Convex after lapping?

[chaloux]

I have lapped a few CPUs now, first my P4 2.8ghz, which was perfectly flat and worked beautifully after.

The second CPU, which is an e4400 (current rig) was concave by this definition.

So, after some lapping with 400/800/2000 grit sandpaper, I put the CPU back in. To my surprise, I noticed no temp drop at all (with a Scythe Mini Ninja cooler).

My train of thought was, "well, it didn't get worse, so I can't really complain"

Then I read a few things about how to apply AS5, and so I decided to take my CPU out and re-apply AS5. Well, I checked the processor again, and again to my surprise I find that the processor is now convex. I mated the cooler and processor without any paste, to see what the contact was like. Sure enough, there were gaps on the edges as only the middle was making contact. I rotated the processor to make sure it wasn't only one edge, and it was convex no matter which side I looked at it.

So I lap the processor again, this time with only 400/800 grit. No difference. The only thing I can think of is that when I lap the CPU, I don't go over EXACTLY the same path: it varies by a centimeter or two, thus rubbing the edges of the CPU with fresher, raised sandpaper, which ends up sanding them down more.

So I apply my AS5 in a straight line along the cores, fire it up, and what do you know - my temps are now worse by 3 degrees. This is where I sigh. I ordered a Thermalright Ultra 120 Extreme to be done with it, and I will attempt to lap the e4400 once more, as carefully as possible to minimize the variation so the whole CPU is sanded by the same paper.

Any thoughts on this? I lapped the CPU first by only going back and forth and not rotating it, and then the second time I tried rotating it 4 times to make sure each side was done evenly.

 

[Nenu]

If you can fit the sandpaper to a flat piece of wood/melamine you should be able to take off the high edges no problem.
It may take some time so be patient.

 

[RangerSVT]

I'm sorry but I neglected to see what type of surface you used for this project?

 

[chaloux]

Sorry, I used a 8x6 pane of glass. I assumed it would be flat and not the cause, and therefore didn't check it, but I suppose it could be the culprit. I will check it when I am at home tomorrow morning.

 

[Nenu]

This is tricky and can damage the CPU pins easily so may not be a viable solution...

If you can have the sandpaper on a flat desk surface, you can sand the cpu by moving it over the sandpaper.
I would only advise doing this if you have an old dead motherboard you can steal the cpu socket from, to protect the CPU's pins.

This will prevent the sandpaper being as loose which could cause bunching of the sandpaper in the centre and will stop you applying too much pressure to one edge.

Otherwise, make sure the sandpaper is as tight to the glass as possible while sanding.

 

[Thor]

I'm sorry but I neglected to see what type of surface you used for this project?

indeed, should be useing something that is flat to begin with, and not just a 'table top' or the likes, i personaly use a piece of glass out of a picture frame when im at home, (which i back with a wooden cutting board as the glass is rather thin and to much pressure will crack it), put a little water , just a few drops, between the board anfd the glass, it should form a slight suction to help keep the glass from moveing around. if im away from home and find a badly matched cooler, i have used marble and granet tiles, as well as glass cutting boards in a pinch.

i think the biggest problem you are haveing might be paper curling, as you'er useing the paper your only useing the center try rotateing the paper 90° every few passes to help keep it from curling to much, as the paper curls it rubs more on the edges and makes it convex, you could try to tape the curling edges down, but tape+water=no good. (to be honest i have yet to really figure out a good way to deal with paper curling, aside from just going for a fresh piece.)

remeber, the work is done on the lower number grits and the pretty is done on the higher number grits, any thing above 800 isnt really needed if your just going good contact, if you must go to 1200 and skip the 2000 and go straight for notebook paper or oak tag (i prefer oaktag) after a few hard passes, the high spots will shine and the low spots will still be dull, back up to 800 grit, work them down, 1200, oaktag, you get the idea.

 

[Arcygenical]

This is tricky and can damage the CPU pins easily so may not be a viable solution...

If you can have the sandpaper on a flat desk surface, you can sand the cpu by moving it over the sandpaper.
I would only advise doing this if you have an old dead motherboard you can steal the cpu socket from, to protect the CPU's pins.

This will prevent the sandpaper being as loose which could cause bunching of the sandpaper in the centre and will stop you applying too much pressure to one edge.

Otherwise, make sure the sandpaper is as tight to the glass as possible while sanding.

