We Made the Perfect CPU Cooler - Intel vs. AMD Curvature & Coldplate Engineering
video description
Date: 2024-05-06
Related videos
Comments and reviews: 20
steadmansmith5591
so, First off I'd like to say that while this is extremely interesting the retail sample didn't really perform bad. And that it still did a great job cooling. And for all purposes this is a practice in extreme optimization that most people building their own PC's really shouldn't lose sleep over. The CPU deforming when mounted with a cooler is one of the reasons why thermal paste is a thing.
Now for my input on what to do with this information with 3 sockets all giving different topography changes when the cooler is mounted, and how cpu cooler manufactures generally needing to manufacture coolers that will mount on multiple different sockets with the aid of mounting brackets. I'm not sure it would be prudent for most manufactures to create a SKU for custom shaped thermal surface.
What would be great if there could be a guide on how an end user might be able to polish the cooling surface by hand to customize it for their socket. For example in the past I when I have solid copper thermal surface/ heat sink block, I've busted out the brasso to smooth out the surface and remove oxidation by with a soft cloth resulting in a shiny reflective surface. I wonder if this same technique could be used shape the block for a socket's bowing pattern.
reply
so, First off I'd like to say that while this is extremely interesting the retail sample didn't really perform bad. And that it still did a great job cooling. And for all purposes this is a practice in extreme optimization that most people building their own PC's really shouldn't lose sleep over. The CPU deforming when mounted with a cooler is one of the reasons why thermal paste is a thing.
Now for my input on what to do with this information with 3 sockets all giving different topography changes when the cooler is mounted, and how cpu cooler manufactures generally needing to manufacture coolers that will mount on multiple different sockets with the aid of mounting brackets. I'm not sure it would be prudent for most manufactures to create a SKU for custom shaped thermal surface.
What would be great if there could be a guide on how an end user might be able to polish the cooling surface by hand to customize it for their socket. For example in the past I when I have solid copper thermal surface/ heat sink block, I've busted out the brasso to smooth out the surface and remove oxidation by with a soft cloth resulting in a shiny reflective surface. I wonder if this same technique could be used shape the block for a socket's bowing pattern.
reply
Lysdexis
Could you make a jig to clamp the cpu down and lap the cooler/cpu to match each others curvature with an either polishing or lapping compound. Assuming a polishing compound has a much much finer grit it should remove slight amounts of material but enough to make a difference while keeping a mirror polish finish. Depending on how thick the nickel plating is vs how much material is removed you most likely would not need to get it plated again as we are talking tiny tiny amounts of material removed. Generally nickel plating is between 12 and 25 microns. If the entire curvature of the surface is less than 12 microns on the flattest sample you have then plating is not necessary after lapping.
Thus both surfaces are matched to each other, confirming this with second scans and pressure paper test after this process to see if it actually makes a difference.
reply
Could you make a jig to clamp the cpu down and lap the cooler/cpu to match each others curvature with an either polishing or lapping compound. Assuming a polishing compound has a much much finer grit it should remove slight amounts of material but enough to make a difference while keeping a mirror polish finish. Depending on how thick the nickel plating is vs how much material is removed you most likely would not need to get it plated again as we are talking tiny tiny amounts of material removed. Generally nickel plating is between 12 and 25 microns. If the entire curvature of the surface is less than 12 microns on the flattest sample you have then plating is not necessary after lapping.
Thus both surfaces are matched to each other, confirming this with second scans and pressure paper test after this process to see if it actually makes a difference.
reply
Vennnaya
But there is more to the story, no
At least as far as thermals goes, and specifically, TIM.
Its all well and good seeing how the cold plate and the IHS deforms under pressure....But surely that shape would affect TIM spread patterns etc, no
For example, you noted how AM5 has a kind of ridge down the middle. TIM is usually applied in the middle, too. So how exactly does that affect coverage Couldn't it be conceivable that, depending on application, you get microscopic air pockets
On that note, it would have been nice to see how TIM was applied when you were doing the thermal testing. 1-2c difference can be explained by a slightly different ammount, or application, of TIM, is that not true
reply
But there is more to the story, no
At least as far as thermals goes, and specifically, TIM.
