Is push-pull still important for CPU radiators?

[yoadknux]

Back in the day it was recommended to put fans in push-pull configuration. When I had a liquid cooled 1080Ti I saw clear advantages doing push-pull and having fans in high RPM. This happened because the delta between the liquid temperature and the GPU die was within about 10c. However nowadays when I'm trying to cool my overclocked 13600k I notice that the CPU easily gets to 90c+ under continuous stress, while the air coming out of the radiator is barely hot, and liquid temperature is often in the 30-40c range.
I have come to the conclusion that the bottleneck in cooling here is not the cooling of the radiator, but rather the heat-transfer interface between the CPU and the waterblock, therefore making any airflow redundant. Is this statement correct? Is push-pull pointless now?
Also, how is it that my GPU consumes 400W+ at ~70c, while the CPU reaches to ~90c at 200W? Why is there such a big difference between them?

Would love your insight on this! Thanks!

 

[Voodoo Rufus]

You're on the right track. The issue is power density: how many watts per square millimeter of die area you're trying to extract. A slab of metal in your waterblock can only take heat out of a tiny CPU die so quick, but it can absorb a ton from a large GPU die more easily.

13600k: 257 mm^2, 181W PL2. .7 W/mm^2
1080Ti: 471 mm^2, 250W TDP. .54 W/mm^2
3080: 628 mm^2, 320W TDP. .51 W/mm^2

The other issue is that of the IHS on the CPU. You have the solder interface, the copper IHS, a paste interface, then your block. Each one induces another thermal resistance loss, and the 'cooling engine' of your block is further away from the die heat source. More interfaces, more resistances, more distance. Your GPU has just the thermal paste directly attached to the waterblock, giving you one interface and very little distance between heat source and cooling engine.

 

[EarthDog]

Push/pull was a good thing if you had high FPI rads for sure.

That said, the problem isn't the cooling or push/pull, but the higher density of the die itself and getting heat out them that's the problem.

Going push/pull because of the CPU isn't going to help much because of the density and surface area. Instant ramps to 90C are common with modern chips, and OK.

 

[Zerileous]

Just want to add to all of the great comments above, comparing temperatures between generations is not apples to apples. There is a lot that goes on under the hood in terms of where the thermal sensors are located, how their data are interpreted by software, and what ultimately gets reported as your temp. Your GPU may well have a 90 degree temp somewhere that is just hidden behind the scenes, or not even measured. The point of this to say, don't be uncomfortable with a number just because you're not used to it. If it's within spec then its okay to run.

 

[Voodoo Rufus]

To add to EarthDog's info, the different radiator/fan configurations all revolved around the same concept of moving more air through the rads. Once people got away from heater cores (oh the dark ages....those things pretty much needed squirrel cage blowers for enough static pressure), things got easier.

The different concepts:
Push/pull
Thick frame fans
Shrouds (helped with noise as well as the 'dead zones' around the fan hubs, spread out the airflow more)

Or just crank more rpms, giving you the performance of push/pull with fewer fans. Unless you have something nuts like an Alphacool 80mm thick radiator, there's limited utility to push/pull. Better off just going with a bigger rad for performance value. Less fans to buy than P/P on a smaller rad, more heat dissipation area.

 

[EarthDog]

Absolutely... all of this great information (and more!) can also be found on our water cooling sticky threads too!

 

[Nebulous]

Push/pull? Of course! Plus the ladies like it because it's sexy!

I don't always push/pull, but when I do, it's with great power

And with great power, yadda yadda, you know the rest

 

[stompah]

Yeah Dino, great power but short stroke and never bottom out.

My belief is that today's age cooling has shifted from a large air cooler with premium thermal paste to overclock, to hard mods to get a similar affect. Meaning delidding and custom loop.

Manufacturers have caught on to us overclockers and started binning chips that overclockers used to seek out for their overclocking potential. Hence why needed more mods to overclock and gain less percentage of overclock.

 

[Voodoo Rufus]

The binning and adaptive clock rates and voltages combined with the already sky high clock rates have basically obsoleted OCing except for bench junkies. Undervolting seems to be the game now since some motherboards just like to throw tons of power around.

And when chips like the 7800X3D only use 50W while gaming, a nice 92mm HSF is totally fine!

 

[yoadknux]

You're on the right track. The issue is power density: how many watts per square millimeter of die area you're trying to extract. A slab of metal in your waterblock can only take heat out of a tiny CPU die so quick, but it can absorb a ton from a large GPU die more easily.

