If you’ve ever watched a large game installation or a Steam library migration start fast and then crawl to a fraction of the initial speed, you’ve seen SSD throttling. Consumer NVMe drives — and even SATA SSDs — shed write speed when their SLC cache fills, when the controller gets hot, or when background tasks compete for I/O. The good news: while you can’t change the silicon, you can often recover a meaningful slice of lost throughput. Realistically, expect to eliminate most of the post-cache “cliff” on drives that were never thermally limited, and to regain 10–30% of sustained write speed on M.2 drives that were cooking in a cramped motherboard slot. On drives already at their physical limit, tuning gains will be modest — this guide tells you how to find out which camp you’re in.
Quick Wins
These changes take minutes and address the most common throttling triggers.
| Change | Effort | Expected gain |
|---|---|---|
| Add or improve the M.2 heatsink | Low | Substantial if drive was hitting 70 °C+; little if already cool |
| Improve case airflow over the motherboard | Low | Moderate on airflow-starved cases; small otherwise |
| Leave 10–20% of the drive unpartitioned | Low | Moderate — extends SLC cache behavior and reduces slowdowns on near-full drives |
| Disable Windows Search indexing on the game drive | Very low | Small but free; removes background write bursts during transfers |
| Update SSD firmware and chipset drivers | Low | Occasionally substantial — some firmware releases specifically improve sustained writes |
| Pause OneDrive/Dropbox sync during large transfers | Very low | Small to moderate if the transfer folder overlaps synced paths |
Step-by-Step Tuning
1. Fix drive thermals first (biggest impact for most NVMe drives)
What causes it: NVMe controllers hit a thermal throttle point — commonly around 70–80 °C composite temperature — and the firmware deliberately cuts throughput to protect itself. Many motherboards place the primary M.2 slot directly under the GPU, which acts as a radiator.
How to check: Open CrystalDiskInfo and watch the “Temperature” reading while running a sustained write (a large file copy is fine). If the composite temperature climbs past roughly 70 °C and your transfer speed drops in the same window, you’ve found your cause. HWiNFO64 (Sensors section, under your drive’s entry) also shows drive temperature and throttling flags on supported controllers.
What to do: If the drive shipped with no heatsink, install one — a basic finned aluminum M.2 heatsink with a thermal pad is inexpensive and usually sufficient. If your motherboard has a factory M.2 heatsink (check for a small metal plate with a retained screw near the slot), make sure its protective film was removed and the thermal pad is seated. If the drive sits directly under the GPU, consider moving it to a secondary slot or repositioning GPU spacing if your board allows. Also add at least one rear exhaust fan and confirm front intake fans actually feed the motherboard area.
How to undo: Remove the heatsink (reinstall any factory shield) or move the drive back to the original slot. There is no software state to revert.
2. Verify the drive is running at full PCIe link width and speed
What causes it: A drive that negotiates x2 instead of x4 lanes, or PCIe Gen 3 instead of Gen 4, will show much lower sustained transfer ceilings — often mistaken for throttling because speed also degrades as the drive fills.
How to check: In CrystalDiskInfo, look at “Transfer Mode” — it shows two values, for example “PCIe 4.0 x4 | PCIe 4.0 x4” (current | maximum). If current is lower than maximum, the link is degraded. Also check your motherboard manual: some boards drop the primary M.2 slot to x2 or disable it entirely when certain SATA ports or a second slot is populated.
What to do: Reseat the drive, update your BIOS to the latest version, and install the current AMD chipset driver or Intel Chipset Device Software from your motherboard vendor’s support page. If the slot shares lanes with SATA ports, move cables off the affected ports per the manual’s sharing table.
How to undo: Revert BIOS settings or repopulate the original slot if you intentionally changed slots for another component.
3. Keep 10–20% of the drive free
What causes it: When a drive approaches full, the controller has fewer free blocks for wear leveling and the dynamic SLC cache shrinks dramatically, so sustained writes fall off a cliff long before the drive is technically full.
How to check: Open File Explorer and look at the drive’s free space. If it’s under ~15% free and writes slow down badly on that drive specifically, this applies to you.
What to do: Move games or archives off until at least 15% is free. For a permanent safeguard, shrink the volume: right-click Start → Disk Management → right-click the partition → Shrink Volume, and shrink by 10–15% of total capacity. Windows will then never fill that region, and the controller always has headroom. On a 2 TB drive, that’s roughly 200–300 GB.
How to undo: Return to Disk Management, right-click the unallocated space, and choose Extend Volume to reclaim it.
4. Confirm the drive isn’t secretly in a degraded state
What causes it: Firmware issues, failing NAND, or an exhausted drive can force the controller into slow, cautious write modes that look exactly like throttling.
How to check: In CrystalDiskInfo check the health status (should read “Good,” ideally 90%+ health) and look at the “Available Spare” attribute in HWiNFO or the vendor tool — if it’s below threshold, the drive is reallocating NAND. Also run the vendor utility: Samsung Magician → “Performance Benchmark” and “Drive Details”; WD Dashboard → “Drive Details”; Crucial Storage Executive → “Drive Details.”
What to do: If a firmware update is pending in the vendor tool, apply it — several controller vendors have shipped firmware that specifically improves sustained (post-cache) write behavior. Back up first, don’t power-cut during the update, and let the tool complete. If the drive reports low spare capacity or “Caution” health, replace it rather than tune it.
How to undo: Firmware downgrades are generally not offered; document the previous version before updating if you want a rollback option through the vendor.
5. Stop background I/O from competing with the transfer
What causes it: Windows Search indexing, antivirus real-time scans, cloud sync clients, and Windows Update delivery optimization all generate reads and writes. When they overlap your transfer, the queue depth contention looks like the drive slowing down.
