Two SSDs both advertised at 3,000MB/s diverge at exactly one moment: when the SLC cache runs dry. While the cache holds, TLC and QLC write at almost identical speeds. Once it drains, TLC settles to roughly 300-1,500MB/s in published measurements, while QLC falls as low as 50-150MB/s. That stretch of the curve never appears on a product page.
The complication is that the drain point is not fixed. Modern drives borrow empty cells to build the SLC cache dynamically, so the fuller the drive, the smaller the cache and the sooner the cliff. Feeding the published speed ranges into a simple model, the same 100GB copy swings from just over a minute to nearly twelve.
Eight voltage steps or sixteen
Everything starts with how many bits go into one NAND cell. As Synology puts it, TLC holds three bits per cell and QLC holds four. Each added bit doubles the number of voltage states the controller must distinguish: three bits means 2³, or eight levels; four bits means 2⁴, or sixteen. Slicing the same voltage range into sixteen bands halves the margin between them, so reads need finer measurement and writes need more careful programming. That is where both speed and endurance go.
The same document states the general rule that denser cells last less — MLC outlives TLC, TLC outlives QLC. Published figures place QLC's program/erase cycles below 1,000.
The SLC cache temporarily undoes this structure. As TEAMGROUP explains, part of the TLC or QLC array is operated in single-bit SLC mode to absorb incoming writes quickly, then folded back into native mode later. With only two voltage states to hit, it is inevitably fast. The price is capacity: run a QLC region in SLC mode and 4GB of physical space converts into 1GB of cache. Building a 100GB cache therefore consumes 400GB of physical area.
| Item | TLC | QLC |
|---|---|---|
| Bits per cell | 3 bits | 4 bits |
| Voltage states to distinguish | 8 levels | 16 levels |
| Capacity cost in SLC mode | 3GB physical → 1GB cache | 4GB physical → 1GB cache |
| Write speed after cache exhaustion | 300-1,500MB/s | 50-150MB/s |
| P/E cycle class | Higher than QLC | Under 1,000 |
How full the drive is decides where the cache ends
A dynamic SLC cache borrows empty cells. An empty drive yields a large cache; a full one has nothing to borrow, so the cache shrinks. On a 1TB drive that might mean 60GB of cache when empty, 15GB at 80% full, and 6GB at 90% full. This shrinkage is why benchmarks look good on a fresh drive and disappointing on one that has been in service for a year.
Applying the published speed ranges, here is a 100GB copy modeled at three fill levels. The model assumes 3,000MB/s inside the cache and post-cache rates of 900MB/s for TLC and 120MB/s for QLC — the midpoints of the published ranges — and treats 1GB as 1,000MB for arithmetic convenience.
| Drive utilization | Assumed cache (GB) | TLC, 100GB copy | QLC, 100GB copy | Gap (x) |
|---|---|---|---|---|
| Empty | 60 | 1 min 4 sec | 5 min 53 sec | 5.5 |
| 80% used | 15 | 1 min 39 sec | 11 min 53 sec | 7.2 |
| 90% used | 6 | 1 min 46 sec | 13 min 5 sec | 7.4 |
The arithmetic is simple. For the QLC drive at 90% full, the 6GB cache clears in 6,000÷3,000 = 2 seconds, and the remaining 94GB consumes 94,000÷120 = 783 seconds. That totals 785 seconds, an average of 127MB/s. The TLC drive under the same conditions finishes in 2 + 94,000÷900 = 106 seconds. Both drives carry the same 3,000MB/s headline figure.
An SSD's peak speed is the cache's speed. What you actually feel is the speed after the cache ends.
Note how gentle TLC's cliff is. Even at 90% full it takes 1 minute 46 seconds, only 66% longer than when empty. QLC goes from 5 minutes 53 seconds to 13 minutes 5 seconds — a 2.2x stretch. TLC's native speed is tolerable and QLC's is not, and that difference amplifies as the drive fills.
The same modeling approach applied to mobile storage standards appears in a comparison of UFS 5.0 and 4.1 bandwidth.
When QLC is still the sensible pick
On published specs, QLC's weaknesses cluster in sustained writes and endurance. Read performance, by contrast, is not far apart, and the capacity gained per won is real. So the write pattern decides the choice.
Short bursts finish inside the cache. Saving documents, browsing, launching games — anything that writes a few gigabytes and stops — feels essentially identical on either type. Move a 50GB video file or install a large game, though, and the transfer runs straight past the cache. Dumping camera footage, serving as a backup target, or hosting virtual machine images are the workloads where QLC is at its worst.
- Capacity per won matters most — read-heavy game and media storage
- Writes arrive in bursts of a few gigabytes at a time
- You can keep utilization below roughly 70%
- Sustained writes are frequent — video editing, full-drive backups
- You habitually fill the drive close to 90%
- One drive serves as both boot and working volume
What to verify before buying
Product pages often omit the NAND type entirely. Working through the indirect clues on the spec sheet in order is faster than guessing.
- Sustained write figure — if only a peak number is listed, look for a whole-drive fill graph in a review
- TBW across capacities in the same series — unusually low TBW per gigabyte points to QLC
- DRAM cache presence — a separate matter from NAND type; drives without it (HMB designs) lose additional ground on random access
- Capacity choice — within a model, larger capacities get bigger caches and more parallel channels; compare sustained write figures for 1TB and 2TB side by side
- Headroom planning — given the dynamic cache, size the purchase around a 70-80% utilization ceiling
- Warranty terms — coverage ends at whichever comes first, years or TBW, so start by converting daily write volume into expected lifespan