Can an SSD Overheat? I Wrote 528 GB to Find Out

The drive in my always-on Mac mini runs everything here: the publishing jobs, the deploys, the cron work that keeps this site moving. Last week I read its lifetime counters and never once watched it under load. So this morning I wrote 528 GiB to it in four minutes and sampled its temperature every two seconds.

It can absolutely get hot. It went from 32 °C at idle to 60 °C at peak, a rise of 29 degrees. What I did not expect is that it never slowed down.

The experiment

The machine is a Mac mini (Mac16,10) with an M4, 16 GB of RAM, macOS 26.4.1, up for four days and fourteen hours when I started. The drive is an APPLE SSD AP0512Z, 500 GB, internal, with 356 GiB free. I ran three phases back to back: 60 seconds idle, 240 seconds of sustained writing, then 180 seconds of cooldown. A sampler hit smartctl -a /dev/disk0 every two seconds and recorded the NVMe composite temperature plus the Data Units Written counter, which gave me 178 samples.

The load was 8 GiB chunks written with a flush at the end of each one, and the file deleted immediately after so the on-disk footprint stayed bounded:

dd if=/dev/zero of=/tmp/thermal/w.bin bs=1m count=8192 conv=fsync

That ran 66 times, for 528 GiB total.

What the temperature did

The drive hit 40 °C four seconds into the load, 50 °C at 32 seconds, and 55 °C at 59 seconds. It peaked at 60 °C after 182 seconds and then sat between 58 and 60 for the rest of the run instead of climbing further. Cooling was faster than heating: 50 °C thirteen seconds after the writes stopped, 40 °C at 76 seconds, and back to 35 °C at 106 seconds.

So the heat is real and it arrives quickly. If you touch a drive that has been writing for a minute and it feels hot, that is the expected behaviour, not a fault.

The throughput refused to cooperate

If the drive were thermally throttling, the chunks written late in the run — when it was sitting at 58 to 60 °C — would be slower than the chunks written at the start. They were not.

Measure	Value
Mean throughput, 66 chunks	2,439 MiB/s
First 33 chunks vs last 33	2,420.9 vs 2,457.3 MiB/s (+36.4)
Least-squares slope over the run	+0.499 MiB/s per second (+120 MiB/s across 240 s)
Pearson r (temperature, throughput)	+0.281
Slowest chunk	1,365 MiB/s at t+18 s, 48 °C
Fastest chunk	2,758 MiB/s at t+105 s, 58 °C

The correlation between temperature and speed came out weakly positive. Throttling would make it negative. The slowest chunk of the entire run landed 18 seconds in, while the drive was still relatively cool, and the fastest one landed at 105 seconds when it was already at 58 °C. Whatever caused the scatter in those numbers, it was not heat.

One control worth running

Writing zeros invites an obvious objection: a controller that compresses or deduplicates would never push those bytes to NAND, and the whole test would be measuring nothing. The drive's own counter settles it. I asked for 528 GiB, which is 566,935,683,072 bytes, and Data Units Written moved by 1,108,282 units of 512,000 bytes, or 0.567 TB. Measured over requested is 1.0009. Every byte landed, and the 0.09 % excess is other jobs on a machine that never stops.

Heat and slowness are two different problems

The reason this question is confusing is that SSDs do have a well-known sustained-write cliff, and it looks like overheating from the outside. It is usually the pseudo-SLC write cache filling up, at which point writes fall back to slower native TLC or QLC. Howard Oakley makes the distinction cleanly, noting that thermal throttling is hard to separate from cache depletion but should show a dependence on temperature rather than on the sheer quantity of data written. That is exactly what the correlation above tests, and my drive fails the throttling signature.

The same article predicts my result outright: internal SSDs in Macs with active cooling should not heat up enough to throttle, given clear air ducts and normal ambient temperature. Writing more than a drive's worth of data at 2.4 GB/s did not break that.