How to check: Open Task Manager → Performance tab → your SSD → watch “Active time” and throughput while the system is idle. If the drive is busy at rest, something is hammering it. Resource Monitor (Win+R → resmon → Disk tab) shows exactly which processes are doing it.
What to do: For a game/data drive, disable indexing: File Explorer → right-click the drive → Properties → uncheck “Allow files on this drive to have contents indexed…” → Apply, choose “Include subfolders,” and let it finish. Temporarily pause OneDrive: click the cloud icon in the system tray → Settings gear → “Pause syncing” → 2 hours. In Windows Security → Virus & threat protection → Manage settings → Exclusions, add the destination folder for large transfers if you accept the trade-off below.
How to undo: Re-check the indexing box, resume OneDrive from the same tray menu, and remove the exclusion from the same Windows Security screen. (Do keep real-time protection on for your OS drive and downloads.)
6. Make sure write caching is enabled
What causes it: If Windows write caching is disabled, every small write goes straight to NAND, which tanks transfer speed and can mimic cache exhaustion.
How to check: Win+X → Device Manager → Disk drives → right-click your SSD → Properties → Policies tab. “Enable write caching on the device” should be checked (it is by default).
What to do: Enable the checkbox. For most modern systems, leave “Turn off Windows write-cache buffer flushing” unchecked unless your drive has a capacitor-backed power cache and the vendor recommends it — an unexpected power loss with aggressive flushing disabled can corrupt in-flight data.
How to undo: Uncheck the box and reboot.
7. Keep the drive’s firmware-level over-provisioning intact (SATA and some NVMe)
What causes it: Some drives (notably SATA Samsung models) let users reserve an “Over Provisioning” region in Magician. Deleting partitions without restoring OP, or manually wiping with the wrong tool, removes that headroom and hurts sustained writes.
p
How to check: Samsung Magician → “Over Provisioning” — it shows the current OP size and whether it’s active. Other vendors expose similar options or none at all.
What to do: If OP is not set and the vendor recommends it (Samsung suggests different percentages based on usage), set it via the slider and apply. Do this when the drive is mostly empty, since setting OP erases data on some models.
How to undo: Drag the OP slider back and apply — the reserved space returns to the usable partition.
Measure Before and After
Tune without measuring and you’re guessing. Do this before touching anything:
- Baseline thermals: Start a large copy (or a vendor-tool benchmark that writes steadily), and log drive composite temperature with HWiNFO64’s sensor logging. Note peak temperature and the transfer speed at the start versus minute five and minute ten.
- Baseline sustained write: Use CrystalDiskMark with a realistic profile — set the data pattern to a large value (e.g., 16 GiB) so the test outlasts the SLC cache, and run the “SEQ1M Q8T1” write row plus a mixed “SEQ1M Q1T1” pass. Record the numbers.
- Apply one change at a time and repeat both measurements under the same conditions: same drive fill level (within a few percent), same background apps closed, similar ambient temperature.
- Judge on the sustained portion. Peak speeds are marketing; the number that matters is what the drive holds after 60–120 seconds of continuous writing. If your post-cache plateau rises, your fix worked.
Keep a simple note of fill level alongside each result — a drive tested at 80% full will always look slower than the same drive at 20% full, and that confound ruins more comparisons than any other single factor.
What Not to Do
- Don’t disable write caching to “reduce heat.” It lowers sustained performance and invites data corruption on power loss.
- Don’t install the drive without any thermal solution because a review showed yours “doesn’t throttle.” Controller thermals depend heavily on airflow and slot placement; your case is not the test bench.
- Don’t run aggressive third-party “SSD optimizer” utilities that promise to clean or tune the drive. Windows’ built-in TRIM (run via Optimize Drives: Start → “Defragment and Optimize Drives”) plus vendor firmware is all a modern SSD needs.
- Don’t manually defragment an SSD. Windows 10/11 correctly runs TRIM instead on SSDs; forcing a traditional defrag adds pointless write wear.
- Don’t chase the drive’s peak spec. If your sustained plateau reaches its rated sustained figure, the drive is healthy — remaining “gaps” versus headline numbers are SLC caching by design, not a fault.
- Don’t stack every tweak at once. If indexing off, firmware, heatsink, and OP land in the same week and something regresses, you won’t know which change caused it.
FAQ
Is SSD throttling harmful to the drive?
No. Thermal and cache throttling are protective firmware behaviors, not failures. Sustained operation at high temperatures can shorten lifespan over years, which is why a heatsink is worth it — but the throttling itself is the drive protecting itself, and the main cost is your time waiting on transfers.
Why does my transfer start at full speed and then drop?
Almost certainly the SLC cache. Most consumer NVMe drives use a portion of fast pseudo-SLC flash as a write buffer. Once it fills on a long transfer, speed drops to the drive’s native TLC (or QLC) write rate — which can be a fraction of the headline figure on QLC drives. Keeping the drive from getting too full and checking thermals keeps that plateau as high as the hardware allows.
Do heatsinks interfere with the motherboard’s M.2 shield?
Use one or the other, not both. If your board has a factory shield with a thermal pad, it’s usually well-matched to the slot’s clearance — reinstall it correctly and skip the aftermarket sink. Stacking a chunky aftermarket heatsink under a GPU or over a factory shield can create clearance and pressure problems rather than solving them.
How hot is too hot for an NVMe SSD during transfers?
Most controllers begin throttling somewhere in the 70–80 °C composite range, and sustained operation above roughly 85 °C is worth fixing regardless of throttling. Check your drive’s datasheet for its exact limits, and treat any reading that climbs steadily toward 80 °C during a long copy as your cue to improve cooling.
