IHS Surface Flatness

[Masterjuggler]

I'm specifically asking about the CPU IHS, but this could really apply to any thermal interface...

 

Has anyone gotten any real measurements on how IHS flatness affects heat transfer to the cooler, assuming some thermal compound applied? I'm not talking about your regular "check the paste spread pattern for high spots," I mean has anyone actually taken a CPU to a surface plate with a dial test indicator and taken actual measurements of flatness? 

 

I did see a video about a YouTuber called Penrowe lapping a Ryzen CPU, like by the machinist definition of lapping, and sending it to someone for testing (probably Gamers Nexus). I'm interested to see how that works out, but I haven't seen anything happen with it yet. Apparently they got it down to 0.3 microns across it's surface, or about 10 millionths of an inch.

 

I'm not entirely sure how an IHS is made right now. If it's stamped, well, I can understand why I see people going and sanding it down. Machining would produce a much flatter, if rougher, surface. The new ryzen 5000 CPUs are soldered, so this really only applies to the top surface at this point, where it meets the cooler's cold plate. The less contact there is, the more space the thermal compound has to fill, and the worse it will perform.

 

I'm planning a PC build right now (as is everyone else LOL) with the Ryzen 5600x. I'll prob start getting parts for it in Jan or Feb once this round of GPUs are actually available. I have a surface plate and a tenths indicator (0.0001") myself, so while I don't really intend on sanding or lapping, I'm interested to check out the one I get.

[emosun]

lapping is a pretty normal thing people used to do back in the day with cpu's. it's kinda funny how people think stuff lives on the internet forever but then one day I see a post about how lapping might be a new thing. lol

oh and to my knowledge it does a bit but not much , remove the ihs all together if you wanna see any sorta real difference in numbers

[Masterjuggler]

I guess I was a bit too long winded and didn't really get my point across, sorry. 

 

First, people haven't been lapping, they've been sanding. Sanding something on a flat surface makes it visually flat, but not compared to a lapped surface like that of a gauge block. Lapping is where you take a softer material than the part to be lapped, charge it with loose abrasive material, then rub the two together. Lapping copper is extremely difficult and as far as I know, has only ever been done by Penrowe. Here's the video: https://www.youtube.com/watch?v=sQw20rHvpRw

 

The point of this thread is asking if anyone has actually taken measurements of surface flatness, whether they sanded or lapped. I'm curious to know if there are actually any standards for IHS surface flatness, either in the industry, or in overclocking groups for example.

[emosun]

I mean whats the end goal here? The chip isn't going to run any cooler than just ditching the IHS. You can get the surface as flat as you want but it still is gonna transfer heat worse that an object that doesn't need to transfer in the first place.

Even then , I think a stock cpu using colder water will run cooler than a lapped cpu running hotter water. think it'll all just come back to the same old same old , turn on the room AC.

[Masterjuggler]

Well yeah, there's a lot of things you can do to make a processor run cooler. Why don't we all just use liquid nitrogen? Why is anyone doing any of this crap to their computers at all if in a year or so we get faster hardware anyway? I think you're missing the point here.

 

Like I've been saying, It's a topic of curiosity. I want to know if anyone has taken an actual measuring device to a processor before grinding it down, measured it again, then seen how it affects their thermal transfer. 

[Guest]

13 minutes ago, Masterjuggler said:

I guess I was a bit too long winded and didn't really get my point across, sorry. 

 

First, people haven't been lapping, they've been sanding. Sanding something on a flat surface makes it visually flat, but not compared to a lapped surface like that of a gauge block. Lapping is where you take a softer material than the part to be lapped, charge it with loose abrasive material, then rub the two together. Lapping copper is extremely difficult and as far as I know, has only ever been done by Penrowe. Here's the video: https://www.youtube.com/watch?v=sQw20rHvpRw

 

The point of this thread is asking if anyone has actually taken measurements of surface flatness, whether they sanded or lapped. I'm curious to know if there are actually any standards for IHS surface flatness, either in the industry, or in overclocking groups for example.

So I normally don't go past 3000 grit sanding paper. But on Crankshaft journals 8000 grit wet sanding paper and finishing with cloth. 

 

But at 3000 grit, on a thick machined flat mirror, you can get a cpu to suction cup to a waterblock. 

 

The above video is way overkill imo.

And I've wasted many hours of lapping time lol... So turned to de-lidding.

Started doing that back in 09' when AMD started soldering processors. Then I lapped silicon instead. 

[Masterjuggler]

Alright, so talking about grits gives you some idea about surface roughness. That's also an interesting topic, but wasn't really my original question. I'm interested to know if anyone has taken measurements of surface flatness. 

[Guest]

2 hours ago, Masterjuggler said:

Alright, so talking about grits gives you some idea about surface roughness. That's also an interesting topic, but wasn't really my original question. I'm interested to know if anyone has taken measurements of surface flatness. 

