Question 9900k is thermal-throttling at stock setings running Prime95 ?

[shiftyeyedkirk]

I've been experiencing some BSODs after long gaming sessions, so I started stress testing my CPU to see if my overclock was causing it. When I reverted to stock, the BSODs issues went away and I discovered that over time my cooler was just not cutting it. After 10-12 minutes of Prime95, my max hits 100c and averages are 95c. It then starts thermal throttling (still better than a BSOD I guess). This seems abnormally high for stock settings. I've tried re-seating the cooler, but still the same result. My cooler is an NZXT Kraken X63. Running on an MSI Z390 Gaming Pro Carbon.

 

[TerryLaze]

What do you consider stock settings?!
If you just did reset settings in bios then that aint it, mobos often use extreme settings for defaults, you have to put in all the settings by hand to be sure that they are stock.
MCE (locks all cores to the highest clocks) and TAU (turbo duration) are the biggest issues alongside PL1 and PL2 / short and long duration power limits.

 

[shiftyeyedkirk]

I'm using the motherboard defaults. With the overclock, it was pinned at 4.8 Ghz. Without, the cores clock down but can boost to 4.7. They don't appear to go higher than that even under load. Voltage doesn't exceed 1.25v. I'm less interested in dialing in the settings on the CPU until I have determined if my cooler is working properly. When my CPU temp is skyrocketing to 90c-100c, the AIO liquid temps are reporting at 41c. I know they'll be markedly lower usually and that they take time to equalize, but that seems low.

 

[hotaru.hino]

AIO coolers can degrade in performance over time, mostly due to the liquid evaporating, though it shouldn't be super terrible. Like you're not going to lose half the water in a 3-4 year period, but you're likely to hear a bit of slosh in the radiator.

Some things to do to make sure the cooler is working as intended:

• Make sure the CPU block is lower than the highest point in the system. Ideally it should be the lowest point
• You may want to take the cooler out and hold it in a way so the radiator is at the top, then give the CPU block a bit of wiggling to dislodge any air bubbles that might've gotten trapped there.
  • I'm not sure if you're supposed to do this while the system is running, but you may need to do this while it's on since water will be flowing

 

[shiftyeyedkirk]

So I removed the cooler and radiator and held the block and gave it a good shake. I didn't hear much sloshing around when I shook the radiator. Re-applied thermal paste (Kryonaut Thermal Grizzly) and remounted. Re-ran Prime95. Set all fans to run at 100% throughout the entire test just to see how much heat I could dissipate. The CPU thermal throttled at 7 minutes.
Ambient Starting Temp: 24c
AIO Liquid Idle Temp: 30c
CPU Idle Temp: 33c
CPU immediately after starting test: 58c
AIO Liquid temp immediately after starting test: 31c
CPU after 5 minutes: 73c
AIO liquid after 5 minutes: 39c
AIO Liquid temp at point CPU thermal throttles (95c): 42c

These results seem to indicate that my AIO has seen better days unfortunately. Though the 9900k does put out a LOT of heat.

 

[Deleted member 2947362]

what are the voltages set at for the CPU?

 

[shiftyeyedkirk]

Voltage varies from 1.182 to 1.328 with the current settings.

 

[Phaaze88]

gaming sessions ≠ Prime95

That was quite the jump, considering those loads are leagues apart, so there was no need to use Prime95 at all. Prime95 can bring any cooler to its knees - no surprise with a 9900K, although the settings used in Prime are unknown, not that it matters much, as even the lightest settings on Prime 95 are extreme.
It's puzzling that cpu core temperatures in the games wasn't shared, but this impractical application was.

'Stock' settings vary between motherboards - some are already overclocked, and others might have reduced power settings.


TL;DR: Provide the in game core temperatures.

 

[Deleted member 2947362]

Voltage varies from 1.182 to 1.328 with the current settings.

I don't know really anything about your CPU but that's where I would start looking and monitoring.

monitor the max voltage while your testing and look for info on your CPU's voltage characteristics to see if they are in line, maybe part of the cause?

 

[shiftyeyedkirk]

That was quite the jump, considering those loads are leagues apart, so there was no need to use Prime95 at all. Prime95 can bring any cooler to its knees - no surprise with a 9900K, although the settings used in Prime are unknown, not that it matters much, as even the lightest settings on Prime 95 are extreme.
It's puzzling that cpu core temperatures in the games wasn't shared, but this impractical application was.

