QUESTION REGARDING STOCK CLICK BIOS SETTINGS...

[jpaleologo160d02eb]

Ok so...

I recently updated my mobo and cpu from z390 MEG ACE / 9900K to z690 MAG CARBON WIFI / 12900K and did so without having to reinstall windows (not sure if that part matters but please read on).

When I bought the new CPU and MOBO, I had the store mount the cpu and new ram and thy updated to newest bios for me.

Now, if I remember correctly, when I overclocked my old 9900k I had to go into the OC settings in the BIOS and manually change the Over Clock setting from Normal to Expert to proceed with the actual ovwrclock but on the new Carbon board, even with OPTIMAL BIOS SETTINGS turned back on, that setting sits on Expert out of the box.

What I'm wondering is... I don't have any OC on the 12900k atm (except for xmp on ram) and when running Cinebench r23 10 minute test, I get slight thermal throttle and I'm wondering if the BIOS itself may be "messed up" in its settings out of the box?

Also, I monitor with AIDA64 Extreme and also have Core Temp on thr machine along with NZXT CAM (mostky fir lighting) and the temps between these 3 softwares are NOT consistant at all.

At idle, AIDA shows the CPU at say between 30 and 35C and CAM for example shows like 26 or 27C.

Im sorry for what is likely a noob type question but can someone assist me or give me advice here? Can not reinstalling Window ls cause these types of quirks and is it possible that certain BIOS settings from the z390 board may be somehow "hung over" to the new z690 board?

Please advise...

THANKS!!!

 

[jpaleologo160d02eb]

Cool... Ok so I will have to wait till tomorrow / the weekend to really get into the CPU Lite thing as I don't have the time right now to run all the tests and ensure it's stable. That being said, I think I will check the thermal paste coverage on the cooler/ cpu as well and perhaps repaste / remount it incase it's not mounted well and I may also try the little mount block as well and see what it may do to help things as well... I will advise you as I go and look forward to more of your insight as I get this done...

Hey man... So I'm running linpak and I went straight from mode 9 to mode 5. I have also set the wattage back to 200/220w.

When linpak started up it asked me if I wanted to disable sleep and I said yes (you hadn't mentioned that option but I set it to not sleep).

It also asked if I wanted to run hwmonitor and I accidentally clicked yes as I thought it meant hwinfo64.

When the test started and hwinfo opened it showed most cores shoot to 100 but I'm not sure if that meant temp or frequency so I shut it off and opened hwinfo and it showed most cores maxed at between 55c and 67c. This was while running at a clock speed of 4973mhz.

I don't know if this is good or not but it's still running as I type this.

What I can say is that even though I am set to 200/220w PL, the max wattage it's pulling during this test is 145.98 watts as I'm writing this. It also shows no throttling of any kind BUT shows the max thread usage at 100% while the TOTAL cpu usage is only a max of 86.7%. My Times in linpak also seem really long compared to the example screenshot you provided. In your screenshot your times are like 77.xxx with gflops in the 300's and mine are like 122.xxx for times and in the 230's for gflops. Something doesn't seem right...

I can also say that in my bios there is a setting for CPU Lite Load Control to switch it from Normal to Advanced. Am I supposed to set this to Advanced before changing the CPU Lite Load?

Why isnt it pulling more wattage or using the whole cpu? Is this normal? Did I mess something up?

Please advise...

 

[citay]

When linpak started up it asked me if I wanted to disable sleep and I said yes (you hadn't mentioned that option but I set it to not sleep).

->

Run Linpack, select 2 (Stress test), 5 (10 GB), set 10 times, press Y to use all threads, 2x N, and let it do its thing.

What I can say is that even though I am set to 200/220w PL, the max wattage it's pulling during this test is 145.98 watts as I'm writing this.

Yes, that's the desired effect. You are lowering the voltage-per-frequency curve so it uses much less power at the same frequency/load.

Do not compare my times or GFlops to yours, i have a completely different CPU (no E-cores for starters) that makes it use different parameters, and it's not a benchmark, it's a stress test.
Also, i have some of the best DDR4 RAM, with manually optimized, very tight timings, which is something Linpack really likes. But again, it's not a benchmark.

Please do the following again: First let HWinfo64 run a bit on its own to get the minimum numbers.
Then let Linpack Xtreme run and make a screenshot of HWinfo64 after 10 minutes or so after it heated up enough.

 

[jpaleologo160d02eb]

->





Yes, that's the desired effect. You are lowering the voltage-per-frequency curve so it uses much less power at the same frequency/load.

