of my guide.

First of all, you should probably update the BIOS to the latest beta version from https://www.msi.com/Motherboard/PRO-Z690-A-DDR4/support
Currently, your board does not have the BIOS with the latest 0x129 microcode yet, but at least the 0x125 one, and that BIOS should also come with a revised cooler selection prompt. The topmost option should no longer be called "Boxed cooler", it should be "Intel Default Settings" for example. But even then, it's not the ideal choice. I will explain.

When you select one of those presets, they are not specific to your cooling whatsoever. They might be overly low or overly high, depending on the preset. Neither Intel nor MSI have any idea how much heat your cooling can deal with, so the numbers are more or less arbitrary for that purpose. But when you have thermal throttling like you had with the 288W/288W/512A preset, what is the main goal? To bring the temperatures to a safe level, without sacrificing too much performance in the process. How do you want to achieve this when there are only three presets available, the bottom two already being way too high for your cooling?

We can also deduct why it doesn't hit the 181W power limits: The IccMax (CPU Current Limit) must simply act before the power limits can ever step in. There is no power-limit-throttling whatsoever, so the thing doing all the limiting here is the CPU Current Limit of 194A. That's why i also like to keep IccMax a bit more out of it (maybe setting it to 307A is a good idea) and purely work with the power limits when it comes to the cooling. The CPU Current Limit works with internal CPU currents and is not easy to adjust properly, plus sometimes it is overzealous. You can see it essentially caps things at around 150W power draw for you.

Now, it would be important to know what cooler you have, and to see all the sensors at once, like i describe in my guide (having three columns of sensors side-by-side). The CPU fan speeds under full load seem quite low at 1300 RPM, and you use passive cooling in idle (perhaps inadvertently), so check my Fan Curves guide to set it up properly.

While performance is suffering at ~150W, your cooler is currently at slightly over 80°C there, so it's not like you have crazy high cooling reserves. But a bit more than this is doable, especially when you have not yet have good fan curves set. We should be able to get into the ballpark with the Cinebench scores again. But also, at 22K, a couple hundred points is almost nothing if you convert it into a percentage, so we shouldn't hunt for exact scores here.

Lastly, about CPU Lite Load, you don't just lower it to a random number, you have to test for stability there. If it's stable, you might be able to lower it even more. This will always improve the score when you are dealing with any kind of limits like power limits, which by the looks of it are absolutely inevitable to set for your cooler.

 

[alex20079x154d02d6]

This is not really in the guide (at least not in mine).
I highly recommend to determine good values yourself, which apply to your specific situation. This is covered by the whole step 1) of my guide.

First of all, you should probably update the BIOS to the latest beta version from https://www.msi.com/Motherboard/PRO-Z690-A-DDR4/support
Currently, your board does not have the BIOS with the latest 0x129 microcode yet, but at least the 0x125 one, and that BIOS should also come with a revised cooler selection prompt. The topmost option should no longer be called "Boxed cooler", it should be "Intel Default Settings" for example. But even then, it's not the ideal choice. I will explain.

When you select one of those presets, they are not specific to your cooling whatsoever. They might be overly low or overly high, depending on the preset. Neither Intel nor MSI have any idea how much heat your cooling can deal with, so the numbers are more or less arbitrary for that purpose. But when you have thermal throttling like you had with the 288W/288W/512A preset, what is the main goal? To bring the temperatures to a safe level, without sacrificing too much performance in the process. How do you want to achieve this when there are only three presets available, the bottom two already being way too high for your cooling?

We can also deduct why it doesn't hit the 181W power limits: The IccMax (CPU Current Limit) must simply act before the power limits can ever step in. There is no power-limit-throttling whatsoever, so the thing doing all the limiting here is the CPU Current Limit of 194A. That's why i also like to keep IccMax out of it and purely work with the power limits when it comes to the cooling. The CPU Current Limit works with internal CPU currents and is not easy to adjust properly, plus sometimes it is overzealous. You can see it essentially caps things at around 150W power draw for you.

Now, it would be important to know what cooler you have, and to see all the sensors at once, like i describe in my guide (having three columns of sensors side-by-side). The CPU fan speeds under full load seem quite low at 1300 RPM, and you use passive cooling in idle (perhaps inadvertently), so check my Fan Curves guide to set it up properly.

While performance is suffering at ~150W, your cooler is currently at slightly over 80°C there, so it's not like you have crazy high cooling reserves. But a bit more than this is doable, especially when you have not yet have good fan curves set. We should be able to get into the ballpark with the Cinebench scores again. But also, at 22K, a couple hundred points is almost nothing if you convert it into a percentage, so we shouldn't hunt for exact scores here.