Its all well and good seeing how the cold plate and the IHS deforms under pressure....But surely that shape would affect TIM spread patterns etc, no
For example, you noted how AM5 has a kind of ridge down the middle. TIM is usually applied in the middle, too. So how exactly does that affect coverage Couldn't it be conceivable that, depending on application, you get microscopic air pockets
On that note, it would have been nice to see how TIM was applied when you were doing the thermal testing. 1-2c difference can be explained by a slightly different ammount, or application, of TIM, is that not true
reply
gamersnexus
GamersNexus, thank you for everything you do. I've been a fan for several years now (almost a decade). Thank you for investing so deeply into so many aspects of reviews of the space: products, results, manufacturing, R&D, financial, adjacent company stories, and even customer support experiences.
Where some may always promote nix as their daily driver, I promote you as my reviewer.
Being very upfront with methodology and why that's your method is so important (and referring to video/post X vs explaining every video is good).
If you get something wrong, you report on that too.
The honesty, effort, and commitment is why I support you.
reply
GamersNexus, thank you for everything you do. I've been a fan for several years now (almost a decade). Thank you for investing so deeply into so many aspects of reviews of the space: products, results, manufacturing, R&D, financial, adjacent company stories, and even customer support experiences.
Where some may always promote nix as their daily driver, I promote you as my reviewer.
Being very upfront with methodology and why that's your method is so important (and referring to video/post X vs explaining every video is good).
If you get something wrong, you report on that too.
The honesty, effort, and commitment is why I support you.
reply
ltsmash7527
hey ive just seen that you are promoting the arctic liquid freezer. although i never used it (just used their air cooler arctic freezer i32 CO for LGA 775 with my old i5 6600k and it still runs greatly today) i appreciate this, they make good products. The MX4 and MX5 thermal paste is legendary! Not because its conductivity, but because of its longevity. You can lay down a 20g pack after you built something and it will be still the same after years...if it has been opened maybe it crumbles on the first push but if you resealed it on the right way its just like fresh one..
reply
hey ive just seen that you are promoting the arctic liquid freezer. although i never used it (just used their air cooler arctic freezer i32 CO for LGA 775 with my old i5 6600k and it still runs greatly today) i appreciate this, they make good products. The MX4 and MX5 thermal paste is legendary! Not because its conductivity, but because of its longevity. You can lay down a 20g pack after you built something and it will be still the same after years...if it has been opened maybe it crumbles on the first push but if you resealed it on the right way its just like fresh one..
reply
The_Mole
I've been doing it the other way around for high performance rigs since a couple of years:
1.Find a compatible socket from a dead board.
2. Dremel it out and insert CPU
3. Fix sand paper to a desk with duct tape
4. Grind the CPU down, check with a straightedge until even.
5. Repeat with finer paper up to 1000 grit.
6. Do the same with the cooler.
7. Clean the surfaces and build the rig, using a THIN layer of liquid metal.
8. Burn in.
9. Enjoy optimal heat transfer.
10. For disassembly, you'll need a pipe wrench to twist off the cooler.
reply
I've been doing it the other way around for high performance rigs since a couple of years:
1.Find a compatible socket from a dead board.
2. Dremel it out and insert CPU
3. Fix sand paper to a desk with duct tape
4. Grind the CPU down, check with a straightedge until even.
5. Repeat with finer paper up to 1000 grit.
6. Do the same with the cooler.
7. Clean the surfaces and build the rig, using a THIN layer of liquid metal.
8. Burn in.
9. Enjoy optimal heat transfer.
10. For disassembly, you'll need a pipe wrench to twist off the cooler.
reply
gravitycat7874
This test is tough. The closer you look the more variables come into play. I assume when surfaces were measured they were allowed to acclimate to the measuring atmospheres environment. Also were they checked beyond just their free state condition (ie running temp) You'll also change the thermal conductivity as you modify the wall thickness of both cpu lid and heat pipe base. Beyond that you are at the mercy of manufacturer tolerance variation. You could acquire a killer spec unit that cools better than 99% of other units just by chance.
reply
This test is tough. The closer you look the more variables come into play. I assume when surfaces were measured they were allowed to acclimate to the measuring atmospheres environment. Also were they checked beyond just their free state condition (ie running temp) You'll also change the thermal conductivity as you modify the wall thickness of both cpu lid and heat pipe base. Beyond that you are at the mercy of manufacturer tolerance variation. You could acquire a killer spec unit that cools better than 99% of other units just by chance.
reply
Cinnabuns2009
Would have been nice to also include in the data a lapped Intel and AMD IHS along with a lapped HSF to mate to each. Obviously they're going to move when clamped down but how much and removing mfg warp from the IHS would be nice to know as a separate control.
This would be great to know because the home user can do this at home and possibly or not improve their temperatures or hurt them depending on clamping method.
