Your Cheap QLC SSD Is Two Drives in a Trench Coat

Last updated: October 4, 2026 · 5-minute read

Every cheap QLC SSD ships with a secret second drive inside it. The box promises 5 GB per second; the drive delivers that speed from a fast region that shrinks as you fill the disk — and when it is gone, the same hardware crawls at hard-drive speeds. This is not a defect or a scam. It is the entire reason QLC drives are cheap, and understanding the mechanism takes ten minutes that will save you from the most common storage mistake of the AI-era hardware market.

What QLC actually means

Flash memory stores bits by trapping charge in cells, and the naming counts the bits per cell: SLC holds one, TLC holds three, QLC holds four. Four bits per cell means sixteen distinct voltage states the controller must resolve on every write — that is why QLC offers the lowest price per terabyte on the market, and also why writing to QLC cells directly is slow. More states per cell means finer voltage discrimination means more time and more error-correction per operation. Density and speed pull in opposite directions by physics, not by marketing.

The second drive inside

The workaround is genuinely clever. The controller takes a slice of the same QLC NAND and runs it in single-bit mode — treating 4-bit cells as if they held one bit each. Single-bit mode writes almost as fast as real SLC flash, so everything you copy lands in this fast region first, and the drive folds it down into dense four-bit storage later, when the drive is idle. The result is the two-drives-in-a-trench-coat structure: a fast drive that handles your bursts, hiding a slow drive that handles the long run.

The catch is arithmetic. A cell in single-bit mode holds a quarter of what it holds in four-bit mode, so the fast region always borrows capacity from the same pool your files occupy. Every gigabyte you keep is a gigabyte that can no longer serve as cache — which means the more the drive fills, the smaller the fast region gets. The cache is dynamic, carved from free space, and its shrinkage is proportional to how full you run the drive.

The numbers that make it real

TechPowerUp's test of the HPFX700 — a $100, 2 TB, PCIe Gen 4 QLC drive — puts numbers on the collapse. The drive held close to 5 GB/s at the start of the write. Sustained speed lasted until about 54 GB had been written, then dropped to 3.5 GB/s. After 460 GB of the 2 TB had been filled, performance collapsed to 130 MB/s — a figure the reviewers compared directly to a mechanical hard drive. Same drive, same test, one order of magnitude apart, purely because the cache architecture ran out of room. The Intel 660p, the first mainstream QLC PCIe SSD, described exactly this dynamic design in its product brief: cells reconfigure to balance capacity against performance.

Read that sequence again if you have ever wondered why "the drive was fast when I tested it" and "the drive is unusable when my Steam library is full" describe the same product. Both observations are true, at different fill levels.

How to buy and how to behave

The buying rule is one sentence: match the drive to your write pattern. QLC with its burst-friendly cache is fine for boot drives, app drives, and general desktop use where writes come in bursts and idle time lets the fold-down happen. QLC is the wrong choice for sustained writers — video work, constant large backups, game-library drives you fill to 90 percent — where TLC's lower density but steady write speed is worth the premium, especially in a market where every gigabyte is repricing upward.

The behavioral rule matters as much: never fill a QLC drive near capacity. Leaving 20 to 25 percent free is not superstition; it is reserving the raw material the cache is carved from. A 2 TB QLC drive that holds 1.5 TB is a different machine than the same drive at 1.95 TB.

How much free space should a QLC drive keep?

Twenty to twenty-five percent is the practical floor, and the mechanism explains the number: the SLC cache is carved from unoccupied cells, so free space is literally performance budget. A drive at 75 percent full retains a meaningful fast region; the same drive at 95 percent is running almost entirely in four-bit mode, where the HPFX700's 130 MB/s floor lives. If a QLC drive must run near full, over-provisioning by leaving a unpartitioned slice achieves the same effect less conveniently.

Does this affect SATA SSDs too?

Yes — the NAND logic is identical across SATA and NVMe, and some of the most cache-dependent drives ever sold were SATA QLC models. The interface only changes the ceiling, not the collapse: a SATA drive tops out around 550 MB/s, so its post-cache floor sits closer to its peak, which ironically makes the cliff less dramatic. The NVMe drives with 5 GB/s peaks have the farthest to fall, which is why the collapse is a modern-gen story.

TL;DR

• QLC = 4 bits/cell: cheap and dense, slow to write directly.
• Controllers hide a single-bit fast region carved from the same NAND — it shrinks as the drive fills.
• Measured collapse: ~5 GB/s to 130 MB/s on a 2 TB drive partway filled (TechPowerUp).
• Match the drive to your write pattern, keep 20-25% free, or buy TLC for sustained writes.

How to test your own drive

Watch the throughput number across the write. A drive that starts high and steps down — and recovers after idle — is showing you its cache architecture in real time. The test costs you nothing and tells you which of your two hidden drives you were actually using.

The one-line summary for the comments section: QLC is not bad, it is conditional — and the condition is written in your fill level, not on the box.

---

Not affiliated with TechPowerUp, Intel or any SSD vendor. Sources: XDA Developers (Adam Conway, Oct 3, 2026), TechPowerUp's HPFX700 review, Intel's 660p product brief as cited.