Lapping for "flatness" versus "nickel elimination" versus "thickness"

[BonzaiDuck]

This could easily go in the CPU forum, but it's a cooling topic.

The recent IB (and now Haswell) de-lidding experiments showed some significant differences in thermal improvement, either by replacing the IHS lid or removing it entirely.

In earlier years, some people had lapped off the entire top layer of their IHS to expose the solder on the die (and hopefully nothing else).

Otherwise, any focus on the IHS is a matter of a copper intermediate layer and its thickness.

Consider the old "TCASE" spec. For certain processors (and TDPs) it had been either measured or estimated at about 10C below the core average, give or take. But it depends on the thickness of the layer.

So would eliminating half the thickness of the IHS through lapping yield a 5C improvement?

Lapping is worth at least a few degrees; nickel elimination is worth a few degrees. So I'm wondering about doubling or tripling the time I spend lapping a processor. And what is the best indicator that it's time to stop?

Probably a good reason to use a digital caliper . . .

 

[krnmastersgt]

I actually wonder if lapping the IHS is worth it anymore without also lapping your CPU cooler's contact plate. Ideally both would be flat but I believe the IHS is designed (or just always turns out to be) slightly concave and due to that fact several CPU cooler manufacturers actually purposefully make their plates convex to get a better overall fit on it (also because the contact in the center is more important).

I haven't lapped anything since my Q6600 though because I can't sell it without any of the identifiers on it which is a shame since all I've had since then is Ivy and Haswell. I'd go to the bare copper if I was going for the best possible lapping, but I'd still probably de-lid and re-mount the IHS first just because of the enormous temperature drops we see with the past 2 gens. There's also always bare die but I'm too afraid I'd crack the die from mounting a cooler on too tightly

Also speaking of my Q6600, haven't seen you since I was originally building around/OCing it Bonzai, good to see you're back I believe I used your old thread on the best way to mount the original Intel quad's and the application of the TIM for the best possible temps so good to see you're back at it!

 

[BonzaiDuck]

I actually wonder if lapping the IHS is worth it anymore without also lapping your CPU cooler's contact plate. Ideally both would be flat but I believe the IHS is designed (or just always turns out to be) slightly concave and due to that fact several CPU cooler manufacturers actually purposefully make their plates convex to get a better overall fit on it (also because the contact in the center is more important).

I haven't lapped anything since my Q6600 though because I can't sell it without any of the identifiers on it which is a shame since all I've had since then is Ivy and Haswell. I'd go to the bare copper if I was going for the best possible lapping, but I'd still probably de-lid and re-mount the IHS first just because of the enormous temperature drops we see with the past 2 gens. There's also always bare die but I'm too afraid I'd crack the die from mounting a cooler on too tightly

Also speaking of my Q6600, haven't seen you since I was originally building around/OCing it Bonzai, good to see you're back I believe I used your old thread on the best way to mount the original Intel quad's and the application of the TIM for the best possible temps so good to see you're back at it!

I pop in here at least several times a year when I'm not obsessed with new projects. Right now, some may even view me as a pest! I've really got the bug this year . . .

All those lapping exercises were a matter of collecting little marginal improvements that added up. The TDPs have only dropped except for the monster-hex-core 130W processors, while the silicon die has shrunk. I was sure from my attention to measurement that getting rid of the nickel plate was good for a couple C degrees.

And -- yeah -- if you're going to do it to the IHS, it's my view you should do it to the HS-base -- if it's nickel-plated. On the latter, you may still void your "warranty," but nobody would hesitate to buy a good heatpipe cooler second-hand that had already been lapped.

 

[dma0991]

So would eliminating half the thickness of the IHS through lapping yield a 5C improvement?

Lapping to half its thickness is a lot of work. By looking at IDC's diagram, there is about >1mm in thickness for the IHS surface. 1mm may not sound much but you'll be using minimal force (unlike sanding a piece of wood) during lapping. Also, unless you're using a low grit sandpaper (I would only use 1000 grit or higher), it'll take forever to get it to half its original thickness precisely.

Rather than lapping and getting 5C of improvement, delidding and reapplying aftermarket TIM yields better improvement with less time and effort.

 

[PliotronX]

The issue with Haswell is not the IHS itself, which I abhor in its own right, but the TIM underneath it. The crud they use these days is responsible for double digit, sometimes outrageously higher temps, I'm talking 20 degrees celsius. IOW with a simple air cooler you can outperform your friends watercooler with the stock TIM still in place.

 

[krnmastersgt]

The issue with Haswell is not the IHS itself, which I abhor in its own right, but the TIM underneath it. The crud they use these days is responsible for double digit, sometimes outrageously higher temps, I'm talking 20 degrees celsius. IOW with a simple air cooler you can outperform your friends watercooler with the stock TIM still in place.

I recall a while back someone on either OCN or even here in the CPUs and Overclocking section talking about their experiences de-lidding and I don't recall the TIM being all that bad. While certainly not the best iirc the experiment had shown that it performed as well or better than many common TIMs when you got rid of the "large" gap between the IHS and the die itself. Just that gap reduction alone gave way to a large amount of temperature decrease although getting a "better" paste doesn't hurt.

Only paste that I've seen people using on their de-lid experiments consistently seems to be the liquid metal stuff from coollaboratory.

 

[PliotronX]

I recall a while back someone on either OCN or even here in the CPUs and Overclocking section talking about their experiences de-lidding and I don't recall the TIM being all that bad. While certainly not the best iirc the experiment had shown that it performed as well or better than many common TIMs when you got rid of the "large" gap between the IHS and the die itself. Just that gap reduction alone gave way to a large amount of temperature decrease although getting a "better" paste doesn't hurt.

Only paste that I've seen people using on their de-lid experiments consistently seems to be the liquid metal stuff from coollaboratory.

Makes sense, thanks for that. In which case you can shave all the IHS you want, it's not going to reduce that gap or make a lick of difference. With the right mounting, it's all moot and completely optimal sans noobshield.

 

[BonzaiDuck]

Makes sense, thanks for that. In which case you can shave all the IHS you want, it's not going to reduce that gap or make a lick of difference. With the right mounting, it's all moot and completely optimal sans noobshield.

In the case of the IB and Haswell for Z77 and Z87 sockets, you are absolutely right. There's not reason to mess with the IHS at all if you choose to go "bare die" with your installation. If you don't, you'd just lap off the nickel -- especially if you de-lid and re-place.

But the "E" processors still use the fabrication process using solder. You'd have to hit the IHS with a blowtorch to simply remove it, with much greater risk and uncertainty than de-lidding any of the other CPUs.

So grinding down more of the IHS might be an option if you don't mind risking carpal tunnel syndrome :biggrin: