For years, choosing flash storage for a server came down to picking an interface and a capacity. Today there is a third variable that shapes price, endurance and performance: the NAND type. Two SSDs with the same NVMe interface and the same capacity can behave radically differently depending on whether their cells store three bits (TLC) or four (QLC).
The difference fits in a single number: how many bits each memory cell holds. More bits per cell means more density and a lower cost per GB, but also fewer write cycles, weaker sustained-write performance and far more complex error correction. Understanding that trade-off is the key to not overpaying for endurance you never use — and to not deploying drives that wear out before their time.
In this article we explain how each NAND type works, what DWPD and TBW mean, why sustained writes on QLC "fall off a cliff" and which workloads each technology shines in. If you also want context on connection interfaces, pair this read with our SATA vs SAS vs NVMe guide.
Bits per Cell: From SLC to QLC (with PLC on the Horizon)
A NAND cell stores information as an electrical charge. SLC (Single-Level Cell) holds 1 bit per cell: it only needs to distinguish two voltage levels, which makes it blazingly fast and extraordinarily durable (~100,000 program/erase or P/E cycles), but very expensive. MLC (Multi-Level Cell) doubles that to 2 bits and 4 levels; TLC (Triple-Level Cell) moves up to 3 bits and 8 levels; and QLC (Quad-Level Cell) reaches 4 bits and 16 voltage levels within the same physical cell.
Every extra bit multiplies density and lowers the cost per GB, but forces the controller to distinguish twice as many voltage levels in a cell that degrades with every write. The result: fewer P/E cycles, slower programming and much more aggressive error correction (ECC), based on LDPC codes that consume controller cycles and add latency as the NAND ages. On the horizon looms PLC (Penta-Level Cell, 5 bits and 32 levels), already demonstrated by the major manufacturers but with no real presence in the enterprise market yet.
In practice, today's server market is split between two technologies: TLC as the general-purpose standard and QLC as the capacity option. SLC has been relegated to industrial niches and to acting as a cache inside other drives, while MLC has practically vanished from enterprise catalogues. To see how NAND fits within the broader set of disk technologies, revisit our comparison of HDD, SSD and NVMe storage.