13600k: 257 mm^2, 181W PL2. .7 W/mm^2
1080Ti: 471 mm^2, 250W TDP. .54 W/mm^2
3080: 628 mm^2, 320W TDP. .51 W/mm^2

The other issue is that of the IHS on the CPU. You have the solder interface, the copper IHS, a paste interface, then your block. Each one induces another thermal resistance loss, and the 'cooling engine' of your block is further away from the die heat source. More interfaces, more resistances, more distance. Your GPU has just the thermal paste directly attached to the waterblock, giving you one interface and very little distance between heat source and cooling engine.

Thank you for this very insightful comment! It seems that heat density and the additional interfaces between the cooler and the CPU are the main limitations for cooling. So, in that sense, I guess that once you pass a certain threshold, going for larger radiators would not make a difference - for example I was considering upgrading my 280mm AIO to 420mm, but this should not affect the thermals.
Why isn't it possible to just make larger CPUs then? Or do direct cooling interface like the GPU? There is clearly an issue with cooling these new chips.

 

[EarthDog]

for example I was considering upgrading my 280mm AIO to 420mm, but this should not affect the thermals.

It will, certainly, but those insta-blips to high temps won't because of the density. The die/IHS just cant get the heat out fast enough. The mediums (die to paste/solder to IHS) have always been there.

Why isn't it possible to just make larger CPUs then? Or do direct cooling interface like the GPU? There is clearly an issue with cooling these new chips.

1. $

2. You can remove the IHS yourself and direct die cool... there are kits for doing just that! YOu just need to be careful not to break the die because of pressure as Z heights are different without the IHS.

EK-Quantum Velocity² IHS Removal Tool - 1700

EK-Quantum Velocity² IHS Removal Delidding Tool - 1700

www.ekwb.com

3. There isn't a cooling issue with these chips. They are spec'd to run to 100C and then throttle. Intel and Mobo makers are getting every MHz out of these chips out of the box these days. I don't buy into binning of these chips too much, no more than they were before. It's not like the lower class chips overclock significantly more than the flagships.

 

[Voodoo Rufus]

Larger dies make for lower yields per wafer. It's a cost thing. Companies want to ship more processors and maximize yields. It's also why you see more of a push for lower power requirements to help keep temperatures in check.

IHS made it less likely that PC assemblers would damage the die when attaching heatsinks. When you think about the weight and leverage imposed by modern coolers, it makes sense to protect the processor.

Direct die cooling is great, if you feel like your performance or clocks are suffering from excessive temperatures.

 

[Woomack]

IHS made it less likely that PC assemblers would damage the die when attaching heatsinks. When you think about the weight and leverage imposed by modern coolers, it makes sense to protect the processor.

Intel went another way and to reduce RMA rate, they moved pins from CPUs to motherboard sockets ... in the same time Intel stopped manufacturing motherboards. Now they are pushing fragile sockets and if anything happens then it's not their problem. Problems with sockets are happening quite often. Somehow problems with AMD sockets, even LGA are not so common.

 

[Al_bundy]

Push/pull is an big advantage if u use thick radiators.

 

[EarthDog]

Push/pull is an big advantage if u use thick radiators.

True. But it depends on FPI as well. Low FPI on a thick rad, the benefits won't be seen as much compared to a rad with higher FPI. The point of a push pull is to increase the static pressure and airflow.

 

[Al_bundy]

The most of the peoples prefer silent and choose lowpreasure fans.

 

[EarthDog]

The most of the peoples prefer silent and choose lowpreasure fans.

Yep. But that doesn't change anything either of us said. Push/pull helps with static pressure and helps the most with a thick rad and higher fpi.

I've run some low speed yate loons and deltas on old school PA 120.3 (low fpi) and didn't see too much difference, for example.

 

[BugFreak]

I generally use thik radiators (currently Alphacool UT60) so to me it makes a big difference. In my current build I actually started with just push to save space and cleaner tube routing but the temps sucked. Once I added the pull to the radiators my temps were closer to my expectations. It also lowered the temp in the case thanks to the better airflow.

If you are using some thin radiator like most kits then I doubt it would make much difference though. Push or push/pull there isn't enough water inside and surface getting air movement to matter. Just my opinion as I can't think of any time I've used a slim radiator.

 

[Zerileous]

Static pressure of the fan and fin density matter too.

 

[EarthDog]

Indeed. Lots of variables involved.