mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2021-03-31, 07:56   #89
zurek
 
Mar 2021

2 Posts
Default

Thank you for words of encouragement and for insisting, that the problem is the stability of the system, not the Prime95. YES! You were right, of course.
When the overclocking is set precisely, the system is perfectly stable, for hours on Prime, all tests.
First, when writing the original post, I had the IceGiant cooler, not good enough. Now, I have custom water loop, using the largest radiator I know (9x140mm) and EK-Quantum Kinetic TBE 200 D5, using the AM4 optimized block EK-Quantum Magnitude AM4.

Well, this cooling system is capable of running Prime95 for hours with the CCD0 at 44.5x, and CCD1 at 44x. This is the maximum. Any higher overclocking gives CPU overheating in seconds after start of Prime95 smallFFT test. With other benchmark and stress tests, it can run at CCD0 46x, CCD1 45x stable.

This cooling can run at minimum waterflow, and with the radiator set passive (no fans) and the system is always stable even at the smallFFT test, while the water in and the water out sensors show about 25-26°C, sometimes the temp out is 27°C and then returns to 26°C. The room temperature is around 22°C. Increasing waterflow or making the radiator active is of no help, because the cooling medium stays always so cool and never heats up. It does not heat up even after several hours. The CPU stays around 79°C in the most demanding stress test (smallFFT). When idle, it stays around 28°C. Under normal load it has between 35-44°C. At the very beginning of the smallFFT torture test it goes up to 85°C for short time.

Hence the problem: the bottleneck is not the cooling loop and its components, but the weak point is the heat transmission from the CCDs to the CPU heat spreader (is it the right name for the metal CPU cover?) and from there to the water block. The CPU and its CCDs are so tiny, that the heat transfer seems very difficult. I think I have the most optimal parts, and I checked 3 times that the block is set well (the 0.5 jet worked the best), and that it is perfectly attached to the CPU in all places - the thermal paste marks are even all over the water block bottom plate after detaching from the CPU. The thermal paste came with the special water block, it says "thermal grizzly" on the label.

How to improve the heat transfer further on this AMD 5950x CPU, please?
zurek is offline   Reply With Quote
Old 2021-03-31, 08:18   #90
Viliam Furik
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

23·3·19 Posts
Default

Quote:
Originally Posted by zurek View Post
How to improve the heat transfer further on this AMD 5950x CPU, please?
I think there is no simple improvement available, only the extreme and risky ones, such as delidding the CPU - removing the heat spreader (yes, correct name, IHS - integrated HS is even more correct) and exposing the dies, and mounting a special cooler directly onto the dies. This is very risky, for multiple reasons: CPU might not survive the delidding process - dies might crack, pins might bend, the substrate of the CPU can break, basically all sorts of bad can happen; when mounting the cooler, you might accidentally mount it too tight, and crack the dies; you might mount it too lightly, and dies will fry (not very probable, if you do it at least a bit correctly, and don't run the small FFTs)

Another thing is to "lap" the CPU, i.e. sanding the IHS to expose the copper under the nickel plating, and lower the in-metal heat travel distance. With AMD's PGA (pins on the CPU) CPUs, it is extremely risky to do this, since pins bend easily.


Both methods are dangerous and will void your warranty!

Other than that, I think you have pretty much optimal cooling.
Viliam Furik is offline   Reply With Quote
Old 2021-03-31, 08:39   #91
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

3×5×53 Posts
Default

Thermal Grizzly is decent paste but there's different types. Kryonaut is the best paste they do, the only way to do better is to use conductonaut which is liquid metal (mildly scary as it's conductive and liquid so you can badly mess up the application). That's as far as you should go but even that is not really worth it IMO. Beyond that you can delid the CPU to replace the internal solder with liquid metal, but you'll also need to grind down the edges of the heatspreader as the solder is thick (which may impact block contact). Only extreme overclockers should be delidding especially now that solder is the norm (older CPUs used poor quality toothpaste, replacing that gave a large delta whereas replacing solder not so much).

Here's a video of the delid process for AM4 for fun: https://www.youtube.com/watch?v=WXbCdGENp5I
M344587487 is online now   Reply With Quote
Reply

Thread Tools


All times are UTC. The time now is 20:22.

Tue May 11 20:22:52 UTC 2021 up 33 days, 15:03, 1 user, load averages: 1.93, 2.47, 2.26

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

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.
A copy of the license is included in the FAQ.