[SOLVED] 3990x Gets 95C under load

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
Hiya,
I have:
3990x
Gibabyte Auros Xtreme
Edit Noctua NH-U14S TR4-SP3 with 2 140mm fans, 1500rpm and 1200rpm
Corsair Vengence 3600Mhz LPX 18-22-22-42 256GB RAM at 1.35V and SOC 1.1V

I had about 70-80C previously then I upgraded the AMD chipset drivers, previously I used from gigabytes homepage. Then I got MUCH higher temps and also higher clock speeds.
Under heavy load I now get 90-95C for CPU and 90C for VRM MOS. Even though it is so hot the CPU is mostly at 3.8GHz which I think is way to high for this CPU, it dips shortly to 3.3GHz and then jumps up again. Previously the CPU would be at 2.9GHZ all the time under 100% load.
Case temp is 30C and room temp is 27C. I got 7 140mm fans to blow into the case.

Now, what are safe temps to run with? And what should I do to get the temp down. Ideally I would like it to clock down the CPU more when it gets hot, it seems as if it is not trying any more.

I guess I could switch to watercooling, but are they not crazy loud? Some use 2500rpm fans.

Any ideas?
Best,
Andreas
 
Last edited:
Solution
OK. That means, changing the powerplan did not change anything. One last hint: try undervolting the CPU ( negative offset for core-voltage, start trial with -0.05, and depending of stability go up to -0.1 with little steps).

Otherwise I can only recommand a custom-watercooling. f.i. EKWB does offer sets for trx40-socket.

I do not have any other idea, and therefore I am off now. good luck.

Some reviews:

https://www.anandtech.com/show/15483/amd-threadripper-3990x-review/2
https://www.overclock3d.net/reviews...n_threadripper_3990x_64_core_monster_review/2...

Phaaze88

Titan
Ambassador
I'm not familiar with the Threadripper platform, but if I had to guess, the update must've activated some overclocking feature, or it actually 'fixed' your cpu and has it operating normally.
You could try setting bios back to default, or reset CMOS? You should also check to see if the power plan is still on balanced.

I guess I could switch to watercooling, but are they not crazy loud? Some use 2500rpm fans.
There's high performance coolers. Yes, the fans can get loud - if you let them.

Then there's the silence coolers - fan speeds(max) are crippled on purpose to maintain the silent profile. The downside is that what you get out of the box is pretty much it; you can't get any extra cooling out of the them because the fan horsepower isn't there. That also makes them poor options for intakes, except when they are placed against open mesh. They are also weaker than performance coolers in the same class because of this. Terrible options for overclockers, but this doesn't apply to you.

You could just get a performance model and control the fan speeds yourself. Better you set your own curves instead of leaving it on auto.
If for some reason you'd rather not do that, there's the silent models like Alphacool's Eisbaer 360/420 and Corsair's H150i Pro... IF they fit in your chassis, that is.
 

Karadjgne

Titan
Ambassador
Wait. Do you have the standard NH-U14S, or the redesigned NH-U14S TR4-SP3? That makes a huge difference.

Paste on a threadripper. Spread it manually, do not use the pea drop method, you do not get the correct coverage. The dies are very close to edges/corners, not a monolithic center single die. That can make a difference.

Gigabyte and others have chipset drivers that mostly cover the motherboard chipsets, like the PCH, Sata controller, Lan, audio etc. The also have bios revisions, mostly for ram compatability, but include some cpu tweaks and further compatability with the chipsets and OS. AMD has its own set of drivers specifically for the cpu, they include memory controller tweaks, infinity fabric updates etc etc.

So having both, especially after a MS OS Major, quarterly update is usually something that can set the cpu free, whereas prior it was stifled in some way. Which could be what happened, or just as likely the OS got availability to access other criteria previously denied, and can now freely boost the cpu.

Either way, I'd be more interested in finding out exactly what happened, what settings got changed, what it all implies before spending out on additional cooling. You could just as easily limit the boost / power limits to only allow upto @ 3.4GHz and get better performance across the board, for free.

Then again, if you have the standard NH-U14S, I'd seriously advise changing coolers anyway. The base on the standard coolers is nowhere near large enough to compensate for the size of the TR IHS.
 

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
Thanks for the help guys
Yes this is using the Noctua NH-U14S TR4-SP3 so is the correct cooler.
I did think that maybe I applied too much thermal paste, I did apply it with drops, normally I would create a thin layer on the CPU.
So before I updated the chipset drivers the CPU never got hotter than around 80C. To me the biggest problem seem to be that the CPU just does not clock down with the new drivers. I mean you would need pretty insane cooling to be able to run 3.8Ghz on this computer and that is where the clockspeed will stay under full load most of the time.

Another change I did from before I increased the RAM speed to 3533Mhz, it was 3400. I set the SOC voltage to 1.1V. I think I had that before as well.

