News Der8auer admits he 'messed up' with underperforming Thermal Grizzly Heatspreaders and Direct Die coolers

[Admin]

Der8auer addresses concerns around the latest Thermal Grizzly releases and details solutions.

Der8auer admits he 'messed up' with underperforming Thermal Grizzly Heatspreaders and Direct Die coolers : Read more

 

[bit_user]

Based on my humble experiments with backside cooling, I wonder if that would be a fruitful direction for him to find the next set of cooling performance enhancements:

Question - How much does cooling the underside of the motherboard/CPU typically help ?

I have an Alder Lake N97 mini-ITX motherboard which has a soldered-down BGA CPU and non-standard heatsink. Upgrading the heatsink isn't a very realistic option, at least not without some custom machining that I'm not setup to do. I have added case fans, but the CPU is still sometimes pegged at...

forums.tomshardware.com

As I mention in that thread, there are particulars about my situation that yielded such large improvements. Also, I'm dumping most of the heat directly to the case, which wouldn't be a good idea with a much higher-power CPU. But, there's still a lot else you could do.

For instance, I think someone could probably design a CPU socket that's more thermally-conductive and use it on a motherboard with a heatsink that you can mount on the motherboard backside, directly underneath the CPU. Some cases have a cut-out that could accommodate it.

Also, he's known to work with case manufacturers, whom he could encourage to focus some airflow directly on that spot.

O11 Dynamic

<span data-sheets-value="{"1":2,"2":"Space optimization chassis, The O11 DYNAMIC is the combination of modern design and art. Through years of researching, collecting feedback and redesigning; We have successfully craft the best edition of the dual chamber chassis. Experience the full view...

lian-li.com

 

[PEnns]

Just an observation: In case anybody is wondering how to pronounce his name (Der8auer):

It is Der Bauer, which means The Farmer in English.
Carry on.

 

[dos_box]

Just an observation: In case anybody is wondering how to pronounce his name (Der8auer):

It is Der Bauer, which means The Farmer in English.
Carry on.

To fully understand the name you would need to be German. Der Bauer means something like farmer/peasant/person that owns land and has livestock or crop. It can also be used as a slightly offensive word, for someone of low intelect or someone of the working class, but often in a more mocking way and thereby often used for a standard type of guy. So it's not too offensive.

But the twist here is the literal translation. Der Bauer literally means The Builder (which is in German a farmer). But the literal translation is when you understand the true meaning. He is building computer systems. That's the twist here in my opinion. So it's more of a word-game.

 

[Amdlova]

Long long time Ago gotcha a full aluminum case... put a huge thermal pad on back of the mothetboard and sandwich it with the case... droop almost 8 degrees celcius on a 3770k at 5.2ghz

 

[bit_user]

Long long time Ago gotcha a full aluminum case... put a huge thermal pad on back of the mothetboard and sandwich it with the case... droop almost 8 degrees celcius on a 3770k at 5.2ghz

How big was the "huge thermal pad"? Like 50x50 mm or 250x250 mm? I assume it was also about 5 mm thick?

I actually found someone selling a 40x40x6 mm thermal pad, for this purpose. I could've just used that as is, but the thermal conductivity of copper is a couple hundred times as high as thermal pads, so it's better if you can use copper (or even aluminum) to fill most of the gap.

 

[mitch074]

How big was the "huge thermal pad"? Like 50x50 mm or 250x250 mm? I assume it was also about 5 mm thick?

I actually found someone selling a 40x40x6 mm thermal pad, for this purpose. I could've just used that as is, but the thermal conductivity of copper is a couple hundred times as high as thermal pads, so it's better if you can use copper (or even aluminum) to fill most of the gap.

The main problem is the conductivity - traces at the back of the socket would need insulation, which would defeat the purpose.

 

[jp7189]

Based on my humble experiments with backside cooling, I wonder if that would be a fruitful direction for him to find the next set of cooling performance enhancements:

​

​

Question - How much does cooling the underside of the motherboard/CPU typically help ?

