10900k real world temperatures, what is too high/expected? msi z490 Settings

[EarthDog]

I wouldn't go much above that. You likely can't temperature wise anyway.

 

[Sumguy]

Ya i was at 85c on small ftt. on my aorus AIO cooler. normal gaming and w.e else never see above 63. Moving over to a 2 loop set up this weekend when i build it..

And the difference in 200mhz prob never be noticed..

 

[EarthDog]

You have a few C. With P95 Small FFT, I'm happy if that tops out at 90C. 95% of loads are going to be 10-20C+ less than that. The chip starts to throttle at 100C.

 

[Sumguy]

ya.. i have a **** fit if it gets near 70 lol.. I just rmemeber from the old days lots of vcore is bad.. think i ran my 5820k at like 1.38 for 6 or so uyears? never an issue

 

[Sumguy]

I feel like there is other voltages I'm supposed to mess with other then just vcore

 

[markm75]

Just an update here on a few notes:

I was "stupid" and decided to delid. Well, it worked out thankfully I gained the expected "minimal" 5C.

I also decided to push for 5.1ghz (gotta have that extra 100mhzzzz lol)..

I'm now setting the bios to 1.330 LLC4 (msi board middle and lower number is tighter llc), this results in a load of 1.320 (1.325 with load of 1.322 or so was NOT stable, interestingly).
Ring is at auto (43x) for now.

This results in cinebench r20 one run at 6403ish and a max temp of 83C with room temp of 73F, i believe before i was roughly 88C, maybe 87. If you ignore the single core it hit 83C the rest were upper 70s to 80 max. My astro heavy avx integration (51x) ended at 84C with room at 73.6, down maybe more than 5c.
This on the 3 year old thermaltake that i still question if its working well enough though that coolmaster i tried was worse.

I say this because i saw someones stat of 1.31v load (unsure on ambient) and they said one run in cinebench r20 they were at 77C. I guess it could be the stat i saw that these chips can vary no matter what, by 10c in max temp.
edit: another comparison i found was 84C cinebench with coolant at 29C (i cant read mine, no usb), ambient about 24C, so he was about the same, but not delidded, so i'm maybe off 5C unless chip temp lottery. (edit he was using a custom loop so i think my numbers are about right then)

 

[markm75]

My 51x settings previously were not quite stable, i was getting L0 cache errors in hwinfo.



I was able to use occt linpack to quickly test and get rid of those errors (along with the std tests) so now i'm at these for 51x:



51x, ring auto (43), 1.35 bios , llc4, avx auto (51x)
Load 1.339
My astro avx integration hits 85C (at 73F)
Cinebench 84C and 6400 score



**if i up the ring to 47x
My astro avx integratin hits 88C (at 73F)
Cinebench 85C and 6560 score
Passmark cpumark at 47x ring is 27,787

Again all these temps are about 5 or 6C cooler after delidding, i may just leave it at 47X.

 

[Woomack]

Right now I'm testing 10900K on Asetek 120 AIO (single Noctua fan) for the next ITX build. It clearly won't run high because of the radiator's size, but I still wanted to share my results.
5.0GHz, AVX 4.7GHz, RING 4.7GHz, 1.20V, max 168W and 88°C
Stability tested with AIDA64 CPU+FPU, voltages/wattage/temps checked with hwinfo64

I find it really efficient compared to default settings with 200-300MHz higher clock but 100W higher power draw.

At least on my 120 AIO, throttling starts at ~200W what is about 1.27V. This is also what motherboard sets automatically when I leave everything at auto and only limit maximum CPU ratio to x50 and -3 AVX offset.
At 1.18V the CPU passes 30mins of stability test and is freezing. At 1.19V all seems fine but to be sure I set 1.20V. My MSI Z490I Unify at auto keeps the voltage about as high as I set in BIOS so I'm not checking LLC or other options.

