CPU Temps

Message boards : Questions and problems : CPU Temps
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
jrich046

Send message
Joined: 19 Apr 20
Posts: 10
Message 98302 - Posted: 3 May 2020, 14:51:11 UTC

Hello Everyone,

Just finished a new pc build and installed BOINC to start running Rosetta@Home. I wanted to check the cpu and gpu temps so I installed Open Hardware Monitor.

BOINC 7.16.5 on WIndows 10
AMD Ryzen 5 1600 AF using the stock cooler.
Nividia 1660 super

While running Rosetta on 100% of the CPUs I was getting temps around 86 C. Looking at AMD it states the cpu's max temp is 95 C but just about everything I have read on various forums suggest that 86 C is pretty high. I then then ran Einstein@Home on 100% of the CPUs and the max temperature is 65 C. Watching the open hardware monitor software while BOINC is running shows that for both projects the CPUs are running at 100%. The GPU under Einstein does run warmer than Rosetta, I guess I am just surprised that while both run the CPU at 100%, there can be a 20 C difference in temperature. Anyone have any thoughts on the large temperature difference?

Thanks,
jrich
ID: 98302 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 98304 - Posted: 3 May 2020, 16:39:06 UTC - in response to Message 98302.  
Last modified: 3 May 2020, 16:39:34 UTC

It's due to the difference in what they search for, which programming language they used, which language their data files use and whether or not the application is using SSE optimization or not. It's a good example of why you cannot compare projects running under BOINC with each other.
ID: 98304 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98305 - Posted: 3 May 2020, 17:44:53 UTC

Wait until you run a project using AVX2 instructions.
Your CPU will throttle down in frequency with those, due to the load.
ID: 98305 · Report as offensive
jrich046

Send message
Joined: 19 Apr 20
Posts: 10
Message 98311 - Posted: 3 May 2020, 21:49:06 UTC - in response to Message 98304.  

Do you feel that running the CPU at 85C for extended periods is okay? Would you suggest I try to lover those temps with additional case fans or aftermarket CPU cooler?
ID: 98311 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 98313 - Posted: 3 May 2020, 23:45:18 UTC - in response to Message 98311.  

An aftermarket CPU cooler is always better than the fan AMD supplies. And here it doesn't matter if that's air- or water cooling.
ID: 98313 · Report as offensive
jrich046

Send message
Joined: 19 Apr 20
Posts: 10
Message 98315 - Posted: 4 May 2020, 3:53:43 UTC - in response to Message 98313.  

Sounds good, thanks
ID: 98315 · Report as offensive
Nick Name

Send message
Joined: 14 Aug 19
Posts: 55
United States
Message 98319 - Posted: 4 May 2020, 6:21:52 UTC

In the meantime, if your CPU fan isn't running fast enough you might look to change that in the BIOS. Most of the boards I've seen have the fan speeds optimized for lower noise, not performance.
Team USA forum
Follow us on Twitter
Help us #crunchforcures!
ID: 98319 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98337 - Posted: 5 May 2020, 17:35:50 UTC - in response to Message 98319.  

In the meantime, if your CPU fan isn't running fast enough you might look to change that in the BIOS. Most of the boards I've seen have the fan speeds optimized for lower noise, not performance.


Another thing that can be done in the bios, is lower the CPU voltage by a few millivolts (sometimes lowering by 0.010V to 0.020V from stock), might get you 5 or more degrees C lower running temps, without needing to sacrifice anything in speed.
85C is quite high. I would aim for between 60 to 75C running temps.
Some great coolers come from China, that are quite affordable ($15-35).
The ones with like 4 heat pipes, and a large heat sink (about 4to 5 SQ" in size, cooled by a 80mm fan).
ID: 98337 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 98341 - Posted: 5 May 2020, 18:18:50 UTC - in response to Message 98337.  

Another thing that can be done in the bios, is lower the CPU voltage by a few millivolts (sometimes lowering by 0.010V to 0.020V from stock), might get you 5 or more degrees C lower running temps, without needing to sacrifice anything in speed.
Don't undervolt via the BIOS, especially on AMD Ryzen CPUs. Better is to use Ryzen Master to do that.