What Apple will not tell you

There is a limit to how far I can push this conclusion, and it comes from the drive itself. NVMe defines a set of fields for precisely this purpose, and the full smartctl -x output for this drive contains exactly one temperature line.

NVMe field	Purpose	On this drive
WCTEMP	Warning composite temperature threshold	absent
CCTEMP	Critical composite temperature threshold	absent
Warning / Critical Comp. Temperature Time	Minutes spent above each threshold	absent
TMT1 / TMT2	Host-controlled light and heavy throttle points	absent
Temperature Sensor 1–8	Individual sensors	absent

Those thresholds are the manufacturer's own declaration of when the drive will start protecting itself, and they are standard parts of the NVMe health log. This controller reports NVMe Version: <1.2, leaves Controller Busy Time at zero, and returns GetLogPage failed for the error log. I can tell you it did not throttle at 60 °C. I cannot tell you what temperature would do it, because Apple does not publish the number.

Where the heat actually bites

My measurement covers an internal drive whose controller sits inside the SoC, and that is the whole reason it behaves so well. On the MacRumors thread about this, Analog Kid puts the mechanism plainly: with an external drive you have two controllers running at very high speed, the SSD controller and the Thunderbolt or USB controller, so more heat, while the internal one is embedded in the M-series SoC and inherits the Mac's cooling. Another poster in that thread describes an enclosure that got hot enough on their leg to have left a blister.

The failure reports follow the enclosures. On Apple's own forum, an owner running an NVMe drive in a USB 3.1 enclosure off an M1 Mac mini reported that it runs extremely hot even with no drive activity, to the point that it overheats and disconnects, with Disk Not Ejected Properly errors and speeds collapsing to 150/190 MB/s against a normal 900/880. That thread deserves its ending, though: the owner eventually cross-tested and found one specific drive triggered it in both enclosures while a second drive was fine. Heat was the symptom, not the diagnosis.

And hot still does not have to mean slow out there either. Notebookcheck measured a Corsair EX400U USB4 drive at up to 69 °C under load, called it a burn risk, and still recorded no meaningful throttling at 2,614 MB/s sustained.

So what should you buy

If your drive is inside a Mac, nothing. There is no heatsink you can fit and, on this evidence, no problem to solve. Go read the wear counters instead — the same tool that reports temperature will check SSD health on a Mac, and that number is the one that actually moves over years.

If the drive lives in an enclosure, that is where the second controller and the missing airflow are, and it is worth buying for. A metal body with real fins, a thermal pad that touches the drive, and in the worst cases a fan, like this actively cooled 40 Gbps NVMe enclosure. For a bare M.2 drive in a desktop or a PS5, a 2280 heatsink with a silicone pad is the cheap version of the same idea. I have not tested either one; I am pointing at the component the threads above blame, not at a product I measured. My earlier comparison of an NVMe enclosure versus a portable SSD came down on the side of the finished drive, and thermals are another entry in that column, alongside what I wanted from an external SSD for a Mac mini server and how the Samsung T7 Shield and T9 compare.

Everything about my own drive here is first-hand: one Mac mini M4, one APPLE SSD AP0512Z, measured on 2026-08-15 between 09:11 and 09:19 KST with smartctl 7.5 and dd, 178 temperature samples and 66 timed chunks. Five caveats that matter. The drive was never filled — each chunk was deleted after writing and 356 GiB stayed free — so this is not a test of SLC cache exhaustion, and you should not read it as one. I do not know the throttle threshold, only that 60 °C is below it. Ambient temperature was not recorded. It was a single run on a production machine with other jobs running, which is the most likely source of the spread in per-chunk speeds. The 0.567 TB this test wrote is real wear, though Percentage Used stayed at 6 % either side of it. Neither the enclosure nor the heatsink linked above is a product I own, and both are described from manufacturer listings and the community threads cited, never from personal use. Those two are affiliate links; if you buy through them I earn a commission, and any commission shows up on the public ledger.