Measurements beyond a 1000th of an inch is kinda moot point when your gonna stick thermal compound between your two surfaces. So any machine flat shop square edge will do. In the picture below, you can see the slight curve to the IHS plate, even though it looks like a straight edge, punched through a die on a press, it's not really so straight. The Cpu die is flatter after a good "sanding" but really no measurement needed to be taken. 

 

And off to de-lidding and "lapping ... or sanding" I found it fun to "polish" the surface I suppose you would say. But not with sand paper, but with soft cloth. Not removing material more or less making it smooth. 

 

[Masterjuggler]

7 hours ago, ShrimpBrime said:

The Cpu die is flatter after a good "sanding" but really no measurement needed to be taken. 

I seriously thought I'd be able to get a yes or no pretty easily, but this is getting pretty frustrating because I just can't get an actual answer. What you're saying is no one has actually taken any real measurements, and everyone is just going "duuurrrr it looks flat enough I think." Is that correct?

 

Maybe it really doesn't matter. Problem is we don't actually know because no one seems to have taken the time to find out. How can people do stuff like sand down their processors without even knowing how they're changing it? There's a Gamers Nexus video where he had Kingpin sand a 9980XE and got about 3C lower peak on ambient air. I'd guess that it was only a 1-2 thou gap to begin with, but he didn't measure the surface flatness, so we don't actually know.

[freeagent]

It was probably more than 1-2 thou, IHS is very soft. Probably more like .015 or so. I cant say for sure because I haven't measured lol. Just throwing numbers around I guess.

 

Also, you don't just sit there and run your CPU back and forth on some sand paper until you see copper. That would be sanding, not lapping.

[Guest]

32 minutes ago, Masterjuggler said:

I seriously thought I'd be able to get a yes or no pretty easily, but this is getting pretty frustrating because I just can't get an actual answer. What you're saying is no one has actually taken any real measurements, and everyone is just going "duuurrrr it looks flat enough I think." Is that correct?

 

Maybe it really doesn't matter. Problem is we don't actually know because no one seems to have taken the time to find out. How can people do stuff like sand down their processors without even knowing how they're changing it? There's a Gamers Nexus video where he had Kingpin sand a 9980XE and got about 3C lower peak on ambient air. I'd guess that it was only a 1-2 thou gap to begin with, but he didn't measure the surface flatness, so we don't actually know.

Lol. Well then. NO. 

 

I do believe there was measurements taken on thickness of a die when a lapping proceedure was done to make it thinner.

 

Think it was an Intel processor. But NO, the measurement was not taken to see how flat it was.

 

Yes lapping on a machine flat mirror is "flat enough". I've stuck many processors to a waterblock and suction cupped the cpu and lifted off the desk. That's flat enough in my book.

 

Edit: Durrrrrr its flat enough!!! No measurements needed. 

[Masterjuggler]

 

2 minutes ago, freeagent said:

It was probably more than 1-2 thou, IHS is very soft. Probably more like .015 or so. I cant say for sure because I haven't measured lol. Just throwing numbers around I guess.

 

Also, you don't just sit there and run your CPU back and forth on some sand paper until you see copper. That would be sanding, not lapping.

It could be more like 15 thou because it is just nickel plated copper, you're right. I'm interested to know though. You're also right that there is a difference between sanding and lapping, that's what I've been saying. I didn't really intend to harp on semantics, but it seems I have lol.

 

3 minutes ago, ShrimpBrime said:

Lol. Well then. NO. 

 

I do believe there was measurements taken on thickness of a die when a lapping proceedure was done to make it thinner.

 

Think it was an Intel processor. But NO, the measurement was not taken to see how flat it was.

 

Yes lapping on a machine flat mirror is "flat enough". I've stuck many processors to a waterblock and suction cupped the cpu and lifted off the desk. That's flat enough in my book.

Thank you, I appreciate it. I agree, regular sanding on a piece of glass to take off the visible high spots is probably more than flat enough to stop seeing any improvement. It's a shame no one has checked how flat it was before doing any sanding to compare, and that's my issue.

[Guest]

4 minutes ago, Masterjuggler said:

 

Thank you, I appreciate it. I agree, regular sanding on a piece of glass to take off the visible high spots is probably more than flat enough to stop seeing any improvement. It's a shame no one has checked how flat it was before doing any sanding to compare, and that's my issue.

But compare it to what? The un-sanded cooler that's going to be attatched? 

Or sand a heat pipe cooler and compare it's flatness to the IHS surface?

OR just to see how flat it is compared to a gauge? 

 

I think this one technicality, for the less than 1% of people in the world that have done it, isn't interested enough to bother with a flatness measurement.

 

I sand IHS plates.

I sand waterblocks. 

I sand then polish core dies. 

 

I think past a certain grit of wet sanding paper, the paper is softer than the copper. Then this wouldn't be called sanding anymore, it would be polishing, or if you want to call it lapping you can.

