Question MSI z690-a pro wifi + 12900kF + ddr5 + kraken x73 360mm cooler, high temps? bios settings? Fails after 5 hours p95 blend

[markm75]

Im using this combo with the ram being: CORSAIR Vengeance DDR5 32GB (2x16GB) DDR5 5200 (PC5-41600) C40 1.25V (32gb total), ‎CMK32GX5M2B5200C40

I have the lga 1700 bracket, everything is tight, cpu was done two different times with two different pastes/methods, same results. 2nd time now with arctic mx5 and did the spatula spread method.
Pump is running near 2800, cooler fans at their max rpm (2000ish).

In the bios i have only changed the P core and E core values.. set P to51x on all of them, set E to 39x, set the ring to 39x, set avx to -2.. set voltage to 1.30, LLC to 2.
I didnt turn off anything else or change other settings.. let turbo i think it is, to auto etc. Didnt set SA or Dram voltages, left on auto, set ram to XMP and auto on frequency (i think gear 2), 5200mhz.

All that said, p95 blend will fail after 5 hours. temps still randomly spike to 100c on 2 cores, though most of the time its around 87C.. if i do cinebench it spikes to round 100c again, usually avg around 89c.. score ~26,700 package load seems to avg 235-285 during cinebench, however i think it spiked to 350 at one point.. its nowhere near these wattage with P95 blend though

I'd have like to have thought the voltages are too high for this 360mm x 3 cooler, but i've seen 1.35 mentioned with similar configs and 95c, so unsure.

Any thoughts or suggestions, or maybe i need to tweak some bios settings.. for now i've lowered the ram to 4800 and i'm retesting blend to see why it errored at 1.30.
I ran into this with the 10900k with this same cooler, so im thinking worst case i try another brand? (suggestions?).. i had to delid the 10900k, then my worst temps were 89-92c.
I might redo things after doing the washer mod next

 

[Phaaze88]

You have some reading/watching to do...

Spoiler

View: https://www.youtube.com/watch?v=Ysb25vsNBQI

This only really affects the i9s. The i7s and below are far easier to manage.
So even if you find a happy medium now, the later to come warping will just put you back at square one.
There are commercial LGA 1700 contact frames available, but do so at your own risk, as mentioned in the above video.


Intel likely already maxed it out very close to its voltage threshold shift. Can find out more about VTS HERE in section 8. You probably won't get very far without custom loop or LN2; in other words, the gate of entry is much higher now.


Then there's all of these:
https://www.techpowerup.com/review/intel-core-i9-12900k-alder-lake-12th-gen/22.html

i9 12900k temperature problem

I recently bought and installed the i9 12900k with the Corsair iCUE H150i ELITE and I have noticed that my idle temp for the liquid jumps up to 35c and the package jumps around between 38c and 50c. At first I thought it was normal but then I started to play cyberpunk and the temps went up to...

www.techpowerup.com

12900k thermal throttling with 360 AIO

Im trying to use my i9-12900k with Cooler Master ML360 Illusion AIO (with socket 1700 mounting kit) I dont have a dedicated GPU in my system. Using intel integrated graphics for now. Problem is as soon as i start cinebench r23 or CPU-Z benchmark CPU starts thermal throttling at 100 degrees...

www.techpowerup.com

https://www.tomshardware.com/review...-i5-12600k-review-retaking-the-gaming-crown/2

[SOLVED] - i9-12900k cooling

New build with an i9-12900k cpu on a MSI Pro Z690-A Using a Noctua D15 with dual fans. Have a total of 6 case fans, all exhausting. Question is that under stress, cpu hits 100C on first 7 cores (HWMonitor) and 80-90C on the remaining. I know that this cpu does run hot although I don't know how...

forums.tomshardware.com

12900K review, gaming and blender loads, from LTT

Kitguru 12900K review

I'm seeing a pattern here...

Spoiler

View: https://www.youtube.com/watch?v=4jjopjkJzxA

Undervolting plus lower power limits, not OCing, is the way to go with that i9.