Do not compare my times or GFlops to yours, i have a completely different CPU (no E-cores for starters) that makes it use different parameters, and it's not a benchmark, it's a stress test.
Also, i have some of the best DDR4 RAM, with manually optimized, very tight timings, which is something Linpack really likes. But again, it's not a benchmark.

Please do the following again: First let HWinfo64 run a bit on its own to get the minimum numbers.
Then let Linpack Xtreme run and make a screenshot of HWinfo64 after 10 minutes or so after it heated up enough.

Ok man... I will try this again later today and will take a snippet of HWINFO64 and post it after about 10 mins.

One other thing I forgot to mention is that when Linpak was done that pass (and it passed all items while set on mode 5) I ran a quick cinebench23 and it was SLOOOOOOW as molasses. After it's first run it said I had like 14000 points.

I know you said this behaviour is normal for linpak but is it normal for cinebench too? And if so, will it stay like that?

Sorry for all the noobish questions...

 

[citay]

Linpack will cause the so-called EWMA to be pulled to the maximum, completely exhausting the budget for the Short Duration power limit aka PL2.

I have somewhat avoided mentioning the EWMA until this point, because it wasn't that important thus far and will complicate to understand things easily.
But since you like to try out different things and wonder why your performance is suddenly lower, i will have to explain it.

So far, you should have understood this graph:

We allow a higher Short Duration PL2 for a minute or so, and after that, to not overwhelm the cooling on prolonged full (artificial) load, we limit to a lower Long Duration PL1.

But there is also an EWMA active. EWMA is the Exponentially Weighted Moving Average, a statistical calculation. This determines how long the CPU is allowed to stay in PL2 (the higher Short Duration power limit) if it previously had full load already. Otherwise, without that EWMA, even the shortest of idle periods would allow the CPU to draw PL2 power over and over again for the full Tau time.

So the EWMA will make sure that after longer full load, even with some idle time in between, the CPU can't go right back to PL2 for the full Tau time.
In other words, it will clock lower for a while until EWMA has dropped enough, so the CPU can't inadvertently overpower the cooling capabilities.

If you let the CPU sit there for a couple minutes doing nothing, the EWMA will drop again, and once it did, the full PL2 power draw will be allowed again for the full Tau time, improving the score.

Again, i have to point out that Cinebench R23 is not representative of most full load you will normally experience. It's a 10 minute onslaught on the CPU.
Even Linpack has some short breaks with lower power consumption in between, if you pay attention. And a lot of normal full load scenarios aren't maxing out the power draw.

 

[jpaleologo160d02eb]

Linpack will cause the so-called EWMA to be pulled to the maximum, completely exhausting the budget for the Short Duration power limit aka PL2.

I have somewhat avoided mentioning the EWMA until this point, because it wasn't that important thus far and will complicate to understand things easily.
But since you like to try out different things and wonder why your performance is suddenly lower, i will have to explain it.

So far, you should have understood this graph:

We allow a higher Short Duration PL2 for a minute or so, and after that, to not overwhelm the cooling on prolonged full (artificial) load, we limit to a lower Long Duration PL1.

But there is also an EWMA active. EWMA is the Exponentially Weighted Moving Average, a statistical calculation. This determines how long the CPU is allowed to stay in PL2 (the higher Short Duration limit) if it previously had full load already. Otherwise, with even the briefest of idle periods, the CPU could be able to draw PL2 power over and over again for the full Tau time.

So the EWMA will make sure that after longer full load, even with some idle time in between, it can't go right back to PL2 for the full Tau time.

If you let the CPU sit there for a couple minutes doing nothing, the EWMA will drop again, and once it did, the full PL2 power draw will be allowed again for the full Tau time, improving the score.

Id love to tell you that I actually understood any of that but Id be lying if I did... LOL!!!

I think I get the principal behind it all though... Less wattage drawn = Less Heat. Less Voltage Drawn = Less Heat and Less Power (for certain loads)... Am I at least close?

As for EWMA, I really have no clue what it does / how it plays into the end result as per your explanation above... LOL.

I guess my real question is... Say if I am able to get everything stable (all Linpak checks pass) on say Mode 4 as the example, and if the temps shown from that final stability test show sy 50's and 60's, is the next step then to go back to readjust the wattages of the PL1 andd 2 to make up for what appears to be lost power caused by the CPU Lite Mode adjustment?

Oh and again, in my bios there is a setting for CPU Lite Load Control to switch it from Normal to Advanced. Am I supposed to set this to Advanced before changing the CPU Lite Load? What is the difference between the Normal and Advanced Settings in this option? The Bios only says that by changing it you will be changing from Normal to Advanced but does not give an indication as to what it does exactly. I also googled it and there does not appear to be any relevant answers about it there either.