'Stock' settings vary between motherboards - some are already overclocked, and others might have reduced power settings.


TL;DR: Provide the in game core temperatures.

I never stated that my sole reason for running a synthetic benchmark was for better stability in games. It was for the purpose of testing my cooler's ability to dissipate heat when my CPU is under a very high load. Synthetic benchmarks accomplish this task much faster than say AIDA64. The stated goal is to have stable temperatures while running Prime95 blend. That is my benchmark for temperature stability. If you are fine with a lower one, fine; but don't disparage others for choosing differently.

 

[shiftyeyedkirk]

I don't know really anything about your CPU but that's where I would start looking and monitoring.

monitor the max voltage while your testing and look for info on your CPU's voltage characteristics to see if they are in line, maybe part of the cause?

Thanks I'll try maybe a custom undervolt and see if that does the trick.

 

[Deleted member 2947362]

I never stated that my reason for running a synthetic benchmark was for better stability in games. It was for the purpose of testing my cooler's ability to dissipate heat when my CPU is under a very high load. Synthetic benchmarks accomplish this task much faster than say AIDA64. The stated goal is to have stable temperatures while running Prime95 blend. That is my benchmark for temperature stability. If you are fine with a lower one, fine; but don't disparage others for choosing differently.

Run aida64 stability test with just the stress FPU enabled and tell me it does not heat up fast.

I'll bet it does, and if your overclocking I would of thought it would get hot very quickly

 

[Phaaze88]

I never stated that my reason for running a synthetic benchmark was for better stability in games.

Really..? The first sentence in your first post implied otherwise.
"I've been experiencing some BSODs after long gaming sessions, so I started stress testing my CPU to see if my overclock was causing it."
Am I crazy for getting that impression?

Synthetic benchmarks accomplish this task much faster than say AIDA64. The stated goal is to have stable temperatures while running Prime95 blend. That is my benchmark for temperature stability. If you are fine with a lower one, fine; but don't disparage others for choosing differently.

Spoiler

View: https://www.youtube.com/watch?v=NGHiRrQ2AAo&t=281s

I see. Carry on then. I'll leave you be.

 

[shiftyeyedkirk]

Run aida64 stability test with just the stress FPU enabled and tell me it does not heat up fast.

I'll bet it does, and if your overclocking I would of thought it would get hot very quickly

I'm not running an overclock right now, but I suspect the motherboard defaults are scaling the voltage up too quickly for my AIO to handle. I normally do run AIDA64 for CPU testing. Right now I'm testing worst case heat dissipation. I want rock solid stability because I do all kinds of workloads on this PC: premiere exports, CAD, 3d slicing, gaming, streaming, modding, etc. Due to all of that, I value stability far higher than squeezing every last frame out of my games.

 

[Deleted member 2947362]

Really..? The first sentence in your first post implied otherwise.
"I've been experiencing some BSODs after long gaming sessions, so I started stress testing my CPU to see if my overclock was causing it."
Am I crazy for getting that impression?

Spoiler

View: https://www.youtube.com/watch?v=NGHiRrQ2AAo&t=281s

I see. Carry on then. I'll leave you be.

I thought they were testing the CPU 'sstability for overclocking and worried about high temps.

That's why my first thought was check what the voltages are doing as his cooler seem quite a good one and not some £10 budget one.

 

[Deleted member 2947362]

I'm not running an overclock right now, but I suspect the motherboard defaults are scaling the voltage up too quickly for my AIO to handle. I normally do run AIDA64 for CPU testing. Right now I'm testing worst case heat dissipation. I want rock solid stability because I do all kinds of workloads on this PC: premiere exports, CAD, 3d slicing, gaming, streaming, modding, etc. Due to all of that, I value stability far higher than squeezing every last frame out of my games.

with everything stock best idea I can think of is under volt like you have already said.

Is it a good idea to overclock a PC you need for pro work ? not to sure knowing the issue it can cause, if your flipping between the 2 like O/C then Stock is it even really worth the gains? I guess only you will know the answer to this.

I would just keep it stock maybe under volt if it helps, only other thing is if your setup is inherently hot by design is getting cooler lapped and CPU

 

[Phaaze88]

I thought they were testing the CPU 'sstability for overclocking and worried about high temps.

