How to check SSD health without fooling yourself

Wear percentage vs narrative

Vendor-normalized “life left” can flatten scary cliffs into gentle slopes. Track raw counters where exposed and correlate with sudden realloc-equivalent events on SSDs (grown defects). If wear stalls visually but latency climbs during sustained writes, suspect thermal or mapping pressure.

TBW math that maps to your workload

Divide vendor endurance rating by your measured daily writes — video caches and databases skew heavily. Re-evaluate quarterly if roles shift (e.g., scratch disk becomes archive).

Thermal throttling hides latency spikes

Thin laptops throttle NVMe under sustained loads. Sentinel’s thermal curves expose peaks during renders — useful even when SMART wear looks pristine.

NVMe notes

Bus resets and surprise disconnects sometimes precede firmware faults. Pair SMART trends with OS event logs when chasing intermittent disappearances from PCIe lanes.