SSD Thermal Throttling

Overview

SSD thermal throttling is the mechanism by which solid-state drives automatically reduce performance to prevent damage when internal temperatures exceed safe operating thresholds. SSDs are rated for 0-70°C but can run slightly warmer; beyond that, two throttling stages activate. At approximately 77°C (Threshold 1), performance drops to roughly 58% of rated speed. At approximately 79°C (Threshold 2), an emergency brake cuts performance below 1% — effectively halting useful work.

The economic significance is direct and severe: throttling reduces storage bandwidth, which in turn degrades GPU utilization, reducing tokens per second and impacting customer economics for both rented and owned GPU fleets. MLCommons MLPerf Storage 2.0 data across 150 industry submissions demonstrates a strong linear correlation between storage read bandwidth and active GPU utilization, making SSD temperature management a first-order concern for AI cluster efficiency.

Air-cooled SSDs in high-density GPU servers are increasingly vulnerable to throttling as PCIe interface generations advance: Gen 5 devices (~25W) already approach ASHRAE A2 inlet air limits; Gen 7 devices are projected at 60-70W where air cooling is definitively infeasible.