 

But as far as I know in the OC community, it's always been called lapping despite any technicality behind the word.

[Masterjuggler]

Like I just said, to compare the unsanded surface to the sanded surface.

 

People usually sand it for two reasons, both with the same end goal of reducing the amount of empty space: To reduce surface roughness (going up the grits to polish), and to increase surface flatness (using a flat backing like glass). 

 

It really doesn't matter what you call the processes, call them Fred and George for all I care. Whatever the names, they (what I've been calling sanding and lapping) are very different processes with different end goals. In any case, it really wasn't the point, I initially brought it up as the one example I knew of where someone took real look at flatness.

[Guest]

3 minutes ago, Masterjuggler said:

Like I just said, to compare the unsanded surface to the sanded surface.

 

People usually sand it for two reasons, both with the same end goal of reducing the amount of empty space: To reduce surface roughness (going up the grits to polish), and to increase surface flatness (using a flat backing like glass). 

 

It really doesn't matter what you call the processes, call them Fred and George for all I care. Whatever the names, they (what I've been calling sanding and lapping) are very different processes with different end goals. In any case, it really wasn't the point, I initially brought it up as the one example I knew of where someone took real look at flatness.

I'm sorry you didn't really find the answer you're looking for. This is a very niche procedure I think.

Some of the overclocking communities might have some type of results. I'd start looking at OCN forums maybe.

[Masterjuggler]

Makes sense, I hadn't really seen anything about it but wasn't sure if I just missed it. Anyway, thanks for dealing with me lol.

[freeagent]

Again, I’m just throwing some numbers around here but..

 

If current AMD IHS is flatter than Intel, would make sense that top Intel coolers don’t work so well with new AMD CPU’s. So looking at my True Spirit 140 Power, I estimated I would be milling off about 15-30 thou off the top so it wouldn’t be as convex. I probably would have tried in 5 thou increments, because I still like the cooler and didn’t want to mow too much off. Again, just an estimation as I don’t actually own a modern cpu. I would have made a shim to take up the slack. I also got laid off from the machine shop I worked at due to current events.. it was pretty slow. So I don’t know when or if I’ll be going back. Hopefully I will, but manufacturing has takin a real shit kicking in my town. 
 

I would need to have the CPU in hand and maybe rub an indicator over it or at the very least a height gauge.

[Guest]

Dug out some old pictures. AMD IHS plates really haven't changed at all in many many years.

 

 

And that's only 1500 grit, I would call this sanded. You can still see the scratches by naked eye.

 

I think a measurement can be difficult. You see the edges are high, indicates concave center. 

But there's a lot of Nickel plating in there. I'd be interested to see the thousands of an inch measurement....

Didn't do much lapping on IHS plates after this. It was easier to just de-lid them, soldered or not. (Long Before Die-mate was ever invented lol)

[freeagent]

She’s purty.

 

All this talk is making me want to do my 3770K lol. For something that puts out no heat that thing can get crazy hot.

[Masterjuggler]

@ShrimpBrimeWhen I build my new system in a few months I'll be checking it out, if for no reason other than to sate my own curiosity. I don't really plan on sanding the new chip any time soon, I like warranties haha. The system I'll be replacing though has an old fx-6300 that I'm also interested to check out. I might actually try running some tests on that to see if sanding flat does anything meaningful. 

 

Taking a measurement is probably less difficult than you might be imagining. You basically lay out a grid of some arbitrary size on the CPU, then run an indicator over it to see relative change in height. It can get right up to the edges and corners too. This is often done on machine tools to mate precision bearing surfaces, like a lathe cross slide. 

 

@freeagentMan, sorry to hear about the job. I wonder if all IHSs are just stamped out, if so then I can see how none of them would be very flat. Probably the easiest way to make one flat is to surface grind it, but I don't have access to a surface grinder.

[freeagent]

All good my man. I’ve dabbled in a little of this and that so if they can’t find you handsome at least they can find you handy
 

It would be nice to go back, we were a small shop putting out a world class product that we were proud of. We shall see..

[Masterjuggler]

 

8 minutes ago, freeagent said:

if they can’t find you handsome at least they can find you handy

Alright Red Green, lol. Hope things work out.

[freeagent]

I don’t think they are stamped from a sheet, I think they may be laser cut and then pressed, but I can’t say for sure. If they are pressed from a sheet they are using some mighty tight tolerances in their dies. I’d be interested in a manufacturing video if there was one.

[Guest]

22 minutes ago, freeagent said:

I don’t think they are stamped from a sheet, I think they may be laser cut and then pressed, but I can’t say for sure. If they are pressed from a sheet they are using some mighty tight tolerances in their dies. I’d be interested in a manufacturing video if there was one.

If I where to take a wild guess, they are stamped out. It's one of the fastest ways to make plating out of sheets of material. 

 

The nickel plating process I think is to make the etching easier to read. It's horrible for conducting heat.

 

But people still use IHS plates though? Blows my mind! XD