 

[citay]

Read the post again, i edited it with some more explanations. And here are yet more for you.

Think of EWMA as a power or turbo "budget". You know that once you have full CPU load, a timer of approximately one minute is ticking down (the time it will allow the higher power limit for), and after that it will only allow the lower power limit. Now, what would happen without EWMA, without the "power/turbo budget"?

Let's say you had one minute of the highest allowed power draw, and then for example 5 seconds of the lower allowed power draw. Then without an EWMA, if the CPU is idle for just one second after that, it would allow the full minute of highest power draw again. So in this example, you'd have one minute of highest power draw, 5 seconds of lower power draw, 1 second idle, and the full one minute full power draw would immediately be allowed again. This could overwhelm your cooling.

So now the EWMA comes in. During the one minute of highest allowed power draw, the EWMA is tracking it. Then the CPU gets limited to the lower allowed power draw, and then it's idle for a second. But the EWMA remembers that you just has a minute of highest allowed power draw, just a few seconds ago. So the EWMA will not allow the highest power draw so soon again. First there will have to be a longer idle period in between. This ensures that your cooling can always deal with the heat.


Now about your other questions. You don't have to touch the power limits again. We have determined the maximum values that your cooling can handle, so this will always stay the same, because your cooling is the same. We even set the limits a bit high perhaps, but that's ok. It's to get the maximum performance with acceptable downsides of power draw and heat.

Lowering "CPU Lite Load" will make the CPU draw less power at the same frequencies. So if the CPU notices this, and there is some frequency headroom left that it previously couldn't exploit due to higher power draw, then it will exploit that now. So the performance can actually rise a little (up to a certain point). If there was no frequency headroom anyway, performance will at least stay similar. Leave CPU Lite Load on Normal. "Advanced" influences the AC (VRM) and the DC (VID) Load Line seperately, it's not needed.

 

[jpaleologo160d02eb]

Hmmmmm... Ok. I think I have a bit better understanding of this now.

I have some errands to run in a bit and when I get back I will start at CPU Lite Load Mode 4 (as Mode 5 Passed ok yesterday) and will run Linpak and post the HWinfo64 results for you to see after about 10 mins as you have requested and then wait for a reply before I make any further changes to the CPU Lite Load...

 

[jpaleologo160d02eb]

Read the post again, i edited it with some more explanations. And here are yet more for you.

Think of EWMA as a power or turbo "budget". You know that once you have full CPU load, a timer of approximately one minute is ticking down (the time it will allow the higher power limit for), and after that it will only allow the lower power limit. Now, what would happen without EWMA, without the "power/turbo budget"?

Let's say you had one minute of the highest allowed power draw, and then for example 5 seconds of the lower allowed power draw. Then without an EWMA, if the CPU is idle for just one second after that, it would allow the full minute of highest power draw again. So in this example, you'd have one minute of highest power draw, 5 seconds of lower power draw, 1 second idle, and the full one minute full power draw would immediately be allowed again. This could overwhelm your cooling.

So now the EWMA comes in. During the one minute of highest allowed power draw, the EWMA is tracking it. Then the CPU gets limited to the lower allowed power draw, and then it's idle for a second. But the EWMA remembers that you just has a minute of highest allowed power draw, just a few seconds ago. So the EWMA will not allow the highest power draw so soon again. First there will have to be a longer idle period in between. This ensures that your cooling can always deal with the heat.


Now about your other questions. You don't have to touch the power limits again. We have determined the maximum values that your cooling can handle, so this will always stay the same, because your cooling is the same. We even set the limits a bit high perhaps, but that's ok. It's to get the maximum performance with acceptable downsides of power draw and heat.

Lowering "CPU Lite Load" will make the CPU draw less power at the same frequencies. So if the CPU notices this, and there is some frequency headroom left that it previously couldn't exploit due to higher power draw, then it will exploit that now. So the performance can actually rise a little (up to a certain point). If there was no frequency headroom anyway, performance will at least stay similar. Leave CPU Lite Load on Normal. "Advanced" influences the AC (VRM) and the DC (VID) Load Line seperately, it's not needed.

Oh and one more question - not related to this tweaking that you are helping me with, if you dont mind...

As you know, I just updated my Motherboard and Cpu from MSI MEG ACE Z390 to MSI MPG CARBON WIFI Z690 and from 9900K to 12900K, along with ne DDR5 RAM. Please keep in mind that I did all this in August / this month which is relevant to the info below...