Cinebench R23 is better for checking stability, not Prime95 Blend.
For thermal testing, there's Prime95, small FFT only, all AVX options off, or Aida's FPU only test.

Prime95 Blend cycles through smallest, small, medium, and large FFTs. What gets loaded is different too.
The AVX options may not have been disabled, which is just too hard if it's enabled; together with small or smallest FFT, of course it's going to throttle.

That's why my first thought was check what the voltages are doing as his cooler seem quite a good one and not some £10 budget one.

Stock/auto settings generally are too high for Prime95. Custom voltages or bust.

 

[Deleted member 2947362]

The AVX options may not have been disabled, which is just too hard if it's enabled; together with small or smallest FFT, of course it's going to throttle.


Stock/auto settings generally are too high for Prime95. Custom voltages or bust.

Is that to do with CPU work load TDP which would be normal behaviour for CPU's and doesn't that kick in when the CPU cores are under full work load and can no longer boost to the max or I have I got that wrong?

 

[Phaaze88]

Is that to do with CPU work load TPD which would be normal behaviour for CPU's and doesn't that kick in when the CPU cores are under full work load and can no longer boost to the max or I have I got that wrong.

AVX is a separate instruction set from the SSE set that you normally see the cpu operate under. At the same frequencies, AVX is faster, but voltage and power demands are higher.
In games, AVX is used during effects like bloom and explosions. It doesn't run throughout a game.
It doesn't require all cores to be loaded.

Motherboards have AVX offset settings to help you control the heat if the cpu gets too hot during AVX. Downside to that is all the cores clocking down if just one core is running AVX, instead of just that one core. How far of an offset for the cpu's performance in SSE to match AVX, I'm not sure - I think it's 3.

 

[Deleted member 2947362]

AVX is a separate instruction set from the SSE set that you normally see the cpu operate under. At the same frequencies, AVX is faster, but voltage and power demands are higher.
In games, AVX is used during effects like bloom and explosions. It doesn't run throughout a game.
It doesn't require all cores to be loaded.

Motherboards have AVX offset settings to help you control the heat if the cpu gets too hot during AVX. Downside to that is all the cores clocking down if just one core is running AVX, instead of just that one core. How far of an offset for the cpu's performance in SSE to match AVX, I'm not sure - I think it's 3.

well I learned something there cheers

 

[shiftyeyedkirk]

Thanks for all the useful info Phaaze. I undervolted using a -0.01v offset, but for some reason the mobo is still hitting 1.32v under load. Not sure why.

 

[Deleted member 2947362]

Thanks for all the useful info Phaaze. I undervolted using a -0.01v offset, but for some reason the mobo is still hitting 1.32v under load. Not sure why.

my board has it's quirks with voltages on ram xmp set it at 1.35v motherboard was setting it to 1.38v but now seems to sets it at 1.362v and goes up to 1.38v ?

I have to manually set the ram voltage to 1.340v for it to stay between 1.352v and 1.344v

even on stock motherboard setting the all core speed is 3600mhz rather than 3500mhz it's supposed to be and boosts to 4450mhz rather than 4400mhz.

To get it to run the CPU to run at it's stock values, I have to turn stuff off that' been preselected by the motherboard default stock setting.

how dumb is that considering this board is not an overclocking board ? lol

 

[hotaru.hino]

I'm too lazy to catch up on the posts but by thermal throttling, does it drop below base clock speeds?

 

[shiftyeyedkirk]

Yes, it does drop below the boost clock speed.
I managed to undervolt to 1.18. Oddly enough, I had set core voltage via override to 1.275 as the 1.2v was not quite enough power. Then when I checked the voltage after a reboot, it was 1.18v. I also added a -2 offset to AVX workloads and set the base ratio at 47, meaning it should default to 4.7ghz unless it's on an AVX workload. In practice, it seems to have just reduced my overall clock speed regardless of what's running. With this new voltage, it settled out at 90c after 15 minutes and it appears to be rock solid there.

 

[shiftyeyedkirk]

So after I realized I'm a gigantic buffoon and was looking at VID instead of vCore, the troubleshooting got much easier. Ran for 30 minutes so far at 85c with a voltage offset of -0.115. Boosting to 4.7ghz. Thanks for all the advice. I might even try going lower to see how cool I can get it!