Lastly, about CPU Lite Load, you don't just lower it to a random number, you have to test for stability there. If it's stable, you might be able to lower it even more. This will always improve the score when you are dealing with any kind of limits like power limits, which by the looks of it are absolutely inevitable to set for your cooler.

Amazing! Thank you, I will definitely return to this comment in 3-4 hours after I'm done with some University work
My cooler is NH-D15S, and my case is Corsair 4000D Airflow TG with 2 extra PWM Arctic fans added (one front sucking air in, and one back pulling it out), so I assume my cooling is not too bad, but probably needs a fan curve adjustment with your added guide (thank you!).

Right now as I see it I need to do these steps in this order:
1. Update my bios to the latest beta, even though I will have to update it again in a couple of weeks when the new version is available - but that's ok!
2. Remember how to set up XMP again, because I assume the bios update resets all the settings?
3. Use the fan curve guide and to set up fans.
4. Test which lite load is appropriate for my CPU, lowering it until I find a stable number (I ran R23 with LL=3 already and it seemed ok).
5. Pick a default cooling option for PL1/2 and IccMax, and then experiment with the values until there's a decent temperature/no throttling

Only step 5 confuses me a little since I'm not sure how to play with the PL1/2 And IccMax, should I start by upping the IccMax until steady state in multicore benchmark? Then try and move the PL1/2 up? Even though Intel stated that the recommended options are PL1/2=181W, IccMax=200A.

Thank you so much!

 

[citay]

You're welcome.

My cooler is NH-D15S

It should do better than this for sure. Granted, its fan only goes to 1500 RPM. But we'll see once it's properly set up. I have the NH-D15 (with two fans) and it's definitely good for more than 200W, this one shouldn't be that far behind.

As for the steps:
1. Pretty much.
2. Yes, XMP back to Enabled. BIOS update resets everything but the fan curves.
3. Yes.
4. Yes, only that CB23 is not a real stability test, just a benchmark. Run all sorts of things, you can run Cinebench, Geekbench, and of course real stress-tests like OCCT. Any crash, instability or weirdness going on and your mode is probably too low.
5. Set IccMax to 307A so it doesn't interfere, then work only with the power limits. The goal is to stay in the 80°C range for the whole Cinebench Multi run, low 90s is the most you should accept (unless it's hot in the room because of summer, then the power limits will probaby be needlessly low for cooler room temperatures). But even in the summer you want to avoid thermal throttling.

Intel recommended settings don't have much to do with anyone's cooling, just like the MSI presets didn't. Or do they think an average tower cooler can get rid of 288W indefinitely? Maybe for a couple seconds at most. That kind of preset is firmly in AIO water cooler territory, if your CPU really wants to draw that much power.

The IccMax is not relevant if you set decent power limits, the power limits will already take care that things don't get out of hand, IccMax then can only get in the way, like it did on the Boxed preset.

 

[Raxiel]

I have the Z790-P WIFI DDR4 and the 13900kf.
When I imposed the 181W power limits and 200A ICC max it was fine, but when I also lowered lite load, it would very quickly hit "RINGL Max VR Voltage, ICCmax, PL4" and throttle back below the 5.1/3.9GHz defaults.
Raising Lite load or bumping ICCMax to 210A seemed to clear it up.

 

[citay]

Whenever you have significantly lower performance (sometimes halved) after you lowered CPU Lite Load, and when it has nothing to do with the power limits or is too significant to be explained by that, then you normally just need to disable "IA CEP Support", because that is needlessly intervening. Since he already has that disabled, and the performance only dropped slightly (24K -> 22K points), it can be completely explained from the limits keeping everything to ~150W of power draw. Of course, if the CPU is forced to "make do" with that lower power/current budget, it will not quite extract its full performance. But i'm sure we can increase it enough again.

As i said, for the IccMax, i would probably keep it at 307A for whatever CPU model, or if you really want to eke out all the performance of an i9 and have good enough cooling that this becomes a limitation, 400A at the very most. Then just do any limiting you have to do because of your cooling with the power limits. No need to raise CPU Lite Load if the lower mode was stable, as long as IA CEP is not interfering (and then you'd simply disable that, not raise the mode).

 

[Raxiel]

Fair enough, It wasn't a significant drop with me either (IACEP is on 'auto' which is apparently disabled). I haven't tried tuning under these new limits much, I'm waiting until the 0x129 version before I spend any time on it and it was a quick fix without straying from the 'defaults' too much until then.

