How to check flatness of CPUs and coolers - INK and OPTICAL INTERFERENCE methods

[BoggledBeagle]

I think lapping truly flat surface is hard, because when you have truly flat surface the friction of the object with the emery paper is very high. So you need to move the object with very careful and precise force overcoming and fighting with high friction. Most of the time you end up applying the force in an unwanted way.

Another problem is the waste material between the surfaces. Ideally you should wash it away after each move - extremely impractical.

I actually think that the best way is to move the object with the longest possible stroke, because when is the object finaly steadilly moving on the abrasive surface, you can control the force the best.

BTW achieving perfect flatness is IMO not necessary at all, because at least the CPU I was preparing the cooler for is far from flat.

There is another way to test the flatness, when you have reflective or dark surface.

YOU CAN USE CREAM OR MILK instead of ink. I like to use coffee cream because it comes in little containers and you can always have some at hand, and if you will not use it for assessing flatness, you can just pour it in the coffee.

The disadvantage of the milk/cream is that because it is an emulsion, after certain depth you get no information about the layer thickness at all.

Delshay: you might first want to test the flatness of the GPU chip, before becoming too much obsessed with vapour chamber flatness.

 

[delshay]

I think lapping truly flat surface is hard, because when you have truly flat surface the friction of the object with the emery paper is very high. So you need to move the object with very careful and precise force overcoming and fighting with high friction. Most of the time you end up applying the force in an unwanted way.

Another problem is the waste material between the surfaces. Ideally you should wash it away after each move - extremely impractical.

I actually think that the best way is to move the object with the longest possible stroke, because when is the object finaly steadilly moving on the abrasive surface, you can control the force the best.

BTW achieving perfect flatness is IMO not necessary at all, because at least the CPU I was preparing the cooler for is far from flat.

There is another way to test the flatness, when you have reflective or dark surface.

YOU CAN USE CREAM OR MILK instead of ink. I like to use coffee cream because it comes in little containers and you can always have some at hand, and if you will not use it for assessing flatness, you can just pour it in the coffee.

The disadvantage of the milk/cream is that because it is an emulsion, ather certain depth you gen no information about the layer thickness at all.

Delshay: you might first want to test the flatness of the GPU chip, before becoming too much obsessed with vapour chamber flatness.

That's the point I'm trying to put across. The biggest reason why manual lapping to a perfect flat surface is difficult is us. Me, you & everyone else. What's happening is we are moving the downward pressure point to another location on the object being lap. This can lead to errors if the tolerance is not kept tight enough. This is why I think it's best to keep the lapping central as possible to give it a more even lapping across the coldplate, provided the lapping device covers the complete coldplate at all times.

As for sandpaper I would start somewhere within 600 - 800 grit. In this latest lapping it's 800 grit because the coldplate is reasonable flat. Then I go to 1000,1200,1500,2500,5000,7000 & finally 10000.
So, yes I can push all these around with one finger.


Let me give an example about flatness.

Lets say I lap my coldplate in one direction staying within the box I outline. It will be perfectly flat.. ...Flat yes, but it could be a slope flatness like a triangle. This means nothing to the GPU mating surface, but the mounting pressures will be slightly different for each side of the core.

Extreme flatness is important because it has an effect how efficient heat is transferred to the coldplate. But it's even more important on GPU because it has an effect on the GPU Hotspot. Be aware I already have working example here to prove it "preliminary newly lapped coldplate ".

As for cleaning, I brush away the old copper bits with water every time I change to fresh sandpaper. All lapping is actually done with water, so my desk is already wet. Maybe this is why I can lap with one finger.

 

[BoggledBeagle]

I tested DeepCool AG620 cooler. I had a very good impression from the Assassin IV cooler I tested before, unfortunately the base of this cheaper cooler was a spiky cone! You can see there is no hint of some flattening in the middle, it is just a perfect cone with straight sides. The glass was spinning on the tip of the cone.






This is awful. In spite of this, this cooler gets positive reviews.

I flattened it some and now I am using this cooler. It works really well and it is lighter and much less bulky than the Assassin cube.

 

[delshay]

I was right with my lapping technique.. It's around 98% perfect.

I should point out the heatsink moved, so there's more thermal paste where the HBM contact is. So I'm going to explain what are we looking for.

We want to see very little thermal paste where all the contact points are & see more of copper reflection. Plus we need to see a nice wall of thermal paste around the the perimeter of each die. The less thermal paste I see, the better. A perfect flat surface you will not see hardly any thermal paste.

