Overclock.net banner

1 - 1 of 1 Posts

1,362 Posts
Discussion Starter · #1 ·
So I delidded my A10-7700k. It's been done before, though I had to do it myself to take a shot at the best temps I could get for this chip. Well, correction: I didn't HAVE to do it, but I really, really wanted to do it. I've finally got the TIM application between the die and IHS correct (yes, I did a relid), so I can finally report some positive results.

Tracking temperatures on this chip is . . . interesting. I can track either thermal margin or socket temperature. The best data I can provide showing the difference in temperatures before and after delid are at stock, where it was easiest to track thermal margin (I never really did finish tuning a max overclock for the chip prior to delidding, shame on me).

The maximum thermal margin I recorded @ stock before delid was ~14C while running y-cruncher. mprime (Prime95 by it's Linux name) blend only got me up to ~11C thermal margin.

Now that I've slathered on enough CLU between the die and the IHS, my delidded/relidded 7700k shows a ~2C thermal margin running y-cruncher or linpack @ stock. Ridiculous.

Post-delid/relid, I am now able to push the chip up to 4.7 ghz @ 1.508v vcore (as shown by CPU-z) with stability. Not 24h tested yet, but that's just an isolated clockspeed without NB/RAM/iGPU OC to go with it. Since AMD's rated breakdown voltage is 1.49v, I'll probably stay away from that clockspeed for anything but benchmark runs. Regardless, before the delid, even a few seconds of stress testing @ 4.6 ghz or higher brought errors or lockups, so I have some new options clearly.

So, in summary, the delid bought me about 200 mhz clockspeed overhead (I had 4.5ghz isolated stable before, but getting it stable with NB + RAM + iGPU OC wasn't yet finished).

Some of you may be wondering what method I used and what TIM I used in which places. Here's a simple breakdown, without pictures because I was lazy:


Razor method, though I used an X-acto knife instead. In particular, I used a basic X-acto #1 knife with #11 blades. The package came with a total of six blades, which is good because I broke one blade during the cut. I was rather clumsy and nicked the outer edge of the PCB in 2-3 places enough to expose copper traces. The CPU is still quite functional. Regardless, the #11 blades are thin and sharp enough to make short work of the epoxy if you know what you are doing, which I did not at first. I sort of learned as I went. Sadly, it involved a lot of cutting towards my body, which is not exactly safe. Also, once the blade starts to sink into the epoxy, it gives pretty quickly, so controlling the speed of the cut is not easy. What I did was sink the blade in on the corner, and then press on the back of the blade with my thumb while straightening the blade (taking great care to make a mental note of where the epoxy bond ended underneath the IHS so that I didn't cut any resistors). I'd stop the cut about halfway along the edge of the IHS and rotate the CPU 90 degrees to start on a new corner. That, of course, didn't get but half of the epoxy cut, so eventually I had to start cutting away from myself which was much trickier.

Early on I freaked out when I saw metal shavings coming from underneath the IHS. I thought I had started peeling away a lot of the PCB. Turns out the blade was shaving copper off the IHS itself from me tilting the edge upward slightly.

Once complete, I removed the epoxy from the IHS only, and cleaned up the epoxy shavings that were left on the PCB. It turns out that leaving the epoxy behind on the PCB helps to prevent the IHS from sliding around when you are relidding the CPU prior to installing the HSF. After several relids (read below), I have also learned that the force of the HSF mount and the heat from the CPU will basically glue the IHS back on with a very weak epoxy bond thanks to the epoxy remaining on the IHS. It's easy enough to break that bond by twisting the IHS slightly after removing the HSF.

Also of note: I did lap the IHS surface to 400 grit to get rid of the nickel surface coating and work out any imperfections in the copper underneath. As was the case in the last two AMD CPUs that I lapped (x2-3600+, x4-635), the nickel surface was fairly even but the copper underneath was not. I stopped at 400 grit since, in my experience, lapping isn't that big of a deal when CLU is involved. Getting rid of the nickel can help, but going for a mirror shine is just pointless when you're going to mar the surface and even alter the surface texture of both the IHS and HSF anyway.


I used Coollaboratory Liquid Ultra between the die and IHS, and between the IHS and HSF. The HSF is a Noctua nh-d14 which is already well-stained by CLU. I didn't have much left, so I used an absolutely tiny amount on every surface (it is my technique to use a brush with the bristles trimmed very short with scissors, and to paint both surfaces to be joined with a very thin layer). That works for me between the HSF and IHS, but when the die comes into play, it's a whole different ball game. It went like this:

1st application: Temperatures were horrible! I suffered a lot of CPU throttling due to out-of-control thermal margin numbers at anything but stock. I quickly tore down the system and inspected the die, only to notice that the contact pattern showed only a tiny bit of direct contact between the die and the IHS. I found out that most of my remaining CLU was a solid plug that I could not work out of the syringe, so I used what dregs I could scrape off the syringe surface and off a pin I stuck into the syringe (really) to try and fix the problem. In my experience, it is seldom necessary to actually remove previously-applied CLU; you can just paint more on top and it works fine. So, I painted a bit more onto the die and the underside of the IHS, and then a tiny bit onto the IHS and HSF, reinstalled everything, and crossed my fingers.

2nd application: Temperatures returned to what they had been at stock. I was able to reach 4.5 ghz again, but it was not much better than before. Getting 4.5 ghz CPU/2 ghz NB/1028 mhz iGPU/DDR3-2400 still eluded me. I ordered more CLU. When the new CLU got here, I let it sit for four days because I was busy with other . . . stuff. Finally, earlier today, I tore down the system again, and found that the contact pattern was . . . inconclusive. There were spots on the die where CLU had pooled without obviously making any serious contact with the IHS. I applied an excessive amount of CLU to the die, painted a wee bit more onto the underside of the IHS, and then made sure I put the spillage clinging to the syringe onto the IHS and HSF so that it wouldn't wind up crusted onto the cap instead. I reinstalled and . . .

3rd application: Success! Temps were insanely good compared to what they were after the 2nd application. Well, insanely good in my book anyway. Compared to simple lapping jobs I've done in the past, this delid/relid stuff came with serious benefits. I recorded about 4C lower socket temperature and 12C lower thermal margin @ stock, which is simply amazing when you consider how hard it is to drop a chip from 34C to 30C. At stock, the socket temp only runs 5C above ambient. And that's with more voltage (1.2125v) than it realistically needs for 3.4 ghz anyway. I could probably drop the volts and get it to run even cooler.

If you're using CLU between the die and the IHS, you may need to use a fair amount of it to get good results. It is better to go a bit overboard there than to be stingy. I credit Idontcare from the Anandtech forums for (sort of) making this clear to me in his now-venerable 3770k delid thread. He wasn't relidding, but the same principle applied.

Any other Kaveri delids out there? Please tell us all about it.
1 - 1 of 1 Posts