 

[alex20079x154d02d6]

Whenever you have significantly lower performance (sometimes halved) after you lowered CPU Lite Load, and when it has nothing to do with the power limits or is too significant to be explained by that, then you normally just need to disable "IA CEP Support", because that is needlessly intervening. Since he already has that disabled, and the performance only dropped slightly (24K -> 22K points), it can be completely explained from the limits keeping everything to ~150W of power draw. Of course, if the CPU is forced to "make do" with that lower power/current budget, it will not quite extract its full performance. But i'm sure we can increase it enough again.

As i said, for the IccMax, i would probably keep it at 307A for whatever CPU model, or if you really want to eke out all the performance of an i9 and have good enough cooling that this becomes a limitation, 400A at the very most. Then just do any limiting you have to do because of your cooling with the power limits. No need to raise CPU Lite Load if the lower mode was stable, as long as IA CEP is not interfering (and then you'd simply disable that, not raise the mode).

Thank you man, I came with some answers after I haven't been here for a couple of days! (University exams haha).
I did all the steps mentioned above, I updated my bios to latest (0x125 as we speak), changed my fan curves (THANK you btw! I have a few DC fans and a few PWM fans, and they were all set to DC. So I adjusted the curves similar to your guide and the pc is now both more quiet and runs cooler I think), ofc XMP working as intended and all.

BUT, I haven't been able to play with ONLY the Lite load. Because even at Lite load 3 (which is rather low) I would still get close to 100c if I let my IccMax be 300A+.
I tinkered and played with the settings, seeing that my Core VIDs and my Vcore should be +- the same margin (+- 0.02v) so I ended up placing a manual Lite load with:
LL_AC on 0.15 and LL_DC on 0.9, with IccMax current limit of 250A and PL1/2 with 181W as before - Which seems to stabilize my cores on 5Ghz on the Pcores and 3.9Ghz on Ecores with 23.2k on Cinebench and never above +- 90c, which seemed stable enough and fine since I'm mostly gaming and would not have my cores working in multi core mode like that.

I still haven't run a stress test, because my lack of time, but my PC never crashed after 3-4 consequative runes of R23.
I include a picture during the end of one of the runs, I'm still curious as to why I cannot reach the 5.1Ghz with decent temps with comparably low Lite load - but for now it seems fine enough (maybe until 0x129, who knows haha).

Curious for your opinion tho!



EDIT: I just saw that 0x129 is available, any insights from it? Should I change anything?

 

[citay]

They seem to be turning the BIOS with the 0x129 microcode into a stable version for some boards already, so maybe wait a couple days for that, it's not like you have a 14900K. You can reuse most settings like the power and current limits, the fan curves should be preserved, the only question mark is about CPU Lite Load, there's reports it may behave a bit differently under that new BIOS. So this will at least have to be retested, if not readjusted.

About the cooler, i still expected a bit better performance from it, but maybe i just underestimate what difference a second fan on the cooler makes. And maybe it's also a bit hotter than normal in your room. I guess then it's alright. Cinebench performance seems to be in the ballpark too, so no concern there.

 

[alex20079x154d02d6]

They seem to be turning the BIOS with the 0x129 microcode into a stable version for some boards already, so maybe wait a couple days for that, it's not like you have a 14900K. You can reuse most settings like the power and current limits, the fan curves should be preserved, the only question mark is about CPU Lite Load, there's reports it may behave a bit differently under that new BIOS. So this will at least have to be retested, if not readjusted.

About the cooler, i still expected a bit better performance from it, but maybe i just underestimate what difference a second fan on the cooler makes. And maybe it's also a bit hotter than normal in your room. I guess then it's alright. Cinebench performance seems to be in the ballpark too, so no concern there.

Will do!
I'll wait for the stable version, even though the guys at MSI are funny since the 0x125 for my mobo seems to be still beta even after 0x129 beta release

According to my screenshot, is everything seems to be in the ball park when we speak about the voltages? I saw some fellas in other forums suggesting to try and undervolt manually (with offset), but since my lack of experience I didn't even wanna try this route. It's also interesting how much of a role the IccMax plays in my settings. Lowering it even by 15-20A changes the temp by 7-8c+- and throttles my pCores back to 4.9Ghz.

 

[citay]

1.35V maximum VID request and 1.34V maximum VCore is not really concerning, on a 14900K you can see 1.5V and more, just check these screenshots (and that's already with 253W power limits, not sure of IccMax there). Of course, the lower, the better. Speaking of IccMax, it works with internal CPU currents, i think it can regulate things much faster than the power limits, that's why it's a bit difficult to fine-tune if you want some specific behaviour.

