Gamers Nexus starts detailed cooler testing

[GDRRiley]

Gamers nexus has after 3 years of refining testing methodology is going to begin to release CPU cooler tests.

The use a combination of synthetic and real wold.

They cover many of the mistakes made often in testing and what they are doing to eliminate them. From using the same CPU, board, Ram, PSU and GPU to testing on a heater representing a zen 3 CPU. The test using a heater has 3 spots on a ryzen IHS, representing the 2 main chiplets and the IO die
It is highly automated and covers thousands of points.

My thoughts, seems to be the most detailed and crazy testing done by any review website/channel. I hope it will finally answer some of those which cooler are better questions.

 

EDIT:

First review using it is now out.

 

Edited March 10, 2020 by GDRRiley
added more, reworded.

[GrockleTD]

Madladding intensifies

[Danioki]

only tech Jesus can best tech Jesus

[Tenelia]

I remember back in 2012, a thread discussed if it might be best to simply cut out a block of metal to simulate a CPU and heat to 150W, 100W, 50W, and 35W. Then mount each cooler one by one onto that artificial metal block.

 

We've come really far. Wow.

[Dabombinable]

4 minutes ago, Tenelia said:

I remember back in 2012, a thread discussed if it might be best to simply cut out a block of metal to simulate a CPU and heat to 150W, 100W, 50W, and 35W. Then mount each cooler one by one onto that artificial metal block.

 

We've come really far. Wow.

Some reviews actually did that in the past, with various set thermal output.

[Guest]

It baffles me that a lot of reviewers still dont noise normalize their GPU/CPU cooler tests. Thats one of the only ways to really show which cooler is better instead they're just benchmarking which fan spins the fastest.

[kakik09]

Doesn't Steve believe loop order matters, as opposed to Linus who doesn't? I wonder when this will be settled

[mr moose]

10 minutes ago, kakik09 said:

Doesn't Steve believe loop order matters, as opposed to Linus who doesn't? I wonder when this will be settled

You'll likely need to ask a thermal physics expert for a definitive answer, however  If everything I have read is to be believed it does matter as a mathematical technicality, but likely won't make much of a difference due to the overall cooling capacity and the heat sources combined.

[leadeater]

4 hours ago, GDRRiley said:

I hope it will finally answer some of those which cooler are better questions.

The cooler made by Steve 

[GDRRiley]

Just now, leadeater said:

The cooler made by Steve 

He’s going to have all the data so he’ll just rig the gamers nexus tower cooler 9000 to be the best. 

[RejZoR]

FrostyTech has been using a dedicated heating element instead of actual CPU to test coolers. It also allowed them to test based on raw wattage opposed to many things that affect CPU's and CPU coolers. May not be the most realistic representation in terms of workloads vs heat generated, but it basically throws it all away and if there is 250W of heat, that's 250W of heat. If one cooler can deal with this better than the other, it'll most likely scale the same in both down and up direction.

[Soppro]

Wow good stuff, I wonder how it compares to Anandtech's testing methodology! Either way, I respect the huge effort to design something like this just to test CPU coolers  

[TVwazhere]

I'm not actually sure how well they're going to implement this. And I'm not referring to their methodologies..... I'm talking about actually doing them. Mostly because we were supposed to get PSU testing done with them a while back (2018 they had their station set up) and so far I think we've gotten.... two PSU tests done (unless they only post them on Patreon?)

[GDRRiley]

1 hour ago, TVwazhere said:

I'm not actually sure how well they're going to implement this. And I'm not referring to their methodologies..... I'm talking about actually doing them. Mostly because we were supposed to get PSU testing done with them a while back (2018 they had their station set up) and so far I think we've gotten.... two PSU tests done (unless they only post them on Patreon?)

This was even older. They seems to go the do it right or not at all 

[Tan3l6]

I wish that there would be a site like "Frostytech" used to be, for referencing all the coolers ...

Fingers crossed.

[BuckGup]

Good, this is the info I want

[Doobeedoo]

That's neat, always great to see cooling potential of coolers. 