Now in idle Task Manager reports the speed is 3.95Ghz. CPU temp is aroud 80C. Case temp is 32C and room 27C.

All too all a bit confusing.
So just to summarize what changed
  • I got the new chipset drivers
  • I increased the RAM speed to 3533Mhz and set the SOC to 1.1V Not sure what it was before because it reset the bios when booting so lost all the old settings
In the overclocking I tested to change the "temp throttle value" to 80C. That seems to work a bit, seems like when it hits 80C then it will dip a bit, but only for a short while. However it does not seem to do much when the CPU is under load.
 

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
Edit. I will get Ryzen Master and check the PBO settings.

I use the AMD Ryzen processor power plan. Maybe I should set it to 90% max instead?
Ok, I tried putting it at 95%. It set the CPU speed to a fixed 2.7GHz which is of course a bit on the low side.
I had 2 crashes yesterday, under load so better to play it safe until I can have it not burn up.
 
Last edited:

alexbirdie

Respectable
1.Could be wrong AMD-drivers ? do you already have windows 2020/2004 version or 1909 or 1903 ?

Found on gigabyte homepage different version of the amd-drivers for your MB, depending on your windows 10-version.

2. Powerplan changed by installation of the amd drivers ?
 
When applying thermal paste on AMD platforms, you really don't need to apply too little. If anything having too much (within reason) is perfectly fine and tests have proved that this will not increase temps. However risking too little paste will definitely make the chip run hotter.

There's a problem right now with Windows and AMD Chipset Drivers. if you install updated AMD Chipset Drives ON TOP of the old ones. You will have downclocking issues.

I had this exact issue on my 3600, once I reinstalled windows completely fresh and installed the latest chipset drivers my CPU now idles perfectly and cores sleep properly too (I.E Enter C6 state)

So in short, simply do not update chipset drivers at all unless you have having issues. (not sure if this works but) when swapping to updated drivers, 1. Manually Uninstall the old AMD drivers 2. Make sure you delete the AMD folders in the root of your C drive.

Oh and don't touch the AMD power plans, they are set exactly where they need to be. There's a lot of extra code in these power plans behind the scenes that we can't manipulate. And changing anything CPU related in the power plan won't fix anything. You are better off using the standard windows power plans if you want to manipulate the power plans yourself.
 
Thanks for the help guys
Yes this is using the Noctua NH-U14S TR4-SP3 so is the correct cooler.
I did think that maybe I applied too much thermal paste, I did apply it with drops, normally I would create a thin layer on the CPU.
So before I updated the chipset drivers the CPU never got hotter than around 80C. To me the biggest problem seem to be that the CPU just does not clock down with the new drivers. I mean you would need pretty insane cooling to be able to run 3.8Ghz on this computer and that is where the clockspeed will stay under full load most of the time.

Another change I did from before I increased the RAM speed to 3533Mhz, it was 3400. I set the SOC voltage to 1.1V. I think I had that before as well.

Now in idle Task Manager reports the speed is 3.95Ghz. CPU temp is aroud 80C. Case temp is 32C and room 27C.

All too all a bit confusing.
So just to summarize what changed
  • I got the new chipset drivers
  • I increased the RAM speed to 3533Mhz and set the SOC to 1.1V Not sure what it was before because it reset the bios when booting so lost all the old settings
In the overclocking I tested to change the "temp throttle value" to 80C. That seems to work a bit, seems like when it hits 80C then it will dip a bit, but only for a short while. However it does not seem to do much when the CPU is under load.

Making a 2nd post to help organize things a bit lol:

For memory, I would highly recommend for now, just clearning CMOS, run the RAM at it's default 2133 JEDEC standard and stress and see what temperatures are.

If you are having crashes, I'm thinking your memory controller voltage is probably not high enough. Keep in mind you are running 3400mhz on 256GB of RAM, that's a TON of memory.

Here's my recommendation (check my overclocking RAM guides below for more assistance):

  1. Run a SOC voltage of 1.15v and see if that works at all. (I would recommend lowering it back to 1.1v or 1.05v after testing is done if it turns out this is a non-issue.)
  2. Install the Ryzen DRAM Calculator and check what ProcODT value you need to run from the calculator.
  3. What's your XMP profile rated for for RAM Frequency? For that massive amount of RAM, I wouldn't go higher than 3200mhz unless you want to fiddle with timings and voltages.
  4. Run memtest HCI until about 200%.
Optional: Disable XMP and manually put in your RAM Frequency, Timings, DRAM Voltage, and SOC voltage. Ryzen has been finicky with XMP in the past (due to it being an Intel thing), and if you didn't know, the motherboard will usually apply some custom voltage parameters when XMP is enabled. Disabling it makes the motherboard run default voltage parameters on everything. (it is super rare these days for XMP to have issues but you never know...)
 
