Thermal paste! A little dab and all....

[ziggy29]

.... well, those of you who are old enough remember the rest of the old slogan.

Over the weekend I pulled out my Quad G5 to play around a little and to make a couple of tweaks. As I was moving it, I dropped it with one hand and one end of it klunked onto the floor (fortunately not too hard) about 8 inches. I figured all was probably OK, but when I went to start it, it failed with the light on for CPU A overheating. I stopped, checked a few things and tried again. Same thing. Eventually I pulled out most of the RAM (from 16 GB down to 4 GB, just two matched sticks) and got it to boot up. Then I went to reduced performance mode and that allowed my trusty Quad to limp along, with CPU A temps in the low 80s with spikes into the 90s, and CPU B in the 40s and low 50s. OK, turn that sucker off before CPU A completely fries!

So I figured that *probably*, that bump caused something to go awry. I decided to go with the easiest fix, the thermal paste, figuring that maybe the bump caused the last adhering paste to loosen up and stop working. (It is a single pump unit and one CPU was not overheating so I decided the LCS was not the likely source of the problem.) So today, I took the plunge and redid the thermal paste. This was the first time I have done it on the G5; I have removed and swapped out the complete CPU/heatsink assembly but never the thermal paste.

The teardown to remove the entire assembly is well documented. The part about removing the CPU module for repasting, not so much. I eventually figured out that I needed to remove 10 hex screws on each CPU and it came right out. Armed with some rubbing alcohol to remove the old paste -- which looked almost nonexistent -- and reapplying new paste (I went with Noctua NT-H1) I reversed all the steps to rebuild.

It was flaky getting it to restart. But I found that I really had to tighten the CPU/heatsink assembly really well or else the motherboard would complain it could not see one or both CPUs. But finally it started up. Now it still needs time to fully cure, but I ran it for a few minutes and here are the temps now (see attached). In particular, A and B are almost identical. So I feel pretty confident that the problem is corrected and the ancient (probably original) thermal paste was the issue. It should be even better, I hope, in the days and weeks to come as the paste cures.

 

[Totalbearface]

Not sure the bump would have caused any thermal compound to loosen up from the CPU die itself, the pressure that the die meets the heatsink is pretty high, but it does seem to turn into cement over time (at least, the factory stuff does I've noticed). Never a bad idea to change it out though for some fresh stuff! Recently did it on my G5 quad too just for peace of mind and to check the pump / LCS for any leaks, which thankfully there weren't.

 

[ziggy29]

Well, yeah -- my thought was that cement can crumble on impact, and that maybe enough of it crumbled to make the contact between CPU and heatsink entirely insufficient. Once it's already pretty far gone it doesn't take too much more flaking away to cross the Rubicon, so to speak.

 

[B S Magnet]

I’m stealing this thread to make it all about thermal paste.

So far, since buying some Noctua NT-H2, I’ve completed the clean-up and new application of thermal paste from the OEM paste/pads on two of my systems — a PowerBook G4 17" DLSD and an early 2011 MacBook Pro i5 13".

* * *

The first, the PowerBook, was a genuine mess before cleaning it out. It was stunning how filthy it was. The paste on (more precisely, around) the CPU was rock-hard, while the thermal pad for the GPU was basically a lumpy, sticky white mess. I’ll let the pictures tell the rest of the story:





Before [below, left] and after [below, right] idle temps are only marginally cooler, but when watching a video or crunching something in Photoshop, the temps are noticeably lower — on the order of 3–7°C for the CPU. The screen caps are from about a week apart, and ambient room temperatures were cooler a week ago (a window is typically open).



The GPU on it is an oddball, and I’ve never been sure whether it’s an accurate reading. Whenever the system is idle, both before and after new thermal paste, the GPU reports ~97–100°C. When I’m pushing the GPU, the displayed temp actually tends to drop as much as 13–15°C. I’m not sure why.

* * *

Then there’s the early MacBook Pro which, from the day I bought it from Apple as a factory refurbished unit, has run the same build of 10.6.8 and will probably always run 10.6.8 until it dies (though it now has a backup partition set up with Sierra to run Disk Util should anything on the main partition run amok).

The old thermal paste was about what one might expect from Intel-era Macs. Compared to the PowerBook, it was a lot easier to clean up and re-paste.

These screen caps occurred within two hours of one another (the room more or less being the same temperature). As with the before screen cap [below, left], I let the system settle in for about a half-hour before recording the sensor temps again [below, right].



That said, idle temperature differences, compared with the PowerBook, are dramatic. The average drop for all CPU sensors is anywhere between 15–20°C. It’s incredible. As a by-product of the CPU being measurably cooler, the other interior sensors are also a bit lower — ~3–10°C cooler, depending on the sensor.

tl;dr: I must be growing dull in my old age, because I’m finding this fascinating. It’s going to be a while before I’ll attempt this on my Power Mac G5 (mostly because getting to the tower and disassembly is going to be a nightmare), but I’m excited to carry this to my other systems as I’m able. I chose the above two because the performance differences were liable to be noticeable (and there are a fair number of internal sensors to measure overall conditions).

If you’re adventurous and comfortable with tinkering, definitely try some new thermal paste.