Prime 95 Blend runs all of the other tests, one at a time, and each one runs for like 10-20mins before moving on to the next one? It's been so long, I've forgotten.
It is not appropriate for thermal or voltage stability testing.
Small FFT
AVX, AVX2, AVX-512(if supported) all off
Run for at least 15mins
That's the go-to thermal stress. Prime95 is too steady for any kind of voltage stability testing; Cinebench R23 and OCCT are better for that.


Arctic has already discontinued MX-5 because they couldn't keep it consistent: https://www.techpowerup.com/forums/...n-the-market-offers-free-replacements.291072/
Might want to double check your package. How to tell: "If you push the plunger and a bunch of oily stuff comes out that means your TIM has separated and is garbage."




TL;DR: AIO not effective enough for OCing 12900K IF you don't want to touch upon undervolting plus lower power limits(you don't lose much at all if done right)... at that point, you may as well have gotten a 12700K, because that's what you more or less get after doing such, and it's much easier to manage.

 

[geofelt]

What is the make/model of your case, and where is the aio radiator mounted?
Do you also have any hot graphics cards installed?

What is your room temperature?

What will be the normal use for this pc?
Will you be running maximum threads in a batch environment, gaming, or what?

 

[markm75]

You have some reading/watching to do...

Spoiler

View: https://www.youtube.com/watch?v=Ysb25vsNBQI

Thanks for these, i had already stumbled onto much of these but ill take another look.

On the undervolting.. at one point i set the cores to auto ratio for everything (like it was out of the box), i tried adaptive with negative offset, set to -0.035 i think it was (some reported -0.08 being effective).. at any case, i didnt see lower temps not with realbench, still spiking 100c, but averaging 88C id say. I dont quite get the whole offset thing/undervolt.. i mean, if the cpu can run at a fixed 1.27 volt and not crash, why undervolt from say 1.30.. just set to a tighter LLC and 1.27? Cinebench would crash when the volts during the test were at like 1.25 with 5.1ghz/3.9 when i tried looser LLC or lower volts anyway and temps were still spiking. I should have looked at the cpu watts then too, i want to say its hit 350, which is way too high, but during prime95 blend with my fixed 1.30 and 5.1 its 227watts.

On the warping issue.. i have 4 nylon m4 1mm thick washers coming, i was going to do the washer mod, its very easy to do, the avg drop sounded like 5c. I have seen those brackets but wasnt sure which vendor/model was the best.

 

[markm75]

What is the make/model of your case, and where is the aio radiator mounted?
Do you also have any hot graphics cards installed?

What is your room temperature?

What will be the normal use for this pc?
Will you be running maximum threads in a batch environment, gaming, or what?

i forget the exact make and model, ill update here once i dig that up.. but basically, radiator is at the top, fans under the radiator push air up and out. I have fans on the front right side pulling air in. I do have a 3080, but its idle during most tests, or for avx tests doesnt feel like it has much heat going up near the cpu, but im sure there is some.
Granted this whole thing sits 3.5 feet up on a shelf, which isnt ideal, but the case is off for stress testing and i have an additional window ac unit cooling the whole room down pretty nicely, so i dont think thats the issue, pretty sure its mostly just the chip being hot/not sitting flush with that lga1700 issue etc.

Normal use is VR flight sim, which doesnt heat up much at all (even currently). However, i do do some rendering of astro images which can heat up my old 10900k to around 90c on a delidded cpu even.

I'd say if i can drop it down 5c i'd be good enough, or i guess worst case drop from a fixed 5.1 to say 4.9 or 5.0ghz (i was at 4.9 with the 10900k), though it would have been nice to have one core at 5.2 and the rest 5.1 Ill be trying the washer mod.

 

[Phaaze88]

Offset is easier and takes less time than adaptive. One could easily start with a negative offset of 0.050v and go up/down from there.
Adaptive, I've found to be time consuming because you first need to use fixed to find a lower, stable vcore. Once that value is found, it's used as the max adaptive voltage and then you apply a small positive/negative offset on top of that.
But if done right, it's slightly more effective than offset - at least under low; go too far in offset and the system can crash because the voltage dropped too low. Adaptive can push the voltage a little higher and keep that from happening.

