Question 5600X temps low during Prime95 , but spike higher when launching League of Legends???

Nov 13, 2021
44
2
35
0
Hi guys,

Since I swapped my CPU for a 5600X, I've been keeping an eye on the temperatures with HWInfo64.

The temps are around 32C when leaving the computer alone, 40-50 during regular light use, and 60 during gaming. My ambient temp is 27C, so these are all great temps as far as I am concerned.

However, sometimes they spike up to ~80C for a very brief period. I don't like this because I chose the 5600X paired with a two-tower cooler to keep those damn temps down and the PC quiet!!

I'm trying to understand why this might happen, because if I run a Prime95 Smallest FFT torture test, the temperature never rises above 65C. Yet when I boot up the League of Legends launcher (I live in China so its the Chinese version), sometimes the temperature spikes up to 80C. Is this just a case of horrible anti-cheat software? Is it windows defender? However horrible it is how can it be more intensive than the Prime95 torture test?

Looking forward to learning more about this! Specs below.

Cheers,
Pin

MSI Mag B550M Mortar
Ryzen 5600X
2x 8GB Corsair Vengeance LPX DDR4 3200mhz
Sapphire Pulse Radeon 5700XT
EVGA Supernova G3 550
Windows 10 Pro
 

DimkaTsv

Great
Nov 7, 2021
156
23
95
3
I'm trying to understand why this might happen, because if I run a Prime95 Smallest FFT torture test, the temperature never rises above 65C
Try to do it on 4 threads, not 12.
5600x is not very hot chip, but definitely requires good SPEED of thermal trasfer at few points (hotspots), which correspond to 2-4 core load
Games never load 12 threads as well as never give enough load, to make cores underclock themselves by curve.
It can be mounting issue too... I managed to get few degrees off just by reseraching, tightening up bolts more and moving my cooler so copper tubes laid across bit different parts of IHS...

But spikes themselves aren't that huge of a problem iself for Zen3...It is something that exists. Thermal dissipation of cooler isn't tight place, but thermal transfer speed - is.

Btw, are these with stock PBO settings? Like base 76W PBO?

P.S. Will buy PX-3 thermal paste and try to remount my cooler even better soon, maybe i will manage to take down few degrees more)
 
Last edited:

DimkaTsv

Great
Nov 7, 2021
156
23
95
3
Yeah, maybe. I'm hitting 75c with 4.75 Auto OC, max 74w. I just expected it to be a little higher for the OP.
If you want to lower "low load - high frequency" temps, you may look into limiting your EDC, as sometimes it is factor that affects "frequency given"... 90A is stock PBO though.
Prime95 is highly efficient though and doesn't rely on EDC, but mostly on PPT, so... limiting EDC will not lower these temps
 

DimkaTsv

Great
Nov 7, 2021
156
23
95
3
Yeah, maybe. I'm hitting 75c with 4.75 Auto OC, max 74w. I just expected it to be a little higher for the OP.
Well 4 threads is average 79.5 degrees for me with peak temperature of 81. But i have pretty sick UV settings for PBO if you may ask
PBO CO negative 30-20-20-25-25-30
And Vcore offset -0.066. So i am pretty darn efficient in using power there.
Still somehow kinda wish to lower temps a bit more...
When i built system i used like 6-7? if not more years old MX4 from tube. It is working for sure, but i wanna know if swapping it for new paste with higher efficiency manage to get my grip on few more degrees)

Actual cooler placement is a pain though, as i don't have X-Ray vision, so cannot see line when specifically CCD is placed under IHS. I know approximately where, but i kinda want to be more precise there. Well, at least my yesterday work actually was beneficial

P.S. if only AMD chips were turned 90 degrees on motherboard... or at least their insides were turned.
Because CCD is perpendicular to copper tubes of coolers, while being parralel would've been more beneficial imo... On other side though, if i should think deeper. Hottest parts of chip aren't cache, but cores... and they are at sides of CCD. So if i manage to place copper tubes right on cores in CCD, while making aluminium part lay on cache part of CCD, it may be even more beneficial... But precision required to do that is insane concidering you cannot see <Mod Edit> that happens on bottom side of cooler
 
Last edited by a moderator:
Nov 13, 2021
44
2
35
0
Hey guys thanks for the replies.

