Those Darn Pushpins

I earlier documented the construction of my current computer, Juno. Last month it suddenly, overnight, began turning itself off just a few seconds after being turned on. At first I wasn't sure if it was the power supply, or something else, and I was even considering swapping power supplies. However, after being turned off overnight (and cooling off completely), juno would run just long enough for me to log in, and fortunately one of the programs that starts automatically when I log in is a system monitoring application called gkrellm. Thus it was revealed to me that the MCP (media and communications processor) on my motherboard, a combination of an interface to the memory and an integrated graphics processor (IGP), an nForce 730i, was rapidly getting so hot (over 90°C) that it was automatically shutting the system down to prevent damage to itself.

I was now confident of what I would find when I opened up the computer case. The heatsink on top of the MCP had probably come loose, robbing the MCP of any means to be cooled. The heatsink used by this particular motherboard, a Zotac 9300 mini ITX, is fastened atop the chipset by spring-loaded pushpins. Just two of them, on opposite corners. When building Juno, I had accidentally popped one of the pushpins out. Though I had successfully reinserted it, my thought was that this was the likely weak spot. It was.

Just pushing that pin back through the hole didn't work; it would immediately pop out again. It was going to be necessary for me to access the underside of the motherboard, which meant largely disassembling the computer.

Here is the underside of the motherboard. The locations of the two heatsink pushpin holes are circled in red. One is holding a pushpin, and one isn't. The bolted arms at the bottom of the photo are part of the mechanism that holds the heatsink for the CPU -- much more reliable than pushpins.

I inspected the failed pushpin. There was some beveling of the lip; perhaps I could spread the tip, or perhaps I should try something else; if I employed this pushpin again, the next failure might be in two months rather than two years.

The smallest machine screw at the hardware store (#4) would fit through the pushpin hole without forcing, so that was my choice of repair.

Here is the underside. Two thin, flexible washers protect the motherboard.

I carefully reassembled Juno, and pressed the power button. It booted normally, but then I looked at gkrellm and saw that the MCP was running about 15°C hotter than before.

Although technically acceptable, this was distressing. Two possible causes occurred to me. First, that the separation of the heatsink and MCP had scrambled or churned the thermal grease, which might have become thicker or dried out over the last two years. Second, that an imbalance of force exerted by the one remaining pushpin versus the nut-and-bolt might have lifted one corner, a small amount to be sure, but enough to interfere with heat transfer. Especially since only two of the four corners of this heatsink were secured or securable. If it had to have pushpins, why not four?

Both of these possibilities would mean disassembling the computer yet again. Worse, because the CPU heatsink hovers over a third of the MCP heatsink, I might need to remove the CPU heatsink to get sufficient access to the MCP to replace the other pushpin. To clean surfaces and replace the old thermal grease, I would almost certainly have to take off the CPU heatsink. Here's a top-down view.

I didn't feel that the reward would be worth the effort of a complete teardown. The temperature was still within specs, after all. But I enjoy tinkering when there's time, and so decided to try a couple of easier remedies.

First, I replaced the downward-blowing CPU fan, 120mm in diameter, with a 140mm fan. Thus more air would be blown onto the MCP, theoretically reducing its temperature.

The 140mm fan didn't quite fit into the case; I attached it to the CPU heatsink with only one screw so that it would rotate a few mm, allowing the case cover to close. Sad to say, this didn't lower the MCP temperature at all, and might have even raised it a degree or two. The airflow pattern of the larger fan was less suitable than that of the smaller one! I put the 120mm fan back on.

Then I tried attaching an 80mm fan to blow horizontally at the MCP heatsink. It turned out to fasten only in such as way that it could blow air towards the vicinity of the MCP, but not directly at it.

It didn't change the temperature a bit. I reversed the fan, so that it would push hot air out of the case, but that didn't change the temperature either. I then slapped it onto the back of the case, in a location which had some ventilation but still blocked perhaps a quarter of the fan, again to blow hot air out, and that may have improved the MCP temperature by 1°-2°C.

At this point I'd invested enough time. After all, the nVidia settings program considers the MCP to be only one measly bar into the yellow.

Juno is chugging along well, and I'm slowly growing accustomed to the higher MCP temperatures. They haven't interfered with writing this blog post!