Prime 95 in any mode is too 'flat' for vcore stability testing. Testing core thermals(small fft, all AVX options off), sure. Cinebench and OCCT in variable mode are much better for this.
Constantly changing loads = voltage stability.
Steady, heavy loads = thermal stability.

if the cpu can run at a fixed 1.27 volt and not crash, why undervolt from say 1.30.. just set to a tighter LLC and 1.27?

Because it can use less power during light loads... and who cares about that? Apparently, many do, with all the 'my idle temperature is this or that'... some folks really obsess over it, even though it's not any more dangerous than the part running at 80C under full load.

On the warping issue.. i have 4 nylon m4 1mm thick washers coming, i was going to do the washer mod, its very easy to do, the avg drop sounded like 5c. I have seen those brackets but wasnt sure which vendor/model was the best.

About that washer mod, it may only be good short term: https://www.tomshardware.com/news/intel-12900k-world-record-with-hacksaw (scroll down to the last 4 paragraphs below the censored image)

The Thermal Grizzly model is probably the best, since they worked with De8auer for that one, but ehh, there isn't that much info on them all in general.

 

[markm75]

Because it can use less power during light loads... and who cares about that? Apparently, many do, with all the 'my idle temperature is this or that'... some folks really obsess over it, even though it's not any more dangerous than the part running at 80C under full load.

Yeah i was aware on the lighter load part, but i never understood how anyone figured it would help with full load temps, because it shouldnt. Only cure there is lower stable max volts or lowering the ratios.

 

[Phaaze88]

Yeah i was aware on the lighter load part, but i never understood how anyone figured it would help with full load temps, because it shouldnt. Only cure there is lower stable max volts or lowering the ratios.

Adjust the Processor Base Power and Maximum Turbo Power settings. Set them both to 241w, or around 200w.
Intel more or less designed this cpu to run as fast as it can out the gate with little regards to fuel consumption and engine temperature. You have to manually put the throttle on it to a level you're comfortable with - even though the current behavior is technically within spec from Intel's POV.

 

[markm75]

I just noticed something worth trying in the bios, in reading those links above.. someone also with the msi had their cooler set to "water cooler" which sets higher wattage limits.. i guess maybe if i try boxed cooler i may find better results, worth a shot anyway

 

[markm75]

Adjust the Processor Base Power and Maximum Turbo Power settings. Set them both to 241w, or around 200w.
Intel more or less designed this cpu to run as fast as it can out the gate with little regards to fuel consumption and engine temperature. You have to manually put the throttle on it to a level you're comfortable with - even though the current behavior is technically within spec from Intel's POV.

Ah thanks, ill check that as well (i think the watercooler setting is probably doing this, needs to be "boxed cooler" instead, to bring down the wattage.

 

[Phaaze88]

Ah thanks, ill check that as well (i think the watercooler setting is probably doing this, needs to be "boxed cooler" instead, to bring down the wattage.

Oh yeah, I know what you're talking about. The liquid cooler setting is set stupid high on motherboards; the max is 4096w, if I remember correctly? Granted, your cpu won't pull anywhere close to that, but it's still free to go into the 300s if it's that high.

 

[markm75]

Oh yeah, I know what you're talking about. The liquid cooler setting is set stupid high on motherboards; the max is 4096w, if I remember correctly? Granted, your cpu won't pull anywhere close to that, but it's still free to go into the 300s if it's that high.

Yeah i had read that too, 4096w.. odd though, the actual setting for that still says auto, but maybe the water cooler setting just overrides it. This will surely help a few C at least Will test this eve

 

[markm75]

Just for reference, prior to changing to "boxed cooler", i confirmed it shows 4096w in hwinfo

prime 95 blend package levels are at 261 watts max, avg 85-87c, peaks to 100c, 1.270 volts or so loaded
cinebench is at 241 watts (27,423 score), initial 84c, then within seconds 100c, avg 85-87c, 1.264 volts or so load

real world photo / integration render, peak temp of 92C (i dont think it was using avx, which i have at -2)
I think i have the bios around 1.285 or so and llc3 but need to confirm

 

[markm75]

well, changing to boxed type didnt help alot.. though cinebench at first didnt hit 100c, but 97c instead, eventually it did hit 100c, i confirmed i had the bios set to 1.29v and llc3, avg cinebench maybe 83-84 instead of 85-87, watts went from 241 to 227w for this test.

