This is one writer's account, not a lab study. Treat it as a useful data point rather than proof for every SSD.
What the test actually did
The drive started at about 64% full. Rather than jump straight to 100%, White filled it in stages. At each stage she ran a game, copied files and checked everyday use. She also watched temperatures in HWiNFO. That was meant to stop her blaming a lack of free space for slowdowns that were really thermal throttling.
Her first surprise was that little changed. At 90% the drive wasn't struggling, and at 95% it hadn't turned into "a glorified hard drive," as she put it.
The benchmark numbers
At roughly 90% capacity, CrystalDiskMark reported:
| Test | Reported result |
|---|---|
| Sequential read | 7,429 MB/s |
| Sequential write | 6,639 MB/s |
| Random 4K Q1T1 read | 60.77 MB/s |
| Random 4K Q1T1 write | 189.76 MB/s |
She says these matched earlier runs at lower fill levels. They also land at or just above Lexar's published ratings. The NM790 page lists sequential read up to 7400MB/s and sequential write up to 6500MB/s for most capacities, and the manufacturer notes those are based on internal testing and actual performance may vary.
There is a catch, and White flags it herself. CrystalDiskMark was on its default 1GiB test file. That is a tiny workload for a 2TB drive, so it mostly measures the fast cache. The NM790 uses Host Memory Buffer 3.0 and dynamic SLC caching. Lexar describes the latter as writing to single-level-cell memory to maintain speed.
Where the cracks showed: big copies
Smaller file copies stayed fast even at 95%. The trouble came with a much larger batch of files copied from her secondary drive. The transfer started quickly, then became less consistent as it went on.
White attributes this to the dynamic SLC cache shrinking as free space disappears. Her write-up doesn't give the batch size, the sustained speeds or a same-batch comparison at different fill levels. So the cause is her explanation, not something she measured.
The explanation matches how dynamic caches are generally described. One storage explainer says that many SSDs use a dynamic SLC cache that draws on free space, so the cache shrinks as the drive fills and the speed drop arrives sooner. Another source describes a drive that sustains about 6,500 MB/s early in a long write and then falls well below that once the cache is exhausted. That figure is for a different drive, so it shows the pattern and not what the NM790 does. These are secondary sources, not manufacturer documentation. Still, the pattern fits her observation: the drive is fast until the cache runs out.
Gaming and everyday use
Games showed no obvious problems. White watched loading screens and asset loading with browser tabs open and says everything was largely fine. She gives no timed results, so this is an impression and not a measurement.
Her reasoning is sensible, though. Games mostly read from the SSD, and a fuller drive doesn't automatically mean slower reads. The slowdown she saw was in sustained writes. If your workload is mostly installing and launching games, you may notice nothing. If you're regularly moving huge batches of files, you're more likely to hit the cache limit.
The real problem: running out of room
White's biggest takeaway is that the main problem wasn't speed. At the end of the test she had close to 100GB free. In her words, that's "pretty much nothing" by 2026 standards. It might fit one big AAA game. She also felt tempted to delete things hastily, even at the risk of neglecting backups.
That matters for Windows. Microsoft's guidance says the free space needed for an update varies. The same page states that feature updates typically need 6GB to 11GB or more, and quality updates 2GB to 3GB or more. A ~100GB buffer clears those figures comfortably. But Microsoft also explains that updates temporarily need room to download, to apply, and to support rolling the update back, typically for 10 days. Windows may also temporarily disable files such as hiberfile.sys and pagefile.sys to free space during a feature update. The practical risk comes from a fuller drive plus a big game install plus a pending update all colliding.
Is 20% a hard rule?
White concludes that the 20% advice is useful but "not a magic number." Her own experience was that she saw little difference between 60% and 95%. She still recommends following the manufacturers. The evidence supports a softer reading:
- Vendors don't agree on one number. She says most manufacturers recommend about 20%. Other published guidance ranges from 10% to 20%, and one explainer says the right amount depends on whether the SSD has a DRAM or SLC cache.
- Percentages scale with drive size. On a 2TB drive, 20% is about 400GB. That is a lot of headroom to hold back for a drive you might otherwise use.
- The risk is mostly in sustained writes. Heavy workloads such as video editing, virtual machines and large file transfers are where a shrunken cache hurts most.
- No damage was shown. The test doesn't show that 95% harms the SSD, and it shouldn't be read as a drive-health claim either way.
What to do if your system drive is nearly full
Microsoft's documented steps for freeing space are:
- Open Settings, then System, then Storage.
- Select Cleanup recommendations and review the categories. These include temporary files, large or unused files, files synced to the cloud, and unused apps.
- Select Clean up for each section you want to clear.
- Optionally, turn on Storage Sense to automatically remove temporary files and Recycle Bin items.
- Change where new content is saved, under Advanced storage settings, so new files don't land on a full system drive.
- If an update needs more room, Windows Update may offer Fix issues. That lets you use an external drive with at least 10GB free. Microsoft warns you may still be asked to free internal space, and recommends backing up important files first.
Microsoft also notes that after an upgrade, the previous Windows version can be removed within the first 10 days. Doing so deletes the Windows.old folder, and you can't undo it or roll back.
Bottom line
On this one NM790, near-full didn't mean near-dead. Short benchmarks held up and gaming felt normal. A big write got uneven, and the remaining space felt cramped. A sensible reading is to treat 10 to 20% free as a comfort buffer, not a cliff. Keep more headroom if you move large files often. And remember that on a Windows system drive, an update needing space can cause trouble before speed does.