There's two key feature where some user get it wrong, so I'm going to point it out here.

1. When lapping any coldplate, your going to lose mass. So depending on the coldplate, you may lose performance, so the key is not to remove too much copper, but in some cases it may benefit to remove more copper because you will be getting closer to the heatpipe..

2. This is a killer where I think 99% of users got it wrong, "Mounting Force". so i'm going to explain it here.

When you lap "any coldplate" you have already lost mounting force. It's the failure of nearly every single user to restore the original or more mounting force. Most user don''t understand it or don't know how to increase the mounting force for they particular coldplate. .In my case I know how to change/add more mounting force "The Professional Way" ...I will explain how to this another time.

If you get a perfect flat surface contact with the coldplate, you may & most likely need higher mounting force set by the manufacture. That's because you now have a larger surface contact area. So higher mounting force is required for those two reasons pointed out above.


The pictures below not very good, poor lighting & coldplate needs cleaning. it's performing about the same, but the magic is yet to come where full extraction will see where it stands.
Just for the record I used a small mirror to lap the coldplate. ...This project is ready for the next stage


Remember we want to see as much copper surface as possible when flooded with thermal paste. ..I want to pushed-out as much as possible over the side.

 

[BoggledBeagle]

...
When you lap "any coldplate" you have already lost mounting force. It's the failure of nearly every single user to restore the original or more mounting force. Most user don''t understand it or don't know how to increase the mounting force for they particular coldplate. .In my case I know how to change/add more mounting force "The Professional Way" ...I will explain how to this another time.

If you get a perfect flat surface contact with the coldplate, you may & most likely need higher mounting force set by the manufacture. That's because you now have a larger surface contact area. So higher mounting force is required for those two reasons pointed out above.
...

I do not agree with what you wrote here.

A lot of CPU coolers are spring mounted, they are made "foolproof" to function properly in any use scenario. They do not care at all if you remove 0.2 mm from the cooler base. Or if you add 0.2 mm thick heat conductive sheet. The force exerted by them will be nearly identical in all cases.

With the curved base shapes, manufacturers presume that with pressure both parts (cooler and CPU) may deform and become less curved and the contact between them may improve.

When you carefuly prepare really flat mating surface/s, most likely you will need much lower force to make them contact each other well. You go from bruteforcing to a delicate refined mating force tuning process. With well prepared surfaces the force needed may be really small, high enough only to hold the cooler or water block in place, to counter weight of the cooler or the hoses attached.

BTW I would like to see your vapour chamber with a drop of ink under a piece of glass. Shiny does not mean flat.

 

[delshay]

I do not agree with what you wrote here.

A lot of CPU coolers are spring mounted, they are made "foolproof" to function properly in any use scenario. They do not care at all if you remove 0.2 mm from the cooler base. Or if you add 0.2 mm thick heat conductive sheet. The force exerted by them will be nearly identical in all cases.

With the curved base shapes, manufacturers presume that with pressure both parts (cooler and CPU) may deform and become less curved and the contact between them may improve.

When you carefuly prepare really flat mating surface/s, most likely you will need much lower force to make them contact each other well. You go from bruteforcing to a delicate refined mating force tuning process. With well prepared surfaces the force needed may be really small, high enough only to hold the cooler or water block in place, to counter weight of the cooler or the hoses attached.

BTW I would like to see your vapour chamber with a drop of ink under a piece of glass. Shiny does not mean flat.

Don't you get it. If you remove let's say 0.4mn you have already lost mounting force. It's not the same. The springs will not compress to the same level as before, because you have removed some of the copper. The more copper that is removed, the less mounting pressure you will have. Mounting pressure has a big impact how the the thermal paste is pushed/spread out & force into the tiny microscopic pits. That's it's roll to fill the tiny gaps. If I was to change the springs to lighter ones, more thermal paste will remain behind between the die & coldplate, which will result in poorer performance

In my project, the die was completed cover in thermal paste, . What you see is the mounting pressure pushing it over the side. Mounting pressure also play an important roll how heat is transferred. I already know this because one of my cards already has increased mounting pressure & is setting world records, for a air cooled card. ..If I was to go back & change the lapping to my latest method, it will simply set new world records, because now I know I lapped it incorrectly & it shows up with a number of areas around the die filled with thermal pad (PTM 7950). The latest test on spare card shows, there should piratically be no thermal pad whatsoever between the die & coldplate. It's almost void with the naked eye, but i'm sure those microscopic gaps are being filled..