 

[Raxiel]

They aren't alarming, but I'd still say those are rather high for an i5, although it could be measurement error with LL_DC set so much higher than AC. That seems plausible based on the package power. Was the screenshot taken while the system was under an all core load? The Cinebench score seems about right for stock speeds.

I've also found the 200A ICCMAX Intel recommends, limiting but only just. With only a small Lite Load tune 210A mostly got rid of current limit warnings, and 220A got rid of them entirely. I had to fiddle a bit to push the voltage down to around 1.155v before it would run at stock speeds within the 200A limit.

 

[alex20079x154d02d6]

They aren't alarming, but I'd still say those are rather high for an i5, although it could be measurement error with LL_DC set so much higher than AC. That seems plausible based on the package power. Was the screenshot taken while the system was under an all core load? The Cinebench score seems about right for stock speeds.

I've also found the 200A ICCMAX Intel recommends, limiting but only just. With only a small Lite Load tune 210A mostly got rid of current limit warnings, and 220A got rid of them entirely. I had to fiddle a bit to push the voltage down to around 1.155v before it would run at stock speeds within the 200A limit.

Yes it was taken during R23 all core (multi core) full load.
I will wait a couple more days until the 0x129 stable version (non beta) and will try to run some tests with lower Vcore (I would assume I do it by setting a more stable Lite load version, like 5? without doing any manual change to LL_AC/DC?), and I'd try to run IccMax a little above 200A with PL1/2 around 180W again, to see how it goes.

If none of it seems to run my cores at full speed (throttle), how would I push the voltage down if not with Lite load? Sorry if my questions are rather stupid I'm learning haha.

 

[citay]

If none of it seems to run my cores at full speed (throttle), how would I push the voltage down if not with Lite load? Sorry if my questions are rather stupid I'm learning haha.

If you can't lower it much more via CPU Lite Load, then you won't get huge improvements with any other methods either, especially for high workloads like Cinebench. This comparison of undervolting methods has shown CPU Lite Load to be very effective for full-load scenarios. So you just gently have to balance the different limits etc. until you find the maximum that you can get out of it with your cooling.

 

[Raxiel]

Yes it was taken during R23 all core (multi core) full load.
I will wait a couple more days until the 0x129 stable version (non beta) and will try to run some tests with lower Vcore (I would assume I do it by setting a more stable Lite load version, like 5? without doing any manual change to LL_AC/DC?), and I'd try to run IccMax a little above 200A with PL1/2 around 180W again, to see how it goes.

If none of it seems to run my cores at full speed (throttle), how would I push the voltage down if not with Lite load? Sorry if my questions are rather stupid I'm learning haha.

First off, Buildzoid did a video on your specific board this morning, which you might find interesting (be aware its three quarters of an hour long, he doesn't really do short form content, I've not had a chance to watch it all myself, some wierdness though).
(Edit: of, I hope for MSIs sake he just has a broken board.)
Honestly, looking at them again, your figures in #8 aren't too bad. I was getting similar power draw (~150W package power, ~130A SVID IOUT) and a similar CBr23 score - High 22k, just into 23k range with CPU Lite Load 7 (0.6mohm AC/DC) Load Line Calibration mode 6, and a -30mv adaptive+offset setting. During the test (not the load peak in between) the VID was around 1.18v and the VCORE was almost exactly the same at 1.181v. Peaks were 1.223v VID and 1.19v VCORE.

It was still occasionally hitting the current limit with those settings, but not all the time as it had been before. In the end I abandoned the Lite load steps and adjusted the AC/DC values separately, with AC=54, DC=68 (to match what I believe the LLC mode 6 value trims the VRM load line to) and bumped the negative offset to 50mV, that finally stopped it triggering either the ICCMAX at 200A or IA CEP

You could Just disable CEP, but personally, I don't want to go removing protection features after chips have been getting killed. It wasn't necessarily sustained voltages under load that did the damage, but spikes that occurred as the cores were unloaded (lower sustained voltage will still reduce the peaks though). It's still possible to Undervolt with it on, but due to CEP, using the AC Loadline alone might not be the best way to undervolt any more. MSI sets it for so much over voltage that it still has a large margin to come down though, so its still the best place to start. The default load line of the voltage regulator on your board is probably 1.1mOhm, so setting LL to that (mode12 probably) is probably a good start point if you wanted to go back to first principles.

 

[citay]

I've not had a chance to watch it all myself, some wierdness though).

I've watched some of it, it's neither a broken board nor a big issue. Microcode may be loaded during late stages of platform initialization by the BIOS, so there will be a short window where spikes can happen on the factory CPU code, until the 0x129 MCU is loaded into the CPU and the OS takes over. You most likely won't get any degradation worth mentioning from a spike or two during the early boot stage.