[Curufinwe_wins]

8 hours ago, mr moose said:

You'll likely need to ask a thermal physics expert for a definitive answer, however  If everything I have read is to be believed it does matter as a mathematical technicality, but likely won't make much of a difference due to the overall cooling capacity and the heat sources combined.

"Thermal physics expert" here.

 

It is measurably different mathematically, but if the loop flow is sufficient/normal, the dT from each effective pass is small enough that it doesn't matter in any practical sense. The thermal transfer bottlenecks are so huge in other locations that it IS irrelevant. 

 

In a **very** low flow rate situation the problem is dramatically different, and in that case the **safest** (technically least thermally efficient though) method would be to have a radiator between every major heat generating component.

 

 

On topic, cool stuff from Steve, honestly even the Anandtech method of a uniform heat source at given load is actually much better than other testing methods. Doing granularity in heat production is really cool, as long as they have flexibility to simulate other heating profiles since the zen2 version is likely to be less accurate than the uniform approximation for say a TR2 or Intel chip.

[mr moose]

4 hours ago, Curufinwe_wins said:

"Thermal physics expert" here.

 

It is measurably different mathematically, but if the loop flow is sufficient/normal, the dT from each effective pass is small enough that it doesn't matter in any practical sense. The thermal transfer bottlenecks are so huge in other locations that it IS irrelevant. 

 

 

Do I have it right if my understanding is that the first device in the loop will have the best thermal transfer and the last device will have the least, but because the system is more than capable, if you reversed the flow it would make little difference?

 

When you say transfer bottlenecks are so huge elsewere?  which parts of the system have the higher thermal resistance?

 

 

[Curufinwe_wins]

6 hours ago, mr moose said:

Do I have it right if my understanding is that the first device in the loop will have the best thermal transfer and the last device will have the least, but because the system is more than capable, if you reversed the flow it would make little difference?

 

When you say transfer bottlenecks are so huge elsewere?  which parts of the system have the higher thermal resistance?

 

 

It's more that the highest possible dT (between liquid and heatsource/sink) at all possible locations yields the highest performance.

 

For example the radiators are most effective when the water is hottest, and the water blocks are most effective when water is coolest. But if you had in theory a limitless number of heat sources in a row, eventually the water temp would rise enough that even in a single finite pass the liquid would be unable to cool the components, likewise with infinite radiator cooling, the limit is the ambient air temperature. 

 

 

So for example for CPUs, the limit is generally heat transfer through the small die(s) into larger IHS and through thermal paste into the block. Adding a higher performance block and increasing pump speed does relatively little to decrease temperatures at a given load (as load increases, this becomes less true). This is why high end air coolers perform basically the same as AIOs at low load. They hit performance bottlenecks shared between them.

 

For GPUs there is no IHS, and thus direct die cooling combined with a much larger die size means that water cooling is ludicrously more effective and thus temperatures of under 25-30C above ambient are very reasonable to achieve even at very low noise, even while aircoolers struggle to keep GPUs under 50C above ambient.

 

One way to think about heat transfer is through whats called resistance networks. I'm not sure how familiar you are with electrical resistance, but its a very similar analogy. The total thermal resistance (for an one-path problem) can be thought of as the sum of individual resistances. So R_die->IHS+ R_IHS->coolerblock+ R_coolerblock->air=R_total.

 

Given a set Q (energy flow) and set ambient temperature, its possible to calculate the source temperature via the total resistance (this is oversimplifying a bit, but still)

 

For large cpu coolers in particular, the first two resistances are very high and the third resistance is generally low.

 

Tossing out imaginary numbers here, suppose the first resistance is 800 ohms, the second is 200 ohms, and the third is 30 ohms for a total of 1030. Adding an infinite fluid flow and infinite internal block surface area (combined with infinite heat transfer to the air) would reduce the third number to 0 but only drop the total resistance from 1030 to 1000. Hardly a notable change.

[leadeater]

15 minutes ago, Curufinwe_wins said:

But if you had in theory a limitless number of heat sources in a row, eventually the water temp would rise enough that even in a single finite pass the liquid would be unable to cool the components, likewise with infinite radiator cooling, the limit is the ambient air temperature. 

How instant gas hot water heating works, extremely basically.