I guess ill still consider lowering from 5.1 or trying again to lower volts, or still do the washer mod.

 

[markm75]

even at 1.22 load and 5.0ghz (override mode in bios and llc2), i still hit about 96C on the cinebench after a few mins (avg maybe 80c).. also, if you change the ratio, it tries to switch the mode back to watercooler from boxed to the full 4096w
1.25 load was more like 83c but 98C or more (at 5.1ghz), honestly the 1.22 (if its even stable and 5ghz might be closest i can do).

5.1 ghz and 1.22 bsod

These above with the boxed cooler 241w limitation

 

[Phaaze88]

I'm thinking it's the cooler mounting pressure + retention bar at this point. Undervolt + lower power limits should yield some pretty decent results.
See if the paste has pooled to one side of the IHS or not - if you have spare paste on hand to do that.

 

[markm75]

I did make one small progress, sort of, setting the bios to offset, negative 0.030, this seems to ring in at 1.30 or so under NO load and about 1.22 or 1.20 under cinebench load and its somehow surviving.. this at 5.1ghz / 3.9 again and boxedcooler wattage. Max temp now at 92C, avg 78C. When i tried 1.30 LLC4 override, i was getting about 96C under full load (vs 100C if i left it at watercooler mode). So the boxed mode helps by at least 4C no matter what.

Im puzzled by making it even 5 mins into this test with offset and load of 1.20v, i tried a manual llc2 (tight) at 1.23 even and had issues with bsod or cinebench errors. With the offset, idle it will hit 0.66v

It would have been nice to have 5.2ghz here, but if this pans out on 5.1 ill leave it be. (i guess ill try occt small variable no avx and maybe overnight on prime95 blend no avx)

 

[Phaaze88]

Spoiler

View: https://www.youtube.com/watch?v=NMIh8dTdJwI

 

[markm75]

UPDATE:
Well i did put the #6 washers on just now (managed to find locally), changed the thermal grease, noted that half of it was pushed to the side somehow last time (mx5), went with mx4 this time, did 5 dots, x pattern, one bigger one in the middle.

Same settings again, so far in the same amount of time as the last test before, it hit 90c, last was 96c, avg 67-76 (p95 blend avx2 is still on),15 minutes later its still only at 93c. So id say there is a 4-6C drop from the washers (or from redoing the paste

paste after cooler removal: (its as if the cooler wasnt sitting evenly)
https://1drv.ms/u/s!AiciZOUoW-_Xu_FE4ux_KPFv_X1_vg

side note: Im still having the issue that i had with the non wifi z690-a msi board, where if you change the ratios to any other value, you lose the nic lan in windows, error 10, have to shut off and hold in power button for 20 sec and wait a few mins.. however even after redoing the cpu cooler/washers this time, i still have no lan card working this time

EDIT: so far everything working great, my gaming/vr msfs temp only hit 73C. (offset neg 0.030 and LLC4 and boxed cooler mode)

 

[markm75]

So i had about 10 hours of successful prime95 blend with avx2.. max temp 94c during the period (side of case off, otherwise would have been 2-4c higher i think): (this all post washer mod and repaste with mx4)

summary of settings/results:
xmp with vddq at 1.35
ram volts at 1.35
ram speed on auto but at its 5200 mhz

Ratios 5.1/3.9/3.9
Offset mode negative and 0.030
Cooler set to "boxed" 241watt mode

Gaming max temp (MSFS/vr) : 76C (case panel off)
P95 blend 94c 10 hours (i think the draw was something around 1.2v or maybe a little less)
Cinebench r23 10 min morning test today: 83C max temp this time, 241 watts used, 27085 score
OCCT small without avx was hitting around 94c i think (at least last nite)



So at this point id like to figure out a few things.
I saw some posts about folks getting under 200 watt power draw, unsure how that is achieved.