When you undervolt via the BIOS, you lose boosting performance.
ID: 98341 · Report as offensive
Profile Keith Myers
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 17 Nov 16
Posts: 863
United States
Message 98344 - Posted: 5 May 2020, 19:00:35 UTC - in response to Message 98341.  

Another thing that can be done in the bios, is lower the CPU voltage by a few millivolts (sometimes lowering by 0.010V to 0.020V from stock), might get you 5 or more degrees C lower running temps, without needing to sacrifice anything in speed.
Don't undervolt via the BIOS, especially on AMD Ryzen CPUs. Better is to use Ryzen Master to do that.

When you undervolt via the BIOS, you lose boosting performance.

There is nothing wrong with undervolting in the BIOS. Especially if you run an all-core overclock and turn off boosting entirely. Static clocks that don't move around and 10-15° C. less temperatures.
ID: 98344 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 98346 - Posted: 5 May 2020, 20:58:09 UTC - in response to Message 98344.  
Last modified: 5 May 2020, 20:59:59 UTC

The problem with undervolting through the BIOS is that you will lose performance even if you do it in small increments. Your CPU won't boost to the high values anymore, it may even get stuck on a set speed (as it did in my occasion).
In such case it's better to adjust the Power Package Tracking or PPT value, as this is always set higher than the TDP for your processor. Or adjust values via Ryzen Master, because you don't need to reboot, you see your changes immediately, you can set different profiles and easily switch between them and easier to quickly go back to the default settings if something goes wrong.

Now, no need to believe me, and I know you did the undervolting for your own system(s), but then say so that it's something you don't mind doing. Plenty of people out there who advise otherwise, because of that performance loss.

https://www.reddit.com/r/Amd/comments/ce4gm4/psa_stop_nerfing_your_cpu_by_undervolting_stop/
https://www.reddit.com/r/Amd/comments/ceakbs/if_you_want_to_save_powerreduce_thermals_reduce/
https://sff.life/how-to-undervolt-ryzen-cpu/
ID: 98346 · Report as offensive
Profile Keith Myers
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 17 Nov 16
Posts: 863
United States
Message 98351 - Posted: 6 May 2020, 17:04:20 UTC - in response to Message 98346.  
Last modified: 6 May 2020, 17:05:07 UTC

I don't run Windows, so no Ryzen Master for me, nor any other conventional utility. I can run an all core overclock of 4200Mhz on my 3950X at 1.22V. Auto tries running at 1.35V and is only able to achieve 4000-4050 all core clocks. So I am not undervolting by a paltry few tens of mV, I undervolt by over 100mV. In the case of a 24/7 distributed BOINC load, it makes sense to avoid any boosting or Auto setting and just select a manual fixed multiplier at reduced voltage. Better power consumption and better temps. My performance is actually BETTER with reduced voltage.
ID: 98351 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98363 - Posted: 7 May 2020, 17:32:08 UTC - in response to Message 98351.  
Last modified: 7 May 2020, 17:36:11 UTC

I don't run Windows, so no Ryzen Master for me, nor any other conventional utility. I can run an all core overclock of 4200Mhz on my 3950X at 1.22V. Auto tries running at 1.35V and is only able to achieve 4000-4050 all core clocks. So I am not undervolting by a paltry few tens of mV, I undervolt by over 100mV. In the case of a 24/7 distributed BOINC load, it makes sense to avoid any boosting or Auto setting and just select a manual fixed multiplier at reduced voltage. Better power consumption and better temps. My performance is actually BETTER with reduced voltage.

Thanks! This is informative!
My Bios is set to use 1,38V on PBO, using a triple fan cooler.
I now feel safer to probably lower it to 1,30V.

On the other hand, I've seen projects run at 4,1Ghz on all cores, and other projects go as low as 3,6Ghz. I fear setting the voltage to fixed would produce errors on those projects?

Bios voltage only affects turbo frequency. When it's set too low, the CPU might crash. But as long as it's not too low or too high, lower voltage should help make the CPU run cooler, which should help in allowing higher boost frequencies.
ID: 98363 · Report as offensive
Profile Keith Myers
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 17 Nov 16
Posts: 863
United States
Message 98364 - Posted: 7 May 2020, 20:17:36 UTC

Either find the project whose cpu tasks work the computer the hardest and find out what is the lowest voltage that is stable for that project.
Or put the host under a synthetic worst case workload like small FFT's with Prime95, y-cruncher, Intel Burntest or OCCT with small dataset and find the lowest stable fixed clock with the lowest voltage.
The problem with running Auto or PBO is the cpu is always trying to maintain 1 or 2 cores at the highest clocks which requires a lot of voltage. That high voltage is applied to all cores which leads to higher temps and eventual lower overall core clocks for all cores because of thermal or power throttling. This is great for maintaining a high clock speed for one or two process threads running a game and producing the highest framerates.