Conroes have no pins

And I like to spray the glass with soapy water, then apply the sandpaper, then tape it down (wipe down the edges of the glass that's not covered with paper, obviously) with duct tape.

Really stops it from moving IMO.

 

[Thor]

Sorry, I used a 8x6 pane of glass. I assumed it would be flat and not the cause, and therefore didn't check it, but I suppose it could be the culprit. I will check it when I am at home tomorrow morning.

heh i wrote my above post write after the one i quoted, regardless...

you need to remeber that glass will take the form of what ever is under it, as i said above, not only your primary surface has to be flat, your seconedary support surface has to be relativly flat as well, useing water between the surfaces will help minimise the small stuff

 

[chaloux]

Thanks guys. What I had done to secure the paper to the glass was wrap it around the edges of the glass and tape it on the underside, but it would still get loose.

I will try using water/soapy water to suck the paper to the glass, that was a good idea that I never thought of. I'll post the results tomorrow.

 

[chaloux]

you need to remeber that glass will take the form of what ever is under it, as i said above, not only your primary surface has to be flat, your seconedary support surface has to be relativly flat as well, useing water between the surfaces will help minimise the small stuff

Actually this may explain why I'm having this problem, especially since I wrapped the paper around, raising the eges, which would make the glass bend under pressure.

 

[Thor]

Thanks guys. What I had done to secure the paper to the glass was wrap it around the edges of the glass and tape it on the underside, but it would still get loose.

this would be part of your problem, the paper under the glass is causeing high spots on the edges and makeing your glass flex slightly, just enought to make what ever your working on convex

 

[Thor]

Actually this may explain why I'm having this problem, especially since I wrapped the paper around, raising the eges, which would make the glass bend under pressure.

lol

 

[spine]

I always say this in threads about lapping:

Hand lapping WILL make a surface slightly convex.


There's no 2 ways about it. It is simply impossible to ensure that the pressure applied to the object will be applied to the dead centre of the base/IHS 100% of the time as you lap it, and so you'll always end up with the surface being slightly convex.

Only a machine has the precision needed to get true flatness from lapping. However, a hand lapped and slightly convex surface is often preferable to the state some heatsinks and IHSs start off as.


Ultimately, don't worry about it, it's not a problem.

 

[ASIA911]

Make sure to use a waterproof marker and draw an X across the heastink from end to end. This way you'll see if it's flat or not while sanding. Keep drawing an X on it until you have a flat surface and don't see any marker left on the heatsink.

Flatness if MORE important than a shiny heatsink.

 

[Arcygenical]

A bit of convexity never hurt anyone.

Look at bowing the waterblock.

 

[Rabidfox]

I had similar problems the first time I lapped. I used plexi-glass. Horrible Idea. Then I used a 3/8" thick piece of tempered glass. It was perfect after that.

 

[Michael Daly]

Hand lapping WILL make a surface slightly convex.


There's no 2 ways about it. It is simply impossible to ensure that the pressure applied to the object will be applied to the dead centre of the base/IHS 100% of the time as you lap it, and so you'll always end up with the surface being slightly convex.

Only a machine has the precision needed to get true flatness from lapping.

And yet opticians can hand lap an optical flat to RMS errors in flatness on the order of 25 nm.

Use a flat base like a thick piece of float glass. Glue down the sandpaper with a spray glue like 3M Super 77. Then use minimal force. Do not press down hard - let the sandpaper cut on its own.

 

[BillParrish]

If the mating surfaces where not pretty "bad" to start with, many times the results/improvement are under the error tolerance of the temp measuring equipment, even the DTS in the c2d's have a tolerance (I cant remember exactly, one deg C +/- I want to say, it can be found in the Intel docs if someone wants to correct/check me) and variations in AS application and mounting will maks small differences that help mask what is really happening. I lapped my cpu and saw no change with any temp measuring software (and I have most of the utilities) and it was apparent after a few passes on the sandpaper it was not close to flat.

So all I can determine is that even with an IHS that is not good and flat, if you have a good thermal interface, lapping will not show a difference that you can easily measure.

Its always a good thing to do on any high performance setup, just like redoing all the thermals on the board. You might not see any big difference or any difference at all, but its part of the craft and a [H]ard overclocker will do it as a matter of routine.

 

[spine]

And yet opticians can hand lap an optical flat to RMS errors in flatness on the order of 25 nm.

Use a flat base like a thick piece of float glass. Glue down the sandpaper with a spray glue like 3M Super 77. Then use minimal force. Do not press down hard - let the sandpaper cut on its own.

