NVMe SSD shows high latency and occasional I/O errors during sustained writes on Ubuntu 22.04

Environment

Custom desktop with ASUS X670 motherboard, Ryzen 9 7900X, 32 GB DDR5, Samsung 990 Pro 1TB NVMe in M.2_1 slot, Ubuntu 22.04 kernel 6.2, ext4, laptop cooling disabled for test

Symptoms

Background file-copy of a 120 GB test file stalls intermittently; applications block on disk I/O (IOwait spikes). Average write throughput drops from ~3.5 GB/s to <100 MB/s after ~90 seconds; some copy operations fail with 'Input/output error'.

Observations

smartctl reports temperature rising from 45°C to 95°C during the test; dmesg shows repeated nvme timeout/recover errors ('nvme nvme0: I/O 12345 QID 0 timeout, aborting'); iostat shows device utilization 99% followed by large service times; no power spikes on UPS logs.

Already attempted

Updated firmware to latest vendor release (rebooted); tried lower sequential write test with dd (same pattern after longer run); moved drive to a different M.2 slot (M.2_2) with unchanged behavior; disabled aggressive CPU boost governor (no change); mounted with noatime and aligned partitions (no change).

Question

What diagnostics and mitigations should I run to determine whether this is thermal throttling, controller firmware bug, or a motherboard M.2 power delivery issue, and how can I safely collect logs and avoid destructive steps?

Post timeline

COMMUNITY RESPONSES

Verified community troubleshooting

0 replies

No answers yet..

Showing 0 of 0 community responses

AI DIAGNOSTICS

Incident analysis

Updated 1w ago• Trouble ID

Most likely causes

NVMe internal thermal throttling / emergency shutdown when controller reaches critical temperature 90%

SSD firmware bug causing controller resets under sustained sequential writes 60%

Motherboard M.2 slot power delivery or BIOS power-limit/cooling configuration causing overheating 30%

Summary

Symptoms strongly point to a thermally-triggered performance throttle or internal NVMe controller fault: rapid temperature rise to >90°C, nvme timeouts in kernel logs, and sustained throughput collapse all fit a thermal-protection or firmware-level recovery. Less likely but possible are poor M.2 slot cooling, BIOS M.2 power limit, or PCIe lane/link negotiation issues. Collecting controlled telemetry (temperature, SMART error counters, kernel nvme traces) while isolating variables (different slot, passive cooling, reduced PCIe link speed) will narrow the root cause.

Suggested Fixes

Diagnostic step 1

Reproduce with incremental telemetry: run a controlled fio sequential-write test while capturing smartctl -a (every 10s), iostat (1s), and dmesg -w > nvme.log; keep system idle except for test.

Diagnostic step 2

Attach an external IR thermometer or thermocouple to the M.2 heatsink and record surface temperature; compare to SMART temperature. Use passive heatsink or increase airflow (desktop case fan) and rerun test to check if behavior improves (reversible).

Diagnostic step 3

Boot a recent live USB (Ubuntu 24.04 or vendor's diagnostic ISO) and repeat a short stress test to rule out kernel regression; collect nvme-cli logs (nvme list, nvme smart-log /dev/nvme0).