But OTOH, if the computer is mainly spending most of its time crunching cpu tasks for BOINC, then it is actually better for the workloads to maintain a fixed clock frequency, irrespective of the workload and not have the clocks constantly moving around on the cores as the cpu moves the loads to distribute the heat.

You can actually get even better performance if you do per-CCX overclocking on the dual die Ryzen's but that involves some major diving into the BIOS features to configure for that. That allows you to run one Vcore voltage and fixed clock for one die and a completely different voltage and clock for the other die. Useful for 3900X cpus with one good binned die and one bad or mediocre binned die. Even with a 3950X with two good binned die there are always going to be minor differences in how each piece of silicon performs.
ID: 98364 · Report as offensive
Profile Keith Myers
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 17 Nov 16
Posts: 863
United States
Message 98365 - Posted: 7 May 2020, 20:20:07 UTC - in response to Message 98363.  

On the other hand, I've seen projects run at 4,1Ghz on all cores, and other projects go as low as 3,6Ghz. I fear setting the voltage to fixed would produce errors on those projects?

Why suffer low clocks on one project and high clocks on another? Why don't you split the difference and find the fixed core clock and voltages that run all projects on all cores.
ID: 98365 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98371 - Posted: 8 May 2020, 12:19:33 UTC - in response to Message 98365.  

On the other hand, I've seen projects run at 4,1Ghz on all cores, and other projects go as low as 3,6Ghz. I fear setting the voltage to fixed would produce errors on those projects?

Why suffer low clocks on one project and high clocks on another? Why don't you split the difference and find the fixed core clock and voltages that run all projects on all cores.


I'm not sure, but I think that some projects use avx instructions, which tax the CPU higher. Even so, the CPU doesn't increase more than 5C over them.



Linux lscpu' shows 1.1V during crunching. I figured the bios values are when the CPU is mostly passive (deactivated cores), and voltages drop with all cores active.

I've tried running boinc on all cores PBO and on each of the 4 dies of my 3950x. But the numbers jump around all the time, one time favoring higher cpu speeds for one, another time for the other.
I would guess it depends on what WU is assigned to what core.

Manual overclocking (to eg 4.0 or 4.1 GHz) could result in failed WUs on WUs that tax the CPU more. On the other hand, the frequency could be lower than optimal with lighter load WUs.
I also think that the auto adjustment only makes minor adjustments (in the likes of a few hundred MHz at most), and thus the power performance penalty should be minimal.
Unlike projects where CPU load could vary from 0% to 100% on a regular basis, those suffer from tremendous power and performance losses.
As soon as a WU is loaded, the energy performance penalty should be small.

It's only my working theory so far...

I will play around with the CPU voltages for a bit.

Any experience with DDR4 voltages?
The xmp profile auto loads 1.35V for the full 3200 or 3600Mhz, and 1.2V for 21XXMhz. I was wondering if it would work ok running at 1.30V instead?
If that's the case, CPU core voltages could be lowered by the same 0.05V, resulting in a cooler running CPU, and more power routed to the CPU?

Also a working theory, but I'm kind of scared to adjust the wrong parameters. Already once before I got in an infinite pre bios boot loop, due to adjusting the CPU voltages too low.
ID: 98371 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98404 - Posted: 9 May 2020, 16:48:21 UTC
Last modified: 9 May 2020, 16:59:15 UTC

Been playing around a bit with BIOS settings.
On my 3950x, I'm still running at 3,7-3,8Ghz on all cores, but I was able to lower the CPU VCORE voltage down to 1.116V without any noticeable drop in performance. The strange thing is, when I set it to 1,25V or higher, the performance on all cores drops drastically (like 2,7-3,3Ghz).