In other words, if you lap the IHS on Intel/AMD, after they're clamped, what is the map look like
reply
Would have been nice to also include in the data a lapped Intel and AMD IHS along with a lapped HSF to mate to each. Obviously they're going to move when clamped down but how much and removing mfg warp from the IHS would be nice to know as a separate control.
This would be great to know because the home user can do this at home and possibly or not improve their temperatures or hurt them depending on clamping method.
In other words, if you lap the IHS on Intel/AMD, after they're clamped, what is the map look like
reply
oneofmany666
It looks to me that what is going on is that just higher pressure means better thermals. Golden sample seams to be taller and it creates bigger pressure when mounted the same way. If the pressure charts are scaled the same the solid red means you are getting higher value than colors can visually represent.You can verify this by adding additional weight on top of the cooler to see if there is just corelation between pressure and thermals
Overall inconclusive.
reply
It looks to me that what is going on is that just higher pressure means better thermals. Golden sample seams to be taller and it creates bigger pressure when mounted the same way. If the pressure charts are scaled the same the solid red means you are getting higher value than colors can visually represent.You can verify this by adding additional weight on top of the cooler to see if there is just corelation between pressure and thermals
Overall inconclusive.
reply
Stuen4y
Amazing work as always! Your dataset is amazing, I wish it had more data points though. I'm all up for only high power (150W, potentially even 200W) testing for both modern platforms given their IHSs are literally opposite. The reality is that the low power CPU SKUs can be cooled by most $25-30 coolers anyway. The same way you won't bother running a full 4k resolution test suite on IGPs unless you want to illustrate on a single game that it's not really worth it.
reply
Amazing work as always! Your dataset is amazing, I wish it had more data points though. I'm all up for only high power (150W, potentially even 200W) testing for both modern platforms given their IHSs are literally opposite. The reality is that the low power CPU SKUs can be cooled by most $25-30 coolers anyway. The same way you won't bother running a full 4k resolution test suite on IGPs unless you want to illustrate on a single game that it's not really worth it.
reply
duggydo
I fluxed and soldered a cooler and chip together last year. Yes it was extremely risky, but I wanted to try it. It worked and I got much better results. If manufacturers would do that to begin with, cooling could be optimized. The problem is there would be a revenue stream lost for cooler manufacturers and it's a pain in the rear for cpu makers. Try at your own risk, but it's the cheapest way to get excellent results...if you're good enough to not fry the chip.
reply
I fluxed and soldered a cooler and chip together last year. Yes it was extremely risky, but I wanted to try it. It worked and I got much better results. If manufacturers would do that to begin with, cooling could be optimized. The problem is there would be a revenue stream lost for cooler manufacturers and it's a pain in the rear for cpu makers. Try at your own risk, but it's the cheapest way to get excellent results...if you're good enough to not fry the chip.
reply
liiich6175
I wonder if intel washer mod or the aftermarket locking plate would help with the more flat coolers....and is it possible to mount the cooler and then undo the intel latch to get rid of that clamp force
I my self did the washer mod well over a year ago and no issues and I feel it dropped a few Cs too.
Also, back in the day lapping the CPU after it is clamped down was a thing, wonder if a flat cooler and lapped CPU would come closer then 100% stock
reply
I wonder if intel washer mod or the aftermarket locking plate would help with the more flat coolers....and is it possible to mount the cooler and then undo the intel latch to get rid of that clamp force
I my self did the washer mod well over a year ago and no issues and I feel it dropped a few Cs too.
Also, back in the day lapping the CPU after it is clamped down was a thing, wonder if a flat cooler and lapped CPU would come closer then 100% stock
reply
herseem
And here's another thought, Steve: How about having a separate mounting device for CPU's that replicate the pressure the CPU will be under on the motherboard, so that the stop of the IHS can be shaved or milled back to flatness before being re-installed on the motherboard That way it won't matter what shape the CPU forms when it's mounted as you will make it flat, and a properly-flat cooler will fit it better, whichever CPU it is.
reply
And here's another thought, Steve: How about having a separate mounting device for CPU's that replicate the pressure the CPU will be under on the motherboard, so that the stop of the IHS can be shaved or milled back to flatness before being re-installed on the motherboard That way it won't matter what shape the CPU forms when it's mounted as you will make it flat, and a properly-flat cooler will fit it better, whichever CPU it is.
