My home server uses dual Intel Xeon Gold 6138 processors. They’re getting a bit long in the tooth, so I decided to purchase some used Xeons of a newer generation, with more cores and a higher clock speed. This had all the ingredients to be a huge upgrade, and since I had upgraded the CPUs once before (a few years ago), I was confident it would be a fairly straightforward process.
I chose Xeon Gold 6248s for the upgrade because they have a 25% faster clock speed and only draw marginally more power. To accommodate the CPUs, I needed to purchase upgraded CPU coolers, as the ones I am currently using don’t support 150W TDP CPUs. The eBay reseller I purchased the CPUs from accidentally shipped me 13 CPUs instead of 2, so they ordered UPS to return the shipment to them before it made it to me. They then overnighted me my 2 CPUs once they received the return shipment. So my CPUs made a few trips around the US before they made it to me.
The time comes to finally tear the server apart and begin the upgrade. I slide the server forward on its rails, carefully watching the cables to make sure they don’t get caught on anything. Not carefully enough. I heard a ripping sound and immediately stopped sliding the server. At first, I couldn’t figure out what happened. I saw one of the CAT 6 cables disconnected, but they don’t make tearing/ripping sounds when they are disconnected. Then I saw a small black piece of plastic sitting on top of the UPS (the device racked immediately below the server). It took me a minute to figure out where it came from and then I realized it was the part of the RJ45 port that holds the cable clip in place, to keep the cable seated in the port. It had literally torn the plastic out of the port on the server, so that cables can no longer lock into place in that port. Great. We’re off to an awesome start.
I removed the top of the server, then proceeded to remove the cooling duct that covers the CPUs. Upon removing the CPU coolers and CPUs, I realized I was a bit overzealous with the thermal paste during my last CPU upgrade, so there was some excess paste on the edges of the CPUs that I had to clean up. I got everything cleaned up, and began assembling the new CPUs and CPU coolers. This is when I noticed that the underside of the CPUs were not completely clean. At first I thought some of the contacts were scratched off, but I realized there was just some grease or dirt smeared on the bottom of the CPUs. I used a clean rag to wipe the CPU contacts and they looked much better. I applied thermal paste and installed the CPUs and coolers. The cooling duct had some removable panels that had to be removed in order to accommodate the larger CPU coolers. I buttoned everything back up and powered on the server to see the results. About 10 seconds into the BIOS POST, I received a message relating to the RAM:
This was not encouraging. Since I didn’t touch any of the RAM DIMMs, I thought perhaps the CPU contacts were still dirty and for some reason they weren’t recognizing the RAM. Hoping that no damage was done to the server, I proceeded to power it down and remove the CPUs to inspect them. The contacts were still not perfectly clean, so I decided to use isopropyl alcohol and a paper cloth to clean the contacts. This worked very well and I was happy with the results. Hopeful that this was the issue, I reinstalled everything and powered it back on.
Same error. Shit.
I started googling the scenario and came to a rather disheartening conclusion. My DIMMs are too slow for this generation of CPU so my server is failing the memory test. Shit. Shit. Shit.
My only option at this point is to tuck tail and put the original CPUs back in. I revert everything back to the way it was, and power it back on. All is well, but I had to change my VMware vSwitch to use a different physical port for the network, since I ruined the port I was using.
The money I spent on the CPUs and CPU coolers was completely wasted. I also wasted about 90 minutes, ruined an RJ45 port on my server, and accomplished nothing. What a great way to spend a Tuesday evening!