I have an Alder Lake N97 mini-ITX motherboard which has a soldered-down BGA CPU and non-standard heatsink. Upgrading the heatsink isn't a very realistic option, at least not without some custom machining that I'm not setup to do. I have added case fans, but the CPU is still sometimes pegged at...

forums.tomshardware.com

​

As I mention in that thread, there are particulars about my situation that yielded such large improvements. Also, I'm dumping most of the heat directly to the case, which wouldn't be a good idea with a much higher-power CPU. But, there's still a lot else you could do.

For instance, I think someone could probably design a CPU socket that's more thermally-conductive and use it on a motherboard with a heatsink that you can mount on the motherboard backside, directly underneath the CPU. Some cases have a cut-out that could accommodate it.

Also, he's known to work with case manufacturers, whom he could encourage to focus some airflow directly on that spot.

​

O11 Dynamic

<span data-sheets-value="{"1":2,"2":"Space optimization chassis, The O11 DYNAMIC is the combination of modern design and art. Through years of researching, collecting feedback and redesigning; We have successfully craft the best edition of the dual chamber chassis. Experience the full view...

lian-li.com

​

That's a tough one from a general design/ manufacturing point of view because there's no standard (as far as I know) for clearance behind the motherboard, and pretty hard to get air flow back there on most cases. Maybe the cooler backplate could have attachment points for an optional modular sink.

This also could be interesting from socket reinforcement point of view as you could torque both sides together to keep bending forces equalized and possibly improve contact.

I was recently playing with an SP5 socket which retention frame is not strong enough to get contact on all 6096 lga pins. It relies on the heatsink pressure at specific screw torque. That has me thinking of a 12900k which had both cooling issues and memory channel performance issues (theorized it was from poor pin contact). If I had a backside sink to torque at the same time I was torquing the front side, i might have been able to get better contact on both the sink and the lga pins.

 

[bit_user]

The main problem is the conductivity - traces at the back of the socket would need insulation, which would defeat the purpose.

In the thread I linked, you can see my solution to this problem. First, I shaved off the leads and solder blobs poking through from the other side. Then, I painted the area with some non-conductive varnish (you sometimes see people talk about using clear fingernail polish, but I used a product specifically designed to withstand high temperatures and be non-reactive).

Finally, I used thermal pad of the same thickness as the tallest surface-mount component being covered. This was about 0.9 mm, so I had to go up to a 1.0 mm thick thermal pad. I first tried it with 0.5 mm, because I didn't have 1.0 mm at the time. With the 0.5 mm pad, the heatsink made almost no difference, since it was barely making any contact.

It was a lot more trouble than just using a 6 mm thermal pad and definitely voided my warranty on the motherboard. These are yet more reasons it would be better to have a motherboard specifically designed to support backside cooling.

 

[A Stoner]

The most likely culprit is that every CPU is different. Which is why Intel has such large tolerances built into their production cooling solution. Everyone complains that their solutions are not ideal, but none of those complainers take into account that intel is not hand crafting each individual component and carefully modding each piece to be a perfect fit.

If they did that, they would only be able to sell a very few extremely expensive chips and most people would never get to own a computer, cell phone or anything else computish.


Every peice that is put together has tolerance, every assembly adds tolerance, at the end of the day all those tolerances are added together, maybe given a fudge factor loosening tolerances for safety or ran through a process that can tighten those tolerances and ensure that 99.9% of assemblies will work. I have seen some components where performance is critical enough that a 90% scrap rate was the best the manufacturing process would allow, every component had to be nearly perfect. I have also seen where every component has to be measured, marked, and only assembled with matching component measurements. A time consuming and space consuming process where some components end up never being usable. These things only happen where the customer is willing to pay a very hefty premium.

 

[PEnns]

To fully understand the name you would need to be German. Der Bauer means something like farmer/peasant/person that owns land and has livestock or crop. It can also be used as a slightly offensive word, for someone of low intelect or someone of the working class, but often in a more mocking way and thereby often used for a standard type of guy. So it's not too offensive.

But the twist here is the literal translation. Der Bauer literally means The Builder (which is in German a farmer). But the literal translation is when you understand the true meaning. He is building computer systems. That's the twist here in my opinion. So it's more of a word-game.

You're right and it now makes more sense: I forgot that "Bauen" means to build in German. Mein Fehler.

 

[bit_user]

The most likely culprit is that every CPU is different. Which is why Intel has such large tolerances built into their production cooling solution.