reply
gamersnexus
i would prefer to have the retail model that i could use for a decade moving forward instead of a specialist model intended for one socket that saves me 1.8C. this would also most likely drive up prices, because the company would then need to have multiple production lines, multiple spots on the shelves at retailers, and then multiple SKUs which should be explained to the users bla bla.... in short, headache for all
reply
i would prefer to have the retail model that i could use for a decade moving forward instead of a specialist model intended for one socket that saves me 1.8C. this would also most likely drive up prices, because the company would then need to have multiple production lines, multiple spots on the shelves at retailers, and then multiple SKUs which should be explained to the users bla bla.... in short, headache for all
reply
jg7734
GamersNexus understanding the flatter cooler examples were not necessarily the best on a given platform for an otherwise out of the box cpu - I'm curious how it would work out with a lapped ihs on those platforms. I understand mounting pressure is a thing and that may also skew results, but just curious if that's another avenue for pursuit Either way, loving these recent uploads. Great quality and content.
reply
GamersNexus understanding the flatter cooler examples were not necessarily the best on a given platform for an otherwise out of the box cpu - I'm curious how it would work out with a lapped ihs on those platforms. I understand mounting pressure is a thing and that may also skew results, but just curious if that's another avenue for pursuit Either way, loving these recent uploads. Great quality and content.
reply
lebeaudj
This had me thinking from the opposite way - Lapping of the IHS is a common task so why not make a holder that simulates the mounting of the CPU but still allows you to lap it. This would then leave you with a CPU that has an IHS that is flat when mounted (but has curvature when not). An out of the box solution is great but from an enthusiast aspect I wonder if this would yield the same or better results
reply
This had me thinking from the opposite way - Lapping of the IHS is a common task so why not make a holder that simulates the mounting of the CPU but still allows you to lap it. This would then leave you with a CPU that has an IHS that is flat when mounted (but has curvature when not). An out of the box solution is great but from an enthusiast aspect I wonder if this would yield the same or better results
reply
ChrispyNut
Nooo, stop it, please stop. The better the cooling, the less thermally constrained CPUs will be, the more power Intel will dump into them, the more lethal people's rooms will be in the summer months.
1200W summer space heater (600 each for Intel CPU and nVidia GPU). How dreadful
N.B. I'm kidding for those who missing it. I appreciate the effort and insights, with hope for a quieter rig.
reply
Nooo, stop it, please stop. The better the cooling, the less thermally constrained CPUs will be, the more power Intel will dump into them, the more lethal people's rooms will be in the summer months.
1200W summer space heater (600 each for Intel CPU and nVidia GPU). How dreadful
N.B. I'm kidding for those who missing it. I appreciate the effort and insights, with hope for a quieter rig.
reply
snipermerc
How about adding in lapped cpu's get them suckers SUPER flat, say /- .0002. From what I saw on the Intel front with that precision square you used, those plates are what, /- .005 thou which is a TON of surface irregularity. Also wouldn't mind seeing a thermal paste review of the Shin Etsu phase change style thermal materials and see if they still hold up vs current thermal pastes.
reply
How about adding in lapped cpu's get them suckers SUPER flat, say /- .0002. From what I saw on the Intel front with that precision square you used, those plates are what, /- .005 thou which is a TON of surface irregularity. Also wouldn't mind seeing a thermal paste review of the Shin Etsu phase change style thermal materials and see if they still hold up vs current thermal pastes.
reply
75keg75
So an apparatus (cpu socket) that applies pressure to CPU as a motherboard clamp would. Then lap the cpu. Then place a calibrated grinding pad into the apparatus to then grind the cpu cooler so that they are the essentially the same.
Then the thermal paste will come into play. Ie paste consistency. 3d resin prints can do 8k resin. So 40 Micron sized which any flatter is pointless
reply
So an apparatus (cpu socket) that applies pressure to CPU as a motherboard clamp would. Then lap the cpu. Then place a calibrated grinding pad into the apparatus to then grind the cpu cooler so that they are the essentially the same.
Then the thermal paste will come into play. Ie paste consistency. 3d resin prints can do 8k resin. So 40 Micron sized which any flatter is pointless
reply
zeemon9623
This seems especially applicable to custom liquid cooling where taking apart the CPU block is really easy. Instead of the whole cooler needing multiple SKUs you can just make 3 cold plate designs and basically put on the right one last minute or even sell them separately. And custom liquid cooling is a relatively small market anyway so that economies of scale are less impactful.
reply
This seems especially applicable to custom liquid cooling where taking apart the CPU block is really easy. Instead of the whole cooler needing multiple SKUs you can just make 3 cold plate designs and basically put on the right one last minute or even sell them separately. And custom liquid cooling is a relatively small market anyway so that economies of scale are less impactful.
reply
Add a review, comment
Other channel videos