When I bought the board, ram and the cpu, the retailer offered to "mount" the components and to update to the newest BIOS version. This is a free service offered by Memory Express here in Canada (kind of like a small version of Micro Center). As it was a free service, I had them do it for me.

They completed this and told me they updated the BIOS to newest and that was that. I then built the PC back up with the new parts and was able to get everything working on the same Windows 11 OS I was running on the old board / parts and did not have to fresh install the OS.

My question here is... My current BIOS version that the new board is reporting is version 1.70. It shows me this in AIDA64, System Info and in cmd prompt when using the
"wmic bios get smbiosbiosversion" command. Also, I should mention that in System Information, it shows the Bios Date of 29-June-22...

However, when I go to the MSI MoBo website / landing page, it shows me that the newest BIOS for my board is version 7D30v17 (no mention of anything like 1.70) and it shows the BIOS date as 2022-07-12 (July). I can also advise that according to the MSI website, the previous BIOS to this one is / was version 7D30v14 with a release date of 2022-05-24 (May).

As you can see, NEITHER of these BIOS versions appear to match what my system is reporting to me as the current version - not even the release date info and according to the MSI website, there was NOT even a BIOS release in June 2022 so Im not sure what to think here...

Are you able to provide me with any insight on this? Is it somehow possible that windows (and all applicable software) may be reporting my BIOS version incorrectly (perhaps some kind of hangover from the current WIndows build since I did not install a fresh OS)?

Please advise...

 

[citay]

BIOS v17 is 1.70. On my board for example, vA6 becomes A.60 when you read it out in Windows.

Then there is a difference between the release date (when it's released on the website) and the build date (when the BIOS was compiled).
The build date is always older, sometimes days, sometimes weeks, than the release date on the website. Here is an example from my board:

So you have the right BIOS version and the dates make sense too (BIOS was built/compiled about two weeks before it was released).

 

[jpaleologo160d02eb]

BIOS v17 is 1.70. On my board for example, vA6 becomes A.60 when you read it out in Windows.

Then there is a difference between the release date (when it's released on the website) and the build date (when the BIOS was compiled).
The build date is always older, sometimes days, sometimes weeks, than the release date on the website. Here is an example from my board:

So you have the right BIOS version and the dates make sense too (BIOS was built/compiled about two weeks before it was released).

You the man bro!!! Thanks...

Ok... I'm heading to do some running around and will continue the linpak thing when I get back home and will post results for you...

 

[jpaleologo160d02eb]

Read the post again, i edited it with some more explanations. And here are yet more for you.

Think of EWMA as a power or turbo "budget". You know that once you have full CPU load, a timer of approximately one minute is ticking down (the time it will allow the higher power limit for), and after that it will only allow the lower power limit. Now, what would happen without EWMA, without the "power/turbo budget"?

Let's say you had one minute of the highest allowed power draw, and then for example 5 seconds of the lower allowed power draw. Then without an EWMA, if the CPU is idle for just one second after that, it would allow the full minute of highest power draw again. So in this example, you'd have one minute of highest power draw, 5 seconds of lower power draw, 1 second idle, and the full one minute full power draw would immediately be allowed again. This could overwhelm your cooling.

So now the EWMA comes in. During the one minute of highest allowed power draw, the EWMA is tracking it. Then the CPU gets limited to the lower allowed power draw, and then it's idle for a second. But the EWMA remembers that you just has a minute of highest allowed power draw, just a few seconds ago. So the EWMA will not allow the highest power draw so soon again. First there will have to be a longer idle period in between. This ensures that your cooling can always deal with the heat.


Now about your other questions. You don't have to touch the power limits again. We have determined the maximum values that your cooling can handle, so this will always stay the same, because your cooling is the same. We even set the limits a bit high perhaps, but that's ok. It's to get the maximum performance with acceptable downsides of power draw and heat.

Lowering "CPU Lite Load" will make the CPU draw less power at the same frequencies. So if the CPU notices this, and there is some frequency headroom left that it previously couldn't exploit due to higher power draw, then it will exploit that now. So the performance can actually rise a little (up to a certain point). If there was no frequency headroom anyway, performance will at least stay similar. Leave CPU Lite Load on Normal. "Advanced" influences the AC (VRM) and the DC (VID) Load Line seperately, it's not needed.

Hey again man...

Sorry for the late reply of sorts...

I just wanted to let you know that for now, I think Im gonna just leave things as they are with the PL States at 200W / 220W and not fiddle with the CPU Lite Modes just now.

For now I am content (although a bit miffed that a 360 AIO cant cool this CPU) with the way its performing in games and stuff but in the future I may try the CPU Lite things again.