 

[markm75]

My os has bsod a few times, i cant seem to figure out why. 5.1ghz 1.345 (now 1.35 after last bsod) either way load of 1.339, avx 0, ring 43, ram 4000 at xmp and autos.
I can pass 12 hours prime 95 blend (avx fma set to off), i can do 6 hours small ffts (avx/fma off), i can do hours of occt small fft, i can do realbench for 4 hours etc, still a random bsod.

The last one seemed to be related to a western digital driver/hdd (Wdf01000.sys 0x0000010d)
. Before that it was kmode_exception_not_handled.

Hoping that raising to 1.35 prevents this, though it passes the tests all the same and load didnt change (read that sometimes it may like a higher vdroop), so .90mv vs .95mv i guess. I also didnt reinstall the OS after swapping motherboards, so unsure if its related. I may try running the same stress tests in the culprit OS (i run on a clean OS/isolated stress tests).

 

[Woomack]

I guess it's a general problem with these CPUs. If you set more than 5GHz with AVX then even the best chips fail longer high load or when they switch between AVX and no AVX or set higher/lower voltages.
Most reviews and overclocking results confirm about 5.2-5.3GHz on all cores without AVX when AVX offset is by default -3 so 4.9-5.0GHz with AVX in most cases. I had 5.1GHz with AVX on my CPU but I wasn't testing it for more than 1h as I was using it only for tests. Right now I started to test it for stability but in a small ITX PC so I can't give you any comparable results as you expect higher clocks.

Today I switched from 120 AIO to 92 AIO. I want to check how it runs compared to i5-10500 and maybe a bit higher clock keeping a similar power draw.

Edit:

If anyone is interested, here is my result with 10900K and Asetek 92mm AIO closed in a Kolink Rocket ITX case.
[email protected], AVX -3, 1.08V ~125W max with AVX, max temp AVX = 96°C, max temp w/o AVX = 80°C (1h test in a closed case)
Before in the same case was i5-10500 - 95W max at 4.5GHz/4.2GHz AVX, max temps around 80°C

 

[markm75]

I guess it's a general problem with these CPUs. If you set more than 5GHz with AVX then even the best chips fail longer high load or when they switch between AVX and no AVX or set higher/lower voltages.
Most reviews and overclocking results confirm about 5.2-5.3GHz on all cores without AVX when AVX offset is by default -3 so 4.9-5.0GHz with AVX in most cases. I had 5.1GHz with AVX on my CPU but I wasn't testing it for more than 1h as I was using it only for tests. Right now I started to test it for stability but in a small ITX PC so I can't give you any comparable results as you expect higher clocks.

Today I switched from 120 AIO to 92 AIO. I want to check how it runs compared to i5-10500 and maybe a bit higher clock keeping a similar power draw.

Edit:

If anyone is interested, here is my result with 10900K and Asetek 92mm AIO closed in a Kolink Rocket ITX case.
[email protected], AVX -3, 1.08V ~125W max with AVX, max temp AVX = 96°C, max temp w/o AVX = 80°C (1h test in a closed case)
Before in the same case was i5-10500 - 95W max at 4.5GHz/4.2GHz AVX, max temps around 80°C

I didnt think the default avx was -3? if so that could explain why my blender tests are high, about 89 to 91 (after delid) after only 2 of 6 of the benchmark done (74-77F room temp, case side on).

When i switch back to defaults as i had to briefly, i was pretty sure it was set to 0 or maybe just auto for avx and when ran everything was at 49x (or maybe at higher loads it does just stay 49x or -2 in my case), that said i could drop the avx back to -2 or -3 from 5.1 and probably save quite a bit on the load temps for avx.


Today i'm also swapping out my cpu aio cooler from the thermaltake 2017 version to the x73 kraken aio 360. Mainly because the fans they have on the old TT arent being made anymore and any others cant interface with the controller. Maybe ill get lucky and the new one will cool slightly better. I also still use the pea size arctic silver 5 technique on the chip, i assumed tried and true.

