NVMe thermal throttling: temperature, speed and heatsinks

An NVMe SSD starts a long transfer at full speed and then drops sharply. After a pause, the speed returns. Heat may be the cause—but SLC cache exhaustion, garbage collection, a full drive or a slow source can create a similar graph.

The right diagnosis logs temperature and throughput on the same timeline and checks the drive's thermal counters. A hot screenshot by itself proves little.

What thermal throttling does

An NVMe controller consumes power while moving data, running error correction and managing NAND. When its thermal policy reaches a threshold, it can reduce activity or enter a lower-power state so temperature does not continue rising.

NVM Express describes host-controlled thermal management with two configured thresholds [1]:

• below TMT1, normal thermal operation;
• from TMT1 to below TMT2, light throttling;
• at or above TMT2, heavy throttling;
• after cooling below the applicable threshold, the controller can return toward normal operation.

This is a protection mechanism, not automatic evidence of a defective SSD. It becomes an operational problem when the device cannot sustain the performance required in its actual chassis and ambient conditions.

There is no universal temperature number

The NVMe specification defines temperature reporting and thermal-management fields, but vendors choose sensor placement and product limits. Sensor readings can represent implementation-specific physical points, while composite temperature may combine information differently [2].

NVM Express also notes that the same SSD can be configured differently in a desktop and a laptop because their thermal environments differ [1]. Micron describes host-controlled thermal management as letting the host optimize SSD temperature for its workload [4].

Therefore, do not write “all NVMe SSDs throttle at 70°C” into an acceptance specification. Use:

1. the exact model's datasheet and firmware documentation;
2. its warning and critical thresholds;
3. the required performance at the defined ambient temperature;
4. observed behavior in the final enclosure.

How to prove the slowdown is thermal

Run a controlled sustained workload and collect at least:

• throughput at short intervals;
• composite temperature and available sensor readings;
• TMT1/TMT2 transition counts and total throttling time, if exposed;
• ambient temperature and fan state;
• drive fill level, firmware and test-file size.

The NVMe SMART/Health log can report temperature sensors, thermal-management transition counts and the total time spent under TMT1 and TMT2 [2]. If the speed drop repeatedly aligns with a threshold or the counters increase, the thermal explanation is strong.

If speed drops while temperatures and thermal counters remain stable, investigate other causes:

Our SLC cache and sustained-write guide explains the first pattern. Review the complete NVMe SMART health log as well; temperature is only one health signal.

When a heatsink helps

A heatsink creates a path from hot components to a larger surface. It can delay or prevent throttling when that surface can release heat into the surrounding air. Kingston notes that a heatsink is not always necessary; the answer depends on the workload and installation [3].

Check four details:

1. Contact: the thermal pad must touch the intended controller and other components without damaging the module.
2. Clearance: the assembly must fit under the GPU, motherboard shield or device cover.
3. Airflow: a larger hot mass in stagnant air eventually saturates.
4. Coverage: single- and double-sided modules may need different mechanical designs.

Do not stack an aftermarket heatsink under a motherboard cover without checking the board maker's instructions. Excess pressure, poor pad thickness or an insulating label arrangement can make the result worse.

A buyer's thermal acceptance test

For an OEM or system qualification:

1. Test the exact capacity, controller, NAND, firmware and approved BOM.
2. Install it in the final enclosure, not an open bench only.
3. Define ambient temperature, airflow and fan profile.
4. Precondition the drive to the required fill state.
5. Run the real sustained workload longer than its temporary cache.
6. Log throughput, temperatures and thermal counters.
7. Repeat with the proposed heatsink or airflow change.
8. Set an application-level performance floor and maximum throttling duration.

This turns “runs hot” into a measurable system requirement. It also prevents a cooler benchmark platform from hiding a problem that appears only in the customer device.

Bottom line

Thermal throttling protects an NVMe SSD by reducing work when configured temperature limits are reached. Confirm it with synchronized temperature, throughput and thermal-counter data; do not diagnose it from touch or one screenshot. Apply the exact vendor limits, then improve contact and airflow—or add a compatible heatsink—only when the installed workload shows a real need.