I see many posts of up to 5.4ghz and similar, i'd also want to take a shot at at least 2 cores being 5.2 and maybe the ecores at 4.0ghz ..

**i tried initially leaving everything the same and raising to 5.2 on two cores, i found the volts on loading windows were 1.42, way too high.. unsure why the voltage rose with same offsets.

I tried a higher 0.08 offset but that didnt seem to bring the under load voltage down, more like 1.3v
Confused on that part.. also confused because in the past i swear you could enter a voltage like 1.30 and then set its offset, not just have the offset, seems the bios picks its voltage on boot up, i did notice a 1.40 or similar after changing ratios, i guess that is an indicator.

 

[markm75]

I now have ran with offset 0.03 negative.. and 5.2 ghz all core, 4.0 ecore, 4.0 ring, llc4 its about 1.15-1.16 again with small ffts, avx is set to enabled and -2, box water type cooler 241watt max
P95 will hit 90c after about 10 seconds and not exceed 94 on small ffts over 1 hour. Cinebench averages 73C with a max of 83c

After all this, i tried cinebench r23.. i only got 27,400 for a score.. and realized in hwinfo, the max cpu speed is only 4888.0 mhz..

So i tried setting it back to the 4096 watt watercooler setting, sure enough, it was ringing in at like 5099 for the frequency on the pcores and the score was more like 28,400

So much data i've looked at at this point, im pretty sure others with the 360 type coolers were staying 95c or less and getting 28,500 type scores (or even had avx on 0).

So i guess either my cooler is still not sitting evenly on the chip, or in my situation/chip i just cant use the watercooler setting to boost the avx mode any and have to settle for what i get.

When i last switched the paste on the cooler / cpu, i took a pic of how it looked, it was all to one side, so apparently not sitting flush. Probably related to the plastic bracket and not metal on the back. Since then i've done the washer mod as well.. no idea if the same is going on under the cooler right now but its possible. I swear my temps are a bit high for a load of 1.20

 

[Phaaze88]

AVX - all AVX options should be disabled on P95. It's just too extreme.

When i last switched the paste on the cooler / cpu, i took a pic of how it looked, it was all to one side, so apparently not sitting flush. Probably related to the plastic bracket and not metal on the back. Since then i've done the washer mod as well.. no idea if the same is going on under the cooler right now but its possible. I swear my temps are a bit high for a load of 1.20

You should still take a look if you have some spare paste. You'll never get even thermals with coverage being so lopsided like in the screenshot.
Underneath the IHS, the die is a vertical rectangle right down the middle.

 

[markm75]

AVX - all AVX options should be disabled on P95. It's just too extreme.


You should still take a look if you have some spare paste. You'll never get even thermals with coverage being so lopsided like in the screenshot.
Underneath the IHS, the die is a vertical rectangle right down the middle.

For some reason on the 4096w test despite disabling avx p95 hit 100c

On the paste issue I'm not sure how to fix the lopsided nature

 

[Phaaze88]

For some reason on the 4096w test despite disabling avx p95 hit 100c

On the paste issue I'm not sure how to fix the lopsided nature

Disabling AVX should've dropped thermals by quite a bit for the next run. This has to be the mounting pressure causing this.


You can't fix that pressure issue without something like the LGA 1700 contact frame. The washer mod has the potential to cause socket pins to stop making contact as the said socket gradually warps over time.

 

[geofelt]

Do you really have a problem that is worth fixing?
Are you a P95 overclocking competitor?
It seems that your current gaming and rendering runs well.
Even if you hit 100c. momentarily, the cpu does not throttle and you keep running.

I do agree that the thermal grease pattern you showed looks strange and offset.
But, not any die bending is evident.

I might think there was too much residual and a bit less paste would be in order the next time you apply some.
Paste is an insulator, compared to metal to metal contact.
But it is much better than air.
Use as l1ttle as possible.