No matter how flat the under surface is you can't avoid the inevitable. As you drag/push the object, the sandpaper will ripple ever so slightly causing a bias in abrasion. Also, as you push/drag the front side will bear more of the force and be ground more. The only way to avoid any of these little factors is to somehow have a dead flat piece of totally rigid sandpaper on a dead rigid flat surface and have the object mechanically held in place so as to only apply force to the dead centre of the object as it moves. Only a machine can do this.

I'm talking very technically here. Sure in practice you can get it pretty damn flat, and more importantly, flat enough top be worthwhile hand lapping. But fundamentally, hand lapping=convex. At the least if you only drag/push in one direction you'll create a beveled surface rather than a rounded convex one, but that's not really preferable for a heatsink base.


Despite what I write about hand lapping, I always do it. I realise I'm creating a slightly convex surface, but that's ok since I only have to worry about a small area of the heatsink base since I always de-lid my cpu's.

 

[cyclone3d]

And this is one of the reasons I use an electric orbital sander to lap heatsinks and processors.

It is much faster and easier to get right than hand lapping.

Also... wetting the sand paper helps a lot.... and then rinsing it off when it starts to get clogged.

 

[Michael Daly]

Only a machine can do this.

I repeat - opticians can hand lap an optical flat to RMS errors in flatness on the order of 25 nm. That's 1/20 of the wavelength of visible light.

If you glue a sheet of sandpaper down, it will not ripple. If it ripples, you have not glued it down properly. That's why I said to use a spray glue - get a thin, uniform layer of glue.

 

[spine]

I repeat - opticians can hand lap an optical flat to RMS errors in flatness on the order of 25 nm. That's 1/20 of the wavelength of visible light.

If you glue a sheet of sandpaper down, it will not ripple. If it ripples, you have not glued it down properly. That's why I said to use a spray glue - get a thin, uniform layer of glue.

What do you mean by "an optical". Are you talking about a lens? Lens's aren't flat and only need be smooth. No doubt they can hand lap something to be really smooth. But that doesn't make it flat. You'd be surprised how warped an eye piece can be without a person knowing. Definitely not a good test of flatness.

And how are these guys moving this 'optical' then? Just dragging this thing by hand across a surface? As I've explained, that method inevitably creates a convex surface. You have to apply pressure to the dead centre of the object 100% of the time as you move it assuming you have an impossibly flat and ripple free friction surface. I don't know how else I can explain it. It just can't be done by hand. I will accept that you can get it done with acceptable error, but it's never gunna be truly flat. If you truly believe you can hand lap a heatsink base using this method to a 25nm level of flatness you are simply insane!


Tell you what, prove me wrong; get 2 old heatsinks (60mm² base at the least) and lap them by hand to the best of your ability to a pretty decent finish. Then put them together base to base and hold them up to a light. I guarantee you you'll see gaps at the outer edges. Guarantee it.

 

[Michael Daly]

What do you mean by "an optical". Are you talking about a lens? Lens's aren't flat and only need be smooth.

Not all lenses are curved on both sides (in the image here, the flint is flat on the right side) and some optical systems need components that are flat on both sides. An optical flat can be used to test the quality of another lens by using it as a reflecting surface and observing interference patterns.

And, yes, they do this by hand. To achieve this level of flatness, they use two surfaces and switch the flat from one one surface to the other - any error that starts to develop from one grinding surface gets removed by the second. Using the lapping techniques used on heat sinks does not require this level of care.

Beginners in making optics frequently have a problem with rounding of the edges. However, they can learn to eliminate this.

You say you need a machine to get something flat - not true. You need to set up your environment correctly and learn to do the work correctly. You don't need to put the force in the exact centre, you need to learn to use little force and let the sandpaper do the cutting. I've lapped steel components down to micron-grit abrasive papers and don't suffer from rolled edges, so I know it can be done. It's slow, but it's not impossible.

 

[spine]

I don't doubt getting something reasonably flat by hand isn't impossible, I'm just being very technical about it. I'm not saying a hand lapped heatsink base would end up blatantly convex to the naked eye, but that technically it would be. Even if you apply only light force, it's still force and thus would have to be applied to the dead centre in order for a convex shape not to be formed. It's a simple matter of physics.

I think we can agree that it doesn't really matter for heatsink bases, and yes, applying light force will be better albeit at a great cost of time and labour. I was just arguing this on a pedantic level.