I did also adjust the Vcore SOC voltage. I was able to drop it to 0.925V (from 1.100V), resulting in 15C lower temps.
I think for some reason, this messed up my XMP, as the DDR RAM now reverts to 1,2V (21xxMhz).
Resetting it all to auto, doesn't seem to fix it.

Not sure if the RAM speed has a super drastic performance effect when crunching Boinc, since most of the crunching data packets, are rather small, and a lot is held in the CPU L-cache; and even at 3200Mhz the ram can provide data faster than the CPU can crunch it... The Infinity Fabric is another thing I've been struggling with.
I think my Asus TUF motherboard is riddled with bugs!
It'll be replaced soon.
ID: 98404 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98444 - Posted: 12 May 2020, 12:28:18 UTC

On a ryzen 9, it's indeed better to manually overclock.
Not only does it use way less Watts when constantly under load, it also goes higher in frequency.

I spent a good day trying to configure one ryzen, then once I grasped the concept, configured the other in minutes.

It's not beneficial to run auto frequencies, but you'll need to run them to find your optimal CPU VCORE voltage, and the VCORE SOC voltages.
These 2 run in tandem, so that adjusting one, depends on the values of the other.
Not only that, but there's a third factor that comes into play, PBO max wattage. A good setting with high boost frequency, that works on a max wattage of eg:95W, can be lots slower at 90W, or even 100W.
This because PBO works in steps.
To have to configure each voltage, benchmark, reboot, change another voltage, benchmark, reboot, change PBO Watts, bench, reboot, is extremely tedious.

But once you found the CPU VCORE and VCORE SOC optimal voltage settings (lower with worse cooling, higher with better cooling), the most common setting is lower PBO to what your cooler can handle.
In my case, my single 120mm water cooler can handle about 90W in that small case, so I set my PBO wattage to 90, and lower my max throttle temp to 5 degrees below throttling. In my situation 85C (since the board tmp sensors register 5 C lower than the internal CPU temp sensors, which should thermal throttle around 95C).

Then set a fixed CPU speed. In the small case, I could only get it working to 39,50Ghz on all cores, but lowering it by 50Mhz, lowered power at the wall by almost 25W. It now runs at 205W total consumption at 3,9Ghz.

I don't know what people do to get 4,2Ghz on all cores, but I'm a far cry off of it, including on my 3950x, which only gets up to 3,975Mhz on all cores with a triple fan water cooler that feels nearly cold under full CPU load (probably 30-40C).
ID: 98444 · Report as offensive
Ian&Steve C.

Send message
Joined: 24 Dec 19
Posts: 228
United States
Message 98451 - Posted: 12 May 2020, 16:42:47 UTC - in response to Message 98444.  
Last modified: 12 May 2020, 16:45:18 UTC

I don't know what people do to get 4,2Ghz on all cores, but I'm a far cry off of it, including on my 3950x, which only gets up to 3,975Mhz on all cores with a triple fan water cooler that feels nearly cold under full CPU load (probably 30-40C).


Not all chips are created equal. And Keith runs custom water cooling, which performs significantly better than an AIO cooler.

He also runs very high end low latency memory (which is key for Ryzen OC), a high end motherboard and has likely spent more time tweaking all the different settings than you have.
ID: 98451 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 98467 - Posted: 13 May 2020, 4:24:56 UTC
Last modified: 13 May 2020, 4:29:21 UTC

Either I have had the worst binned CPU, or the worst motherboard.
I presume the latter. It was only $150 (vs $200 for most other AM4 motherboards supporting PCIE 4.0), and this might be the reason why.
I sent my return notice, and the seller already reimbursed me, without me even having returned the board.
That speaks volumes!
Gigabyte GPUs have been most reliable to me. Out of most GPUs, the only ones working,are Gigabytes. All the rest either failed, DOA, or broke.
But I can't say the same about their motherboards!
The most reliable motherboard I currently have, is the MSI MPG for intel 7th gen. Never failed me! Bios works like it should, GPUs just simply WORK!

This Gigabyte X570 UD, I have had nothing but problems with it!
It has come so far, that it won't even boot when I try to save the bios.
All I can now do is run it completely stock (like is, after a bios reset).

The board is going to be returned!
ID: 98467 · Report as offensive
1 · 2 · Next

Message boards : Questions and problems : CPU Temps

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.