 

12 hours ago, Curufinwe_wins said:

It is measurably different mathematically, but if the loop flow is sufficient/normal, the dT from each effective pass is small enough that it doesn't matter in any practical sense.

@mr moose

 

Back to this point, which you already covered well, the flow rate is at the core of this the reason loop order doesn't matter. Because the water is flowing through the system so quickly and not in contact with the heat source for very long at all along with all the mixing of the water due to the flow hot areas just aren't being created for the temperature scales and sensitivities here.

 

It's not like a bucket of cold water and poring a jug of hot water in to it, where the area of contact rapidly warms but on the edges of the bucket the water is still cold. You stir that water up in the bucket before poring the jug of hot water and you will get much more even rise in temperature of the water in the bucket.

[mr moose]

52 minutes ago, Curufinwe_wins said:

It's more that the highest possible dT (between liquid and heatsource/sink) at all possible locations yields the highest performance.

 

For example the radiators are most effective when the water is hottest, and the water blocks are most effective when water is coolest. But if you had in theory a limitless number of heat sources in a row, eventually the water temp would rise enough that even in a single finite pass the liquid would be unable to cool the components, likewise with infinite radiator cooling, the limit is the ambient air temperature. 

 

 

So for example for CPUs, the limit is generally heat transfer through the small die(s) into larger IHS and through thermal paste into the block. Adding a higher performance block and increasing pump speed does relatively little to decrease temperatures at a given load (as load increases, this becomes less true). This is why high end air coolers perform basically the same as AIOs at low load. They hit performance bottlenecks shared between them.

 

For GPUs there is no IHS, and thus direct die cooling combined with a much larger die size means that water cooling is ludicrously more effective and thus temperatures of under 25-30C above ambient are very reasonable to achieve even at very low noise, even while aircoolers struggle to keep GPUs under 50C above ambient.

 

One way to think about heat transfer is through whats called resistance networks. I'm not sure how familiar you are with electrical resistance, but its a very similar analogy. The total thermal resistance (for an one-path problem) can be thought of as the sum of individual resistances. So R_die->IHS+ R_IHS->coolerblock+ R_coolerblock->air=R_total.

 

Given a set Q (energy flow) and set ambient temperature, its possible to calculate the source temperature via the total resistance (this is oversimplifying a bit, but still)

 

For large cpu coolers in particular, the first two resistances are very high and the third resistance is generally low.

 

Tossing out imaginary numbers here, suppose the first resistance is 800 ohms, the second is 200 ohms, and the third is 30 ohms for a total of 1030. Adding an infinite fluid flow and infinite internal block surface area (combined with infinite heat transfer to the air) would reduce the third number to 0 but only drop the total resistance from 1030 to 1000. Hardly a notable change.

I use thermal resistance networks to calculate heat sink sizes for discrete electronic components (mainly audio but occasionally voltage regulators.   I figured the thermal resistance would follow the same rules but wasn't sure about whether the highest resistance was between the die and the fluid  or between the fluid and air at the radiator.  

 

Which means no I am not right, the highest points of resistance fall outside of the realm of the cooling loop therefore direction largely means nothing.  It's not because the cooling system can dissipate x times the heat than the whole circuit produces (although that is also true) it's because the bit that makes the most difference is not effected by fluid flow direction while the smallest component is.

[Lathlaer]

The video was quite informative and somewhat interesting and now I wonder how their methodology will affect the end results.

 

In particular, I'm interested to know whether it will validate what everyone knows about coolers from flawed tests ("this is top tier/this one sucks") or whether we will see some upsets

[TheNamelessOne]

While I love his dedication and amount of detail he and his team decided to put in this, I'm pretty sure at the end of the day none of this matters and the results will be within margin of error between testing it the regular way and doing all this. Maybe 2-3 degrees difference.

[Lathlaer]

2 minutes ago, TheNamelessOne said:

While I love his dedication and amount of detail he and his team decided to put in this, I'm pretty sure at the end of the day none of this matters and the results will be within margin of error between testing it the regular way and doing all this. Maybe 2-3 degrees difference.

It's gonna suck really bad if after all those months of testing and thousands of dollars of new equippment the conclusion will be "Noctua coolers are the best"