 

[BoggledBeagle]

If the mounting springs are pre-tensioned and the total travel when they are compressed when the cooler is mounted is say 2 mm, shortening this distance by 0.2 mm lowers the mounting force by single digit percents, not even 10.

BTW I am talking about CPU coolers. GPU mounting mechanisms may work differently.

I tested DeepCool AG620 cooler. ... I flattened it some and now I am using this cooler. It works really well and it is lighter and much less bulky than the Assassin cube.

I just noticed that Anandtech tested this cooler:

The DeepCool AK620 Digital CPU Cooler Review: Big, Heavy, and Lit

www.anandtech.com

It has the same fans and body as AG620 and it confirms my finding that this cooler performs well.



They tested it with the awful stock spiky cone base! I wonder how would their test go it they flattened it before testing.

My flattened AG620 performed aprox. 30W worse than the Assassin IV cooler with stock base. Assassin IV with flattened base would be killing it (the heat)!

 

[BoggledBeagle]

I tried to assess flatness of the Corsair XC7 waterblock with black ink, it did not work at all. With blue ink I got this, the base is bulging out stronly:

View attachment 334023View attachment 334024

The black ink failure was also due to the spike in the middle left there after machining the base. Such spike would either make a little indentation in the IHS or crack the chip if you used the block to cool the cpu chip directly. If the spike did not dig into the IHS, it would greatly increase the amount of paste and worsened cooling.

This bulging cooler base shape may work well with Intel CPUs bent in the stock socket mounting mechanism to the ba- I mean convex shape, but for my CPU which is also bulging out due to the mounting frame, this would not work well at all, so I had to get rid of the worst bulge. The job is not perfect, but it is much better than it was before for my use case.

View attachment 334025

I flattened the block a bit more and this time tested the result:



Blue ink looks very good:



Black ink also looks quite good:



And after applying some pressure the result even with black ink is excellent:

 

[BoggledBeagle]

I tested Alphacool CORE 1 waterblock and the result is AWFUL.

You can see the the block has a significantly bulged base.



By chance you can see an interference pattern in the middle:





The area of true flatness is about 1mm wide and less than the crew head tall.

I investigated the block and its performance and found out that it can be easilly improved. I will post it in a separate thread.

 

[A Computer Guy]

I found this interesting and somewhat entertaining video on lapping that has some interesting points on lapping that I haven't seen in other content. I thought you might find this interesting.

 

[BoggledBeagle]

I found this interesting and somewhat entertaining video on lapping that has some interesting points on lapping that I haven't seen in other content. I thought you might find this interesting.

Very interesting, thank you. (WHERE IS THE ENDING?!)

Seeing how his pieces self adhered, his stuff was way flatter than anything I ever done, no wonder, when he used proper lapping technique. Although the aluminium foil slapped on the lapping tool did not look very good.

How I already mentioned a few times, there is little point in trying to achieve perfectly flat surface, when the CPU IHS significantly bends when mounted in the socket with any mounting method.

 

[A Computer Guy]

Very interesting, thank you. (WHERE IS THE ENDING?!)

Seeing how his pieces self adhered, his stuff was way flatter than anything I ever done, no wonder, when he used proper lapping technique. Although the aluminium foil slapped on the lapping tool did not look very good.

How I already mentioned a few times, there is little point in trying to achieve perfectly flat surface, when the CPU IHS significantly bends when mounted in the socket with any mounting method.

Rumor has it he is still trying to get perfect atomic flatness and his CPU was spotted at the The Large Hadron Collider where they are still knocking off the high spots

-- just kidding, but still I too would have liked to see part 2

 

[Noci]

Very interesting, thank you. (WHERE IS THE ENDING?!)

Seeing how his pieces self adhered, his stuff was way flatter than anything I ever done, no wonder, when he used proper lapping technique. Although the aluminium foil slapped on the lapping tool did not look very good.

How I already mentioned a few times, there is little point in trying to achieve perfectly flat surface, when the CPU IHS significantly bends when mounted in the socket with any mounting method.

So if you watched carefully and understood the physics as well as the applied techniques, then what is the point of 'How to check flatness of CPUs and coolers' as with DIY 'lapping' as we know it in the PC scene is light years away from the ultimate goal direct metal to metal contact with optimal thermal conduction (and this will only be valid in a static thermal condition).
Once temperatures start to fluctuate we have already variables like expansion coefficients of different metals, warping and so on, that might have significant impact. Not mentioning that, theoretical, even if you have the perfect bond for thermal condictivity the TIM used under the IHS will be the next bottleneck.