Citation needed. The reason I'm skeptical of your claim is that modern lithography is a nanoscale affair, requiring exacting precision unlike that found anywhere else in industry. That makes it very hard for me to believe they have such large variations in die height, adding up to hundreds of micrometers.

Standard industry practice is to shave off like 90% of the wafer, after fabrication, to yield the final die. This is an exacting process, during which they should be ensuring the die conforms to their height restrictions for the cooler.

Die placement and soldering should also be very exacting, again on the order of micrometers. It's not clear where the supposed variation you're talking about would come from.

Every peice that is put together has tolerance, every assembly adds tolerance,

Not all tolerances arise from manufacturing variations. To a large degree, they could originate from variations in installation & operating conditions, service life requirements, and statistical margins needed to ensure acceptably low long-term failure rates.

I have seen some components where performance is critical enough that a 90% scrap rate was the best the manufacturing process would allow, every component had to be nearly perfect.

Just to be clear, are we talking about semiconductor products, made on leading-edge nodes, or perhaps some very different sort of product?

 

[Amdlova]

How big was the "huge thermal pad"? Like 50x50 mm or 250x250 mm? I assume it was also about 5 mm thick?

I actually found someone selling a 40x40x6 mm thermal pad, for this purpose. I could've just used that as is, but the thermal conductivity of copper is a couple hundred times as high as thermal pads, so it's better if you can use copper (or even aluminum) to fill most of the gap.

like 3mm. Because the metal hardware of nzxt heat sink. Size same as the socket.

 

[mitch074]

In the thread I linked, you can see my solution to this problem. First, I shaved off the leads and solder blobs poking through from the other side. Then, I painted the area with some non-conductive varnish (you sometimes see people talk about using clear fingernail polish, but I used a product specifically designed to withstand high temperatures and be non-reactive).

Finally, I used thermal pad of the same thickness as the tallest surface-mount component being covered. This was about 0.9 mm, so I had to go up to a 1.0 mm thick thermal pad. I first tried it with 0.5 mm, because I didn't have 1.0 mm at the time. With the 0.5 mm pad, the heatsink made almost no difference, since it was barely making any contact.

It was a lot more trouble than just using a 6 mm thermal pad and definitely voided my warranty on the motherboard. These are yet more reasons it would be better to have a motherboard specifically designed to support backside cooling.

Probably why it's not often found "in the wild", as it would be a very expensive process - and thus, still a problem.

 

[bit_user]

Probably why it's not often found "in the wild", as it would be a very expensive process - and thus, still a problem.

You could engineer a proper solution. Someone like Der8auer has the clout to get a motherboard maker and case manufacturer to make a motherboard with a heatsink that mounts back there and a case which provides enough room & good airflow over it. I already linked a case on which Lian Li consulted with him.

 

[A Stoner]

Citation needed. The reason I'm skeptical of your claim is that modern lithography is a nanoscale affair, requiring exacting precision unlike that found anywhere else in industry. That makes it very hard for me to believe they have such large variations in die height, adding up to hundreds of micrometers.

Standard industry practice is to shave off like 90% of the wafer, after fabrication, to yield the final die. This is an exacting process, during which they should be ensuring the die conforms to their height restrictions for the cooler.

Die placement and soldering should also be very exacting, again on the order of micrometers. It's not clear where the supposed variation you're talking about would come from.


Not all tolerances arise from manufacturing variations. To a large degree, they could originate from variations in installation & operating conditions, service life requirements, and statistical margins needed to ensure acceptably low long-term failure rates.


Just to be clear, are we talking about semiconductor products, made on leading-edge nodes, or perhaps some very different sort of product?

Wafer thickness, process variation on substrates, items being manufactured at different facilities all add up to dimensional variance along with the actual heat spreaders having a tolerance stack.

And yes, the 90% scrap rate I know of is on mechanical devices, but even processors, memory modules and so forth come with scrap rates, particularly at cutting edge technologies as they fine-tune the processes that make them.

Anyways, processors are not just electronic, they are mechanical, and ALL mechanical objects that are manufactured have size tolerances. All those size tolerances add up and determine the design gap that would be needed to ensure that 100% of the products work together.