Last night’s reading: “mercurial” cores, CPUs that very occasionally return the wrong result for some instructions. 😱 Affects a few CPUs per thousand, across vendors.
We’re used to ephemeral bit flips at scale in memory and (some) storage, due to solar rays, which we handle with ECC and CRCs, but we expect a different failure model from CPUs. Maybe no longer.
I mean, how would you ECC compute? Can’t even begin to imagine. Redundant hardware? Starts to feel as big and unfixable as Spectre/Meltdown. ðŸ˜ðŸ˜ðŸ˜
Funny/sad timing, a cosmic ray evidently hit a Certificate Transparency log the other day. Very unfortunate service to hit, damage is evidently unfixable 😢