Overclock.net banner

[Gamer's Nexus] Threadripper Cooler, Thermal Paste Coverage vs. Die Area & IHS

2.5K views 12 replies 9 participants last post by  mmonnin  
#1 ·
http://www.gamersnexus.net/news-pc/3008-threadripper-cooler-and-thermalpaste-coverage-vs-die-ihs

Quote:
Of note, there is an "optimal" orientation for the coldplate: You'd want the microfins long-ways across the widest part of the CPU IHS, as that'd ensure that the most thermal mass sits directly over the dies. We also run into concerns where the coldplate's most effective cooling area sits atop nothing of import -- the center of Threadripper isn't a die, it's spacing between dies. Typically, that center area would also be the center of thermal activity on a traditional single-die desktop CPU. That's why the cooling mass is focused over the center like it is, but that design doesn't apply well to Threadripper. Instead, we have two peripheral hotspots toward the outer perimeter of the coldplate. Although these Asetek CLCs may be adequate for cooling Threadripper, if only because of brute force, we also believe they are less efficient than they could be. There's a lot of room for other manufacturers to improve on this, given the unique parameters of Threadripper. And to be clear, this isn't a fault of AMD or of the cooler makers -- Threadripper and the coolers can exist independently as well-designed products, but that doesn't mean they can exist optimally together. But then again, we'll review it eventually. Can't carry this too far.

Those observations do provide a foundation, though: Here's a look at the die location for Threadripper:


 
#4 ·
Quote:
Originally Posted by 222Panther222 View Post

Why didn't he make a custom bracket so he could test and benchmark it, we want to see temps. Not a piece of copper on top of a cpu.
If you watch the video he is not allowed to yet review data is still under NDA.
 
#7 ·
What are they using in enterprise environments where large processors were already common? Are they just using a simple HS then rack fans with ample cooling of the room itself? If that isn't the case surely there exists some cooling equipment (maybe not with the proper bracket yet) already that can handle larger processors?

I hope when he does the actual testing he tries it a couple of times rotating the cooler 90 degrees to see how much difference that makes since the cold plate isn't uniform and you can't see its orientation without disassembly.
 
#9 ·
Quote:
Originally Posted by Olivon View Post

Heatsink got pre-applied thermal pad on it CasualCat.
I know the thermal paste was pre-applied. Not sure what you're addressing exactly?

edit: The first part of my post is asking not what is included in TR, but what is used today for existing large processors in enterprise environments.

The second part of my post is addressing the non-square water channels of the cold plate whose orientation is hidden from the user. The brackets on most (if not all?) of those AIO cpu coolers are such that the cylindrical pump/plate combo can be rotated.

So what I'm wondering and what Steve wasn't sure about was whether the rectangular water channel section being oriented perpendicular to the dies under the IHS or aligned with them would make a difference or not. You could test that without knowing their orientation with multiple reapplies with rotations. You'll get some variation with reapplies anyhow, but if you start seeing significant repeatable differences after rotations it would be telling.
 
#10 ·
Any word on whether the Raystorm Pro will get a TR bracket? The block (and fin section) seems big enough to provide better contact to both dies versus AIO blocks.
 
#11 ·
Quote:
Originally Posted by CasualCat View Post

I know the thermal paste was pre-applied. Not sure what you're addressing exactly?
Sorry I didn't understand your question.
Cooling on servers is based on heavy air flow from front and go through the heatsinks.
Even Itanium CPU are cooled like that and it's not a problem.
Of course, you need heavy clim to cool the room. Or, like we have at my job is an InRow system who tramsform the servers heat in cold outside the racks. It's called hot aisle / cold aisle sytem.
 
#12 ·
Quote:
Originally Posted by Olivon View Post

Sorry I didn't understand your question.
Cooling on servers is based on heavy air flow from front and go through the heatsinks.
Even Itanium CPU are cooled like that and it's not a problem.
Of course, you need heavy clim to cool the room. Or, like we have at my job is an InRow system who tramsform the servers heat in cold outside the racks. It's called hot aisle / cold aisle sytem.
Ok that is what I thought for enterprise, but wasn't sure. Thanks.
 
#13 ·
Quote:
Originally Posted by CasualCat View Post

Ok that is what I thought for enterprise, but wasn't sure. Thanks.
http://store.supermicro.com/heatsink/2u-heatsink.html

Just larger heatsinks than the circular AIOs. In front there would be a row of fans pushing air through cases and these fins. From the cold to hot aisles as mentioned above.