Thanks again so much for all of your assistance with this man. It is truly appreciated and I am glad that there are knowledgable people out here who will take the time out of their day(s) to help someone like me with issues such as this.

I hope you dont mind if I reach out to you again if / when I decide to do the CPU Lite Load thing my friend!

 

[citay]

Ok, sure, the decision is all yours. I think we already got a pretty good setup now.

The thing about cooling today's top-end CPUs, especially on Intel, is not entirely about the total heat output (although that is becoming worryingly high in the last two generations top models, 11900K and 12900K/KS), it's also about thermal hotspots on the CPU. The CPU surface is pretty small as it is, but underneath the heatspreader, the areas that actually produce the heat (pulling lots of current) are much smaller still, and the heat just can't be removed fast enough from there sometimes. Intel is to blame for configuring the top CPU models so aggressively that even very good water coolers struggle to cool them properly when the CPU is let off the leash.

Anyway, I'm always glad to help. Feel free to write here again when you want to set up CPU Lite Load.

 

[jpaleologo160d02eb]

Ok, sure, the decision is all yours. I think we already got a pretty good setup now.

The thing about cooling today's top-end CPUs, especially on Intel, is not entirely about the total heat output (although that is becoming worryingly high in the last two generations top models, 11900K and 12900K/KS), it's also about thermal hotspots on the CPU. The CPU surface is pretty small as it is, but underneath the heatspreader, the areas that actually produce the heat (pulling lots of current) are much smaller still, and the heat just can't be removed fast enough from there sometimes. Intel is to blame for configuring the top CPU models so aggressively that even very good water coolers struggle to cool them properly when the CPU is let off the leash.

Anyway, I'm always glad to help. Feel free to write here again when you want to set up CPU Lite Load.

Hey man...

So I just noticed something really weird here.

I put my PC to sl;eep overnight and just woke it up now and went to HWinfo64 to see temps after setting all my fans to fixed 100% to see how that affected temps at idle and then was gonna run another Cinebench test but upon opening HWinfo, I am seeing that it is reporting my Power Limits as follows ... PL1 220W / PL2 288W BUT in the bios I set this all back to 200 / 220W yesterady and have NOT changed it again since.

Can you advise why it seems that my PL states seem to have set themselves to different values upon waking from sleep (atleast according to HWinfo)? I have not checked the BIOS again but I am 100% sure that the PL States are still set to 200 / 220W.

Please advise...

 

[citay]

That's weird. And after a reboot?

 

[jpaleologo160d02eb]

That's weird. And after a reboot?

No... Only after waking the pc from sleep. I will reboot the PC now and take a look and advise... Oh and here is a snippet of what HWinfo is saying...

1 minute...

 

[citay]

Yes, that's what i mean, what if you reboot now, is it correcting itself. Then it would be a sleep-related BIOS bug you should report to MSI.

As a workaround, shut down each night for the time being.

 

[jpaleologo160d02eb]

Yes, that's what i mean, what if you reboot now, is it correcting itself. Then it would be a sleep-related BIOS bug you should report to MSI.

As a workaround, shut down each night for the time being.

Ok... I will restart the pc and let you know...

 

[jpaleologo160d02eb]

Yes, that's what i mean, what if you reboot now, is it correcting itself. Then it would be a sleep-related BIOS bug you should report to MSI.

As a workaround, shut down each night for the time being.

So I just restarted the pc and no change. HWinfo is still showing PL1 220W / PL2 288W. Should I try a Shut Down and then wait like 30 seconds and restart? Im gonna give that a try... BRB...

 

[jpaleologo160d02eb]

So I just restarted the pc and no change. HWinfo is still showing PL1 220W / PL2 288W. Should I try a Shut Down and then wait like 30 seconds and restart? Im gonna give that a try... BRB...

I just did a full shutdown, waitid like a minute and restarted the pc and it is still the same... PL1 220W / PL2 288W.

I have to run to the store for a few mins and then will be back. If you have any ideas as to why this is, please advise.

I have not run any benchmark or anything to see what the actual draws will be yet but I will. That being said though, HWinfo should see the PL states I set at 200 / 220W regardless right?

Be back in a few mins...

 

[citay]

Yeah, 288/220 would be no good, as i explained in my first reply already. Intel set up this CPU too aggressively, just to win certain unrealistic benchmarks, and guess what, Cinebench is one of them.
Don't put too much emphasis on Cinebench results, only look at your cooling, and we know that 220/200 is a good solution there. The sweet spot for calculation efficiency (energy spent per task completed) is even lower, somewhere between 190 down to 125W (for the long duration PL). So testing with the sky-high PLs of 288/220 will not lead anywhere, you'd be going backwards.