
EPYC 7513 melted a part of its heat-spreader
One of our servers had been throwing weird memory allocation, memory access and execution errors on a Linux box we were running.
All of it got fixed however, after we disabled one of its physical cores, which was weird, right? Why did disabling one core fix such an odd mix of errors?
So we decided to replace the whole CPU despite only having lost 1 of its 32 cores.
What we discovered, after removing its cooler however, utterly shocked us.
The CPU had literally melted the aluminum layer of its heat-spreader in two spots, suggesting there must have been some kind of a short circuit in one of the cores, or some other kind of a hardware issue.
In all of my 8+ years of working in IT, I have never seen a thing like this.
An AMD EPYC 7513, with two spots of copper showing through the aluminum layer of the heat-spreader