Like the reply you've got from Noctua, all coolers coldplates and CPU IHS's are mass produced within a window of specifications to comply to wide range of variations between them, otherwise they are not able to produce products commercially. Each brand will have done their homework and tested what seemed to be best for them to produce costs efficiently and preferably stand out to the competition.

Like others mentioned before in this thread, it has only a very marginal impact on the CPU temps.

You want to get the best out of your cooling solution, just use the DIY 'lapping' methode as we know it for CPU's and gain some minor results. Just don't make it bigger than it is as it is a dead end, better focus on the cooling solution as a whole and see if there are other improvements that can be made with less effort.

Nevertheless I do appriciate the effort and the time you invested in this project, so don't feel offended .

 

[BoggledBeagle]

I just watched this video:

It turned out, that AM5 CPU is also bulging out like LGA1700 CPU mounted with the optional frame.



I already tried to persuade Noctua to make coolers with truly flat bases, and got reply that they will think about that, but I did not get any commitment from them or any official press statement.

Given the results with AM5 platform, which benefits significantly from a truly flat cooler base:



and the volume of AM5 CPUs sold today and expected to sell in the future, I believe that making coolers with TRULY FLAT BASES is necessary.

Remember, how a stock Assassin IV cooler perfomed really well in my testing? It had a truly flat spot on the base. If the whole base would be truly flat, it would perform even better than that on AM5 CPUs and LGA1700 CPUs mounted with the optional frame.



EDIT:

I jokingly called LGA1700 CPUs mounted with the stock mounting mechanism bananas, it turns out that at 100x magnification, the shape fits the fruit really well...



EDIT 2:

I just realised that AM5 CPUs will be nearly perfectly flat, when mounted with the optional frame (how often do AM5 owners use these now?), and will optimally mate with flat coolers!

From that Gamers Nexus AM5 testing you can clearly see, that the shape of the mating surfaces MATTERS and the resulting temperature differences are clear. I would not call 3°C difference insignificant.

 

[ShrimpBrime]

It turned out, that AM5 CPU is also bulging out like LGA1700 CPU mounted with the optional frame.

-------

I just realised that AM5 CPUs will be nearly perfectly flat, when mounted with the optional frame.

How I see it.

""It turned out bulging nearly ready to split!""

------

Has magical epiphany -

""IHS plates and water blocks are nearly perfectly flat. ""

 

[A Computer Guy]

From that Gamers Nexus AM5 testing you can clearly see, that the shape of the mating surfaces MATTERS and the resulting temperature differences are clear. I would not call 3°C difference insignificant.

A 3°C difference may not be insignificant but getting that extra 3°C seems to be quite cost prohibitive especially if in day to day use the temperature is well within the operating range of the CPU anyway , assuming you are using an appropriate cooler to handle the wattage of your chip.

Just an opinion but, outside of niche use cases, it's totally unnecessary to create AM5 coolers with truly flat (define flat) base plates when currently the chips performance with thermal paste is already very competitive in the market despite the thick unflat IHS. For average Joe 3°C doesn't matter. For the performance enthusiast sure it matters but it's going to cost more, possibly more for average Joe.

 

[BoggledBeagle]

A 3°C difference may not be insignificant but getting that extra 3°C seems to be quite cost prohibitive especially if in day to day use the temperature is well within the operating range of the CPU anyway , assuming you are using an appropriate cooler to handle the wattage of your chip.

Just an opinion but, outside of niche use cases, it's totally unnecessary to create AM5 coolers with truly flat (define flat) base plates when currently the chips performance with thermal paste is already very competitive in the market despite the thick unflat IHS. For average Joe 3°C doesn't matter. For the performance enthusiast sure it matters but it's going to cost more, possibly more for average Joe.

I believe that AM5 CPUs are set up to hit thermal limit, improving thermal transfer thus automatically improves performance. Any improvement would be surely welcomed by 3D CPUs owners, 3D CPUs struggle with high temperatures.

Steve mentioned the temperature difference also with connection to tight competition between cooler manufacturers, when coolers tend to perform similarly to each other.

Regarding the "flat" definition: it does not seem that cooler manufacturers have novadays problems with machining precise shapes on the cooler bases. See those shapes I examined above - exact whole cone and truncated cone, on which the cut off part is (very) flat. There will not be any problems with defining and manufacturing "satisfactorily flat"coolers. Even the Intel CPU is nicely flat, before you mount it in the socket, and the IHS is probably just pressed flat.

