Linux

I'd wager a toe from my left foot that if you look in the Event Viewer on windows you will see similar looking errors (though not as descriptive, no doubt, it might say something like "corrected read error" or something obtuse instead), this is a hardware issue that linux tends to be more aggressive in handling. These errors are on the physical layer and data link layer, so it is likely a communication problem between the drive and the motherboard, but interestingly, they are corrected on retry, so the data the system is calling from the drive is fine even if it sometimes fails to get there in time. This screams electrical connection to me, either thermal expansion is making the contacts wonky (and they might not be seated perfectly), there is a flaw in the traces somewhere, or there is some power management issue affecting your PCIe bus. Can you try running it with one more kernel parameter? Under pcie_aspm=off add nvme_core.default_ps_max_latency_us=0 and watch dmesg while running something heavy.