Yes, all my stuff is up-to-date, this CPU swap came at the end of a prolonged BSOD troubleshoot.

The temps I listed are at stock settings. I've decided to undervolt with PBO2 by a very small amount (-15 all cores curve) which has shaved a few degrees off, especially on single core workloads.

Will try the 4-thread Prime95 test now, great idea!
 
Nov 13, 2021
44
2
35
0
Try to do it on 4 threads, not 12.
5600x is not very hot chip, but definitely requires good SPEED of thermal trasfer at few points (hotspots), which correspond to 2-4 core load
Games never load 12 threads as well as never give enough load, to make cores underclock themselves by curve.
Yup, this did it. At 4 threads temps shot up to 75C right away. Nice, this explains it! Tried 12 threads again and it stayed at ~60C

Low-thread, high freq = more voltage = more temp on the 5600X. More threads = lower frequency = lower voltage.

The low-thread Prime95 test immediately revealed instability on the negative curve PBO2 btw. May be a good way to test it.
 

DimkaTsv

Great
Nov 7, 2021
156
23
95
3
The low-thread Prime95 test immediately revealed instability on the negative curve PBO2 btw. May be a good way to test it.
Personally hadn't found more precise way to notice PBO instabilities. 4 thread Prime95 will crash fast and consistently if you miss your PBO curve by 2 points or more.
If you miss by 1 point - instability will be inconsistent, co you may try to do 8 test run and make repeat by swapping it for 12 thread test and then 4 thread again.

Also you can do 1 thread test and move it via task manager from core to core by manually forcing it to cores.
But beware... if 4 threads fail - most likely 3 and 5 will fail too. 2 is 50/50. And reason for fail can be not your LAST thread core, but 1-st or 2-nd. Learnt it hard way
If you get thread fail on 1-3 test - and second thread fail few tests later it is 65% problem with core 2 (by performance) 30% core 1 and 5% that you missed core 3-4 curves.

For example
My core performance order is 5-1-2-4-3-6.
And current curves set to negative 30-20-20-25-25-30.
You can make cores 5 and 6 have most savage curves because they have highest curves by default.
If i set my cores 1-2 to 21-21, than on 4 thread Prime95 i can randomly get 1-2 thread crash or don't (but crash will happen really quick). 22-22 will guarantee me crash, as well as 21-22 or 20-22. Because second core is actually pretty damn important to make all communications between core working.
But threads that will fail, will not make sense as processor will instantly rearrange threads to fill gap.
And if you look per thread log in Prime95, it will not matter as any thread can fail.
Just wanna warn you. 4 thread instability is really damn precise test. It is SO easy to fail

So my test setup works like this.
  1. 12 threads for temp/frequency/voltage/and correlations
  2. 4 thread test for multicore stability (as easiest to fail)
  3. 4/2 thread test for temps
  4. 1 thread test with per core in Task manager for single core max frequency stability (well with small FFT it will be limited to 4.75 for me instead of 4.85, but still) on top of the curve. (Usually passes much more easily than 4 thread one, but can detect errors per core)
 

DimkaTsv

Great
Nov 7, 2021
156
23
95
3
Throw in 5-7 140mm case fans if your worried or want less noise.
Not everyone can afford to place 140mm fans, you know... 120mm maybe?

But you can just make through air flow with front-cpu(mid)-back fans... With 5600X it will be enough
Just set up curves a bit lower (fan ones i mean)... with 76W there aren't that much heat to dissipate
 
Nov 13, 2021
44
2
35
0
Personally hadn't found more precise way to notice PBO instabilities. 4 thread Prime95 will crash fast and consistently if you miss your PBO curve by 2 points or more.
If you miss by 1 point - instability will be inconsistent, co you may try to do 8 test run and make repeat by swapping it for 12 thread test and then 4 thread again.

Also you can do 1 thread test and move it via task manager from core to core by manually forcing it to cores.
But beware... if 4 threads fail - most likely 3 and 5 will fail too. 2 is 50/50. And reason for fail can be not your LAST thread core, but 1-st or 2-nd. Learnt it hard way
If you get thread fail on 1-3 test - and second thread fail few tests later it is 65% problem with core 2 (by performance) 30% core 1 and 5% that you missed core 3-4 curves.

