Double Ryzen cooler hack for REALLY cheap and desperate people

One of the great cost advantages of AMD's Ryzen CPU's is they all include a functional cooler.  Truth be told, almost any decent entry-level aftermarket cooler like the Coolermaster Hyper 212 EVO or even the $15 DeepCool Gammax 200T will match or outperform the stock Ryzen coolers here.  That being said, the 65 watt TDP Wraith Stealth cooler, which is the cheapest of the stack has a reputation for not leaving any thermal wiggle room beyond stock clock settings.  So, the Stealth is often left unused and floating around in the parts bin.  In this experiment, I will see if it is possible to stack two of these on top of each other to improve thermal performance.




Truth be told, the only time this setup would be economically viable is if you can find a free Wraith Stealth cooler either left over in your parts bin or given to you out of sympathy by a fellow enthusiast because, as mentioned above, just about any other budget after-market solution will outperform this hack, but for the cheapest of the cheap among us, we should find out if there is any improvement to be gained.




The much improved AMD Wraith Spire cooler looks to be very similar to the Stealth except the aluminum core is about twice as thick.  It can be assumed that there is no way a sandwiched arrangement of nearly the same cooler design, fin arrangement, and total mass can possibly out-perform its solid counterpart, but this makes the Wraith Spire the best option for comparing against.


The best value CPU in the Ryzen lineup at the time of writing is the Ryzen 5 3600 which comes with a Wraith Stealth cooler.  Any of the Ryzen 5 CPU's would be a likely use case for this hack, and I will test on the 3600.



The test system will be an MSI B450 A PRO motherboard which I was able to flash to the newest BIOS with BIOS flashback feature that does not require an older generation CPU.  Just the motherboard, power supply, and a USB flash drive.



Even though I am not accustomed to doing this type of comparative testing on the regular, I will try to use some reasonably consistent methods and parameters for my testing, so I will be using the same thermal compound for all tests. This means I need to remove the factory applied thermal paste from all of these coolers because I have no clue what they used.  I'm sure it's decent stuff though.  The "credit card" method works quite well to get the bulk of the thermal compound off followed by 99% denatured alcohol and some paper towels.


Gelid GC-Extreme was my thermal paste of choice previously, but I was not sure if I still had enough to do these multiple re-applications for testing, so I will be using some new KingPin Cooling KPx compound.  People who know better than I do seem to think this compound works as good as any other top tier TIM, and to be honest, it's cheaper than Gelid GC-Extreme so I went with it.

Because I only have two Wraith Stealth coolers, I would need to mount and run my tests on one of them to get my baseline data before hacking the cooler up.

Spread method is used for thermal paste application. . . I have no real opinions on which way is best, but the little spread tool was included in the paste, so I felt compelled to use it.



The installation was uneventful and the system booted without any problems.  With baseline testing out of the way, it's time to remove the cooler, clean it off again, and do some hacking.  Here we have two Wraith Stealth coolers that I will attempt to merge into my Double Stealth Stack Awesome cooler.



First, I remove the plastic shroud from the stock fan assembly.  This model was just friction fit with built-in clips.  Others seem to have screws in various configurations.


With the shroud removed, the fan can be easily unscrewed from the aluminum core.



Apparently, AMD uses AVC and Foxconn fans in their stock coolers.  This one is AVC. . . the other is Foxconn.  I'm sticking with this fan because the initial tests on the stock Stealth were done with this fan.

In order to sandwich the two cores together with good clamping pressure, I will need to use longer length fan mounting screws that pass all the way through one core and into the bottom core.  I decided to try some #4 Phillips Pan Head tapping screws from McMaster Carr.  These are 1-1/2 inch length.  The black screws are the original stock ones.


These screws fit the existing taps perfectly, so I am confident they will work in the new assembly.


Alternately, we can use an M3 tap and use M3 machine screws which I had on hand.  I will think on it a bit and wait and see which method to choose when it's time to make the final assembly.




The next area to address is the CPU bracket mounts.  These will interfere when trying to mate two cores together, so we will need to remove them.


A dremel with a metal cutoff wheel  can be used to zap these off very easily.




With the CPU mount screws removed, fan-mounting holes on the middle core need to be drilled out so the fan screws will pass through unobstructed.


The aluminum is very soft, and these drill out very easily.


The core is now ready to assemble.




I decided to go with the tapped M3 screws option for the fan mount screws.  The M3 screws that I had were shorter than the #4 self tapping screws which were a bit longer than necessary.  My thinking was I could try the M3 threads first, and if those threads fail, I can always throw the self tapping screws in later and still have enough material for the screws to bite on.



Next, a quick test fit to see if everything is lining up as it should.


And everything looks to be lining up just fine.  We have good clamping pressure with the machine screws, and the tapped threads seem to be holding ok.


Now to apply some thermal paste between the two cores before final assembly.



With thermal compound applied, time for final assembly.  Everything should be downhill from here.


I have some squeeze out at the seam indicating I have somewhat even paste distribution between the cores!


All that is left is to reinstall the fan shroud, and humans win!


We have now completed the double-decker, poor-desperate-person Wraith cooler.  Time to see how the performance stacks up.  Please note, I tested the stock Wraith Spire and the stock Wraith Stealth coolers in the same session.  But, the results for the final product here were performed the next evening so my test results should only be used as a general reference because variables like ambient temperature differences were not taken into account.  I hope a tech-tuber with better methodology can follow up.

Since precision boost seems to scale with temperature on the Zen 2 chips, I decided to do some tests at stock settings.  At stock AUTO settings, the CPU should automatically increase operating frequency as temperatures decrease.  Here are the results from the stock Auto settings with XMP profile enabled on my 3200mhz CL16 RAM.  I am running the CPU fan at 100% for all of these tests and the side panel has been removed from the computer making it essentially an open air bench.






Next, I set a manual all core overclock to 3.95Ghz and 1.2375V and ran the same tests.  This was the highest overclock I could set that would actually run on the stock Stealth cooler. . . well, I had to stop the run prior to full temperature stability on the Wraith Stealth, so thermals were getting out of hand as expected, but I think we are close enough to get a picture of the performance differences.  I like science, but I would also like to run this CPU for a daily driver after testing, so I stopped things a little bit short before running the temperatures much above 90C!



Despite the holes in my test methodology stated above, I still feel a general picture of the performance can be deduced.  The double stack cooler lands somewhere between the Spire and the Stealth cooler and if this hack can be done for free, provides just a little bit more cooling capacity.

2 comments:

  1. Nice! I think im gonna try this when my r5 3600 arrives

    ReplyDelete
  2. this is awesome, ill never do it, but awesome none the less. Thank you for this!

    ReplyDelete