 

[Woomack]

Today i'm also swapping out my cpu aio cooler from the thermaltake 2017 version to the x73 kraken aio 360. Mainly because the fans they have on the old TT arent being made anymore and any others cant interface with the controller. Maybe ill get lucky and the new one will cool slightly better. I also still use the pea size arctic silver 5 technique on the chip, i assumed tried and true.

Apply, mount and check how the TIM spreads. How much you need and the way you apply depends on used TIM. That pea size way usually leaves me parts of the IHS not covered. If you use arctic silver 5 then better change it to something else as new TIMs are usually better and spread also better. Something at least like MX4, Noctua NT-H2, Gelid Extreme, Grizzly Kryonaut ... the list is longer.
Most new AIO have well-performing fans. The main difference is usually how much noise they make. The performance depends more on the block/pump and the coldplate design. Because of that I guess that the new cooler should perform a bit better. As I remember, older TT coolers were pretty average but I'm not sure what about yours. When I'm using AIO in my test rig then it's usually Enermax.

 

[markm75]

Apply, mount and check how the TIM spreads. How much you need and the way you apply depends on used TIM. That pea size way usually leaves me parts of the IHS not covered. If you use arctic silver 5 then better change it to something else as new TIMs are usually better and spread also better. Something at least like MX4, Noctua NT-H2, Gelid Extreme, Grizzly Kryonaut ... the list is longer.
Most new AIO have well-performing fans. The main difference is usually how much noise they make. The performance depends more on the block/pump and the coldplate design. Because of that I guess that the new cooler should perform a bit better. As I remember, older TT coolers were pretty average but I'm not sure what about yours. When I'm using AIO in my test rig then it's usually Enermax.

Yeah i did notice the last time i took it off, it wasnt covering the outer edges so maybe i should do a thin X over the top if arctic silver still.

I may have old MX4 on hand, but im not sure if too "Old".

I do have some grizzly liquid metal still avail, but i was nervous about putting that on the outside (of course its on the die when i delidded). Is that a common thing now? And how to apply, i guess i can dig up a video. I read that it shouldnt be used if you have aluminum for the base plate of the cooler however.

Is mx4 applied the same or spread using the tool first?

I dont know if the x73 has paste already on it or not or if its any good? (edit i think it does and some say its good as is)

 

[markm75]

Tried a first blender test with the x73, and it shot up to 100c, something is not quite right so far.

EDIT: installed the software, now my max by mid step 2 of 6 blender is about 85C (vs 91 before), however, these fans are a bit noisy. Not sure why the software needed to be running though.

 

[RJARRRPCGP]

kmode_exception_not_handled.

In my experience, it's the CPU temp being too high when OC'ing and overvolting.

 

[markm75]

Turned out the software cant control the fans at least with the x73 version. I guess msi dragon should, but when i installed it i didnt see fan options, only gaming related things.

I set curves in the bios, set to 80% at 84C, now the noise levels are more managable and closer to that of the thermaltake, but still a tad more, even though similar rpms.

At 100% its very noisy and noticeable, i dont think the Thermaltake was ever that bad even at the worst.

I gain about 2C at the same max rpm (80%, roughly 1550 rpm), gain more like 4c if i dont mind the noise at 100%.

I thought about swapping the fans for quieter ones (noctua maybe?) but they could be less efficient. Ill probably just leave as is, because it only climbs to 84C in those rare avx integration routines i do, infrequently. Blender of course cranks them.

 

[markm75]

Update here, so today i got the 3080 rtx ftw3 video card. It replaced the 1080ti hybrid and its rear exhaust 120 fan and radiator.


I dont know if this is the factor here or not, but now even after putting a single 120 fan back there in its place, the temps are getting about 10-11C warmer on the cpu running cinebench than they were with the 1080ti card and its rear radiator/fan.

Cant figure out whats going on.

 

[EarthDog]

You added around 70+ more Watts with that upgrade. The heat is also being dumped inside the case now. Makes sense to me.