Problems with temperature

justdanyul · 2019-03-11 21:30:00

Hi all,

I got a strange issue relating to temperature. Basically, when stress testing, using mprime, sensors reports temperatures north of 93 degrees Celsius, so, I obviously stop the test and then scratch my head. Because,

- Conducting the same test under Windows 10 peaks at about 76-78 degrees Celsius
- I got a warning in the BIOS which is supposed to trigger at 90 degrees Celsius, but never trigger during the runs
- I can't see my fan speed, sensors-detect found fan1-fan3, albeit they are all running at 0 RPM (which isn't the case, I got a window into my case, and the fan is very much spinning

So.. I'm a bit baffled where to go next? It seems to me that the sensors are miss reporting, mostly based on the fact the BIOS alter isn't triggering. But, its a bit its a quite beefy box, and I bought it predominantly for data processing and the reason I started all this testing, was it's getting very warm (or maybe it isn't) when running longer data processing jobs (which are quite AVX intense).

The machine spec is as follows

i9 - 9900K
Gigabyte Z390 AORUS PRO
32 GB Memory

Any suggestions would be very much appreciated.

Last edited by justdanyul (2019-03-17 16:08:47)

Ropid · 2019-03-12 08:41:05

What driver is measuring that temperature? I'm thinking if it's the "coretemp" one, that should measure things correctly. If it's the driver for the board's SuperIO chip, it might be wrong because of differences with your board exactly and how the manufacturer has wired things up.

If the temperature you see is real, do your cooler and case fans seem react to temperature changes (I mean when you check just by listening to them)? Does the board offer some sort of fan curve setting in the UEFI/BIOS menus and have you set things up so that the case fans run faster when things get hot?

EDIT:

There's another thing that just came to mind:

mprime is using the AVX instructions that the CPU supports. There's AVX, AVX2 and AVX512 instruction sets. The way I remember things from following overclocking forums, starting with CPUs that introduced AVX2 ("Haswell" 4th generation core), people started experimenting with disabling AVX2 in Prime95 because temperatures when stress testing were getting impossible to keep in check. I don't know how people did this. Perhaps they intentionally went back to a version of Prime95 that only supported AVX and didn't support AVX2? You need to find out if the mprime you are using on Linux is behaving the same as the prime95.exe you are using to test on Windows.

Last edited by Ropid (2019-03-12 09:03:43)

ewaller · 2019-03-12 14:25:51

What does the utility i7z report for temperature. That tool should give you the canonical answer as to what the core temperatures are doing.
Check your journal. Are there throttling events occurring? (Note: this is not a bad thing, and is expected under heavy load)

kokoko3k · 2019-03-12 15:18:29

It seems to me that the sensors are miss reporting

Start the "bios" an write down the temperature it reports; then start linux and see if there are differences.

ewaller · 2019-03-12 15:30:52

kokoko3k wrote:

Start the "bios" an write down the temperature it reports; then start linux and see if there are differences.

This may be okay for a rough sanity check, but the "BIOS" (More likely uEFI Configuration TUI) does not work the processor(s) that hard. I would expect something on the order of ambient temperature + 10 degrees or so while in that mode. During boot, the processors might be doing a fair amount of work. With finite thermal resistance from the "junctions" of the chip, through the combined thermal resistance of the case and heatsink (on the order of 10 K/W) to ambient, T(j) could jump 50 to 60K in a matter of a couple seconds under heavy load.

justdanyul · 2019-03-12 20:40:22

Hi guys,

Thanks for all the answers. Right, I'll start answering from an end.

@Ropid: Yes, I'm using coretemp to get the readings. And thanks for the tip regarding AVX, I shall have a look ASAP and report back. But, just to add, when I actually do work on the CPU (as opposed to benchmark), I mostly use the intel python distribution, which contains a bunch of MKL optimised versions of various scientific python libraries such as scipy, pandas, numpy, scikit etc. I see the same temperature discrepancy here, and in these workflows, the same parts of the CPU would be exercised (as in AVX512 will be in play on both platforms).

@ewaller I shall check and report back :-D

@kokoko3k + ewaller I've already peekabooed in the bios quite a bit. But I'm unsure if I can draw any conclusions from it. Basically, I tired leaving it idle in the bios for a while, and see what it temperature it stabilises at. And then I've tried doing the same under both Linux and Windows. Windows is largely consistent with the bios reading, Linux is a small handful of degrees higher. However, I'm not sure how meaning full this comparison is. It feels like there is a ton of factors that could influence this.

Thanks everybody for the input, I'll report back later on when I got to have a look at i7z (on a separate note, I violently dislike having to boot into windows to do my work, so, I've ordered a beast of a 360mm AIO, and I'm secretly hoping that will bring the temps down to a level I'm comfortable looking at, over reporting or not :-D)

Arch Linux

#1 2019-03-11 21:30:00

Problems with temperature

#2 2019-03-12 08:41:05

Re: Problems with temperature

#3 2019-03-12 14:25:51

Re: Problems with temperature

#4 2019-03-12 15:18:29

Re: Problems with temperature

#5 2019-03-12 15:30:52

Re: Problems with temperature

#6 2019-03-12 20:40:22

Re: Problems with temperature

Board footer