From that Gamers Nexus AM5 testing you can clearly see, that the shape of the mating surfaces MATTERS and the resulting temperature differences are clear. I would not call 3°C difference insignificant.

I just learned that Roman in this video:

noticed that EK block is strongly bulged, he then machined it flat and the improvement was 2,5°C.

 

[A Computer Guy]

I just learned that Roman in this video:

noticed that EK block is strongly bulged, he then machined it flat and the improvement was 2,5°C.

View attachment 346490

Are you trying to say the block is bad? (I remember the pricing being outrageous in proportion to the performance.) Remember part of this video is about Roman trying to sell you his contact frame. Take it with a grain of salt.

Correct me if I'm wrong but doesn't the use of the contact frame contradict using EK's block "as-is". In other words without the contact frame the IHS of the Intel CPU concavely bends so EK's block attempts to take this into account by not being flat (convex shaped) attempting to match the predicted shape of the IHS? Oh wait later in the video Roman explains from the response from EK "...the design is made for not using it with a contact frame..." also Roman explains the block in question was made before socket 1700. The takeaway I think you might consider from the video is "contact frame good, EK block bad, oops actually block isn't compatible with contact frame - but still contact frame is good please buy it"

Summary - Yes the shape matters (and mounting pressure) but also you need to take into account if the block you intend to use may have been designed to accommodate IHS characteristics at a different period of time.

 

[BoggledBeagle]

Are you trying to say the block is bad?

The block with its shape was bad for a lot of CPUs on the market, for example all AM5 CPUs, and Intel CPUs mounted with the optional frame.

The good thing was/is that EK offered/offers an optional base for the block, which is flat.

Unfortunately it seems that a lot of other cooler/block manufacturers just make them bulging out with no choice and option for customers to buy a flat version.

Choices of some manufacturers seem really bizarre to me, for example to make a base in the shape of an exact cone - pointy cone.

 

[BoggledBeagle]

Noctua introduced three different cooler base shapes. It is great that consumers now have a choice, but I have no idea why the flattest shape is still not truly flat, when AM5 CPUs and LGA1700 CPUs mounted with the optional frame are bulging out and the optimal cooler base shape for them is FLAT.

 

[BoggledBeagle]

I got a new Noctua NH-D15 G2 LBC - the flattest version to see how flat it really is.

I liked the real screwdriver, disliked the colour scheme of the fans (disgusting and disturbing), it performed 8°C better than my current Noctua NH-U12S redux (ground really flat) with 180W heat load, which is three times smaller, lighter and cheap. I also noticed a buzzing sound coming from one of the fins with high fan speeds. With both fans running (not with either fan running alone) there was also a mechanichal ringing (2 or 3 times per second) coming from the cooler body, not acoustical.




Still far from flat, the blue does not get too blue too quickly though going to the edges...

 

[A Computer Guy]

I got a new Noctua NH-D15 G2 LBC - the flattest version to see how flat it really is.

I liked the real screwdriver, disliked the colour scheme of the fans (disgusting and disturbing), it performed 8°C better than my current Noctua NH-U12S redux (ground really flat) with 180W heat load, which is three times smaller, lighter and cheap. I also noticed a buzzing sound coming from one of the fins with high fan speeds. With both fans running (not with either fan running alone) there was mechanichal ringing (2x per second) coming from the cooler body, not acoustical.

View attachment 355501
View attachment 355502
View attachment 355503
Still far from flat, the blue does not get too blue too quickly though...

There were 3 versions of this if I recall. 1 favoring AMD, 1 favoring Intel, and 1 middle of the road.

 

[BoggledBeagle]

There were 3 versions of this if I recall. 1 favoring AMD, 1 favoring Intel, and 1 middle of the road.

HBC version is for Intel CPUs strongly bent in sockets with stock mounting mechanism, LBC version is for AMD CPUs and Intel CPUs mounted with optional frames.

I ordered all three to compare but the total price was so absurd, that I got just the most interesting version.

 

[BoggledBeagle]

... I also noticed a buzzing sound coming from one of the fins with high fan speeds.

I just identified that one of the lowest large fins buzzes in this spot on my sample:



If I was keeping the cooler I would just bend these little ears so that they touch the fin above:

 

[BoggledBeagle]

With everybody excited about the new Arrow lake CPUs I thought I should check, how flat they are and what do they do in the socket.

A new CPU out of the socket surprisingly is not flat, but it has a bulge in the middle.



In socket, which is the new reduced load type, the CPU again bends significantly, in this case twists a bit too.



I expected a little bit more from the new type of ILM.