For example
My core performance order is 5-1-2-4-3-6.
And current curves set to negative 30-20-20-25-25-30.
You can make cores 5 and 6 have most savage curves because they have highest curves by default.
If i set my cores 1-2 to 21-21, than on 4 thread Prime95 i can randomly get 1-2 thread crash or don't (but crash will happen really quick). 22-22 will guarantee me crash, as well as 21-22 or 20-22. Because second core is actually pretty damn important to make all communications between core working.
But threads that will fail, will not make sense as processor will instantly rearrange threads to fill gap.
And if you look per thread log in Prime95, it will not matter as any thread can fail.
Just wanna warn you. 4 thread instability is really damn precise test. It is SO easy to fail

So my test setup works like this.
  1. 12 threads for temp/frequency/voltage/and correlations
  2. 4 thread test for multicore stability (as easiest to fail)
  3. 4/2 thread test for temps
  4. 1 thread test with per core in Task manager for single core max frequency stability (well with small FFT it will be limited to 4.75 for me instead of 4.85, but still) on top of the curve. (Usually passes much more easily than 4 thread one, but can detect errors per core)
Nice, thank you for the detail here. Don't plan on trying any of this right away but will be a good reference for later.

If I'm failing on a -15 all core negative curve, won't I be failing on a mix between 20-30? Maybe I'm completely misunderstanding here.
 

DimkaTsv

Great
Nov 7, 2021
156
23
95
3
If I'm failing on a -15 all core negative curve, won't I be failing on a mix between 20-30? Maybe I'm completely misunderstanding here.
Ah, yes, surely you will. I just set an example from my values to set reference point for further explanation.
And why these values are working for me? is just i got lucky and my 5600x can be undervolted pretty good, that's why my curve values are so... Low? High? So it most likely will not work for others.



To set another reference point and deepen down information.
This is method i used to tighen up my per core curves to 2 step difference (actually tried to make 1 step, but factual values in the end are depends more on motherboard precision, so 2 step is more reliable)
Single core single thread Prime95 small FFT puts my core to 4.75 gHz approx. And at this frequency my cores should have 1.234V or bit more but no less than 1.230, or i risk a crash.
Value is found by checking 99% stable curve on 12 thread load by comparing per core, and putting cores - voltage of 100% load of which i know under max single thread load... Single thread load is actually trash for equalizing cores by curves, though... So you just getting a glance on values you can become unstable at.

How to find around these comparative VIDs to equalize them? Set rude curves to set generally stable undervolt (99% of time one)
For this unstable curve you just throw away lowest voltage and work to equalise others, but do it approximately to 0.012V (steps by 2 in curve optimizer, as 1 step is 0.006V), better not to do less... Or you can round up curve value to closest 8 or 5, but not too much ( so 26/27>25, 28/29>28 for example). And be sure to check per core stability on ones you changed.
For stable curve, you can try to tighten them up, but be careful with core 1 and 2, as they can be bit more picky and easy to destabilize.
So when i saw 1.264V VID on core, and i knew that commonly stable average is 1.234, i can suppose that lowering curve by additional 5 steps, as 5*0.006=0.030.
Tightening curves up to 1 step is not worth it, unless your motherboard able to give incredible precise voltages, so 2 steps generally will be more stable to find.

You can make rude curve by trying out all cores = same value of curve undervolt with steps of 5 then 3 or 2. This way you will find place, when weakest cores will fail.
Also be informed... That cores are divided by 3 voltage groups. 1+2, 3+4, 5+6 (and, likely, 7+8 for 5800x). 1+2 is lowest base curve voltages, and also easiest to overtune. 3+4 have higher curve voltages, usually by 0.030V more approx. And 5+6 have highest set curves as least performative cores (that's why they are last). And curve set is by around... 0.042-0.048V? higher than 3+4...
Values are approximate, but i wanted to make you point that placing blind bet on negative X/X (cores 1/2), then from these X values +5/+5, +10/+10 is a good place to start


P.S. Ofc personal experience will help you better, but i guess i managed to make a rude algorithm for you. Keep in mind, i am far from professional and all my notes were made from personal experience. Your experience may differ, so maybe using good old way of manual binning of values will be easier for you...
 
Last edited:

ASK THE COMMUNITY