Last edited:

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
So the Auros app told me that "there is updated Firmware". This was a bit odd since I do run the latest bios 4F and I do not see any newer. However I updated with this and did not change any bios settings at all.
Now it instantly clocks down to 2.9Ghz when I get 100% load. After 20 minutes BurninTest I get 68C which is much better than 95C. It has not increased the last 10 minutes so seems quite stable.

The memory currently is at 2133Mhz. I will try increasing this and SOC and see what happens with the temperatures then.

The memory is Corsair 3600 18-22-22-42
https://www.scan.co.uk/products/256...0-3600-non-ecc-unbuffered-cas-18-22-22-42-xmp

Cheers,
Andreas
 
Last edited:

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
<1.Could be wrong AMD-drivers ? do you already have windows 2020/2004 version or 1909 or 1903 ?
After Gigabyte installeed its firmware it is working better now. I have not updated to 2004 yet, the build is 18363 of Windows.

<Found on gigabyte homepage different version of the amd-drivers for your MB, depending on your <windows 10-version.
I will check it out I think I will wait a bit with installing new drivers since for the moment it is better than before.

<2. Powerplan changed by installation of the amd drivers ?
I just use the AMD default power plan now.
 

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
I increased the Memory speed to 3200Mhz. Increased the SOC voltage to 1.15V and set DRAM to 1.35V. I can probably reduce SOC to 1.1V when this has proved to be stable. I think this will be stable what I have now.

I did a 30min CPU stress test and the CPU temp increased to 72/73C.It goes back and forward between them for seems stable. The CPU speed was around 2.9Ghz all the time, sometimes 2.85Ghz.

When I'm idle the temp is jumping between 64C and 74C, in rare cases it can spike up to 82C but then it drops fast. The speed varies between 3.95 to 4.03GHz pretty much all the time which must be why the idle temp is still so high. Some are at 3.3Ghz and 3.6Ghz while the highest cores seem to be around 4.180.

Do these temps seem normal? Is there a way to reduce the speed when I'm not really using the computer. Seems a bit overkill to have 64 cores at 4Ghz when I'm not doing much. I could understand if a few cores were at 4Ghz, but all 64?

Cheers,
Andreas
 
Last edited:

alexbirdie

Respectable
I am using 1usmus ryzen universal plan and it is working great for me.

Sorry, I do not have any details knowhow about energy-saving-plan, but looks like as if some settings in your energy-saving-plan are causing the high idle-values.

Hopefully some forum members more experienced with those plans can help.
 
Last edited:

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
I have to tell an interesting story about this computer. I started off a complicated render in Houdini (a VFX program). It was a complicated smoke explosion with up to 150 million voxels. What was really cool was that I could render this and at the same time play the game Control and I would not notice any framedrops at all. I have NEVER been able to render and play complex games before while playing.

A bit concerning was that the temp was really quite high. Around 82C.

Best,
Andreas
 

alexbirdie

Respectable
In my powerplan the settings for min/max energy are both 100%.

My 3950x does have about 40 degree in idle and goes up to about 70 degrees with full AIDA64-stresstest (without avx and avx2). In ryzen master , if idle, I am seeing all cores sleeping or frequency at about 400-600 Mhz for the few threads , which are working for ryzen master display or working with browser.

I am sorry, that I cannot insert screenshots from my ryzenmaster and the powerplan-settings ( only a link could be inserted for a image, but where should I link it to?).
 

alexbirdie

Respectable
Ryzen master. But if it is idle, 1400 is still to high. Try to enter 100/100 into your power plan. simply remember the current values and then you can change back, if neccessary. you cannot do anything wrong, because you can revert it.
 

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
Ok, I entered 100/100. I do not really notice any difference. The voltages and frequences seems to be pretty much the same. They jump around quite a bit..
Peak core is 1.28-1.3V and Average core is about 0.87-0.95V Seem to be the same in both of them.
 
Last edited:

andreasoberg

Distinguished
Aug 28, 2009
21
0
18,510
Idle temp is around 65C at the moment, sometimes going up to 70C
When playing control the speed is about 4Ghz and the temps are around 85C, it even jumped up to 92C for a little while, so that is a bit warmer than I would ideally like to have it. I just wish it would not clock the CPU so highly.
 

alexbirdie

Respectable
OK. That means, changing the powerplan did not change anything. One last hint: try undervolting the CPU ( negative offset for core-voltage, start trial with -0.05, and depending of stability go up to -0.1 with little steps).

Otherwise I can only recommand a custom-watercooling. f.i. EKWB does offer sets for trx40-socket.

I do not have any other idea, and therefore I am off now. good luck.

Some reviews:

https://www.anandtech.com/show/15483/amd-threadripper-3990x-review/2
https://www.overclock3d.net/reviews...n_threadripper_3990x_64_core_monster_review/2
https://community.microcenter.com/discussion/2279/amd-ryzen-threadripper-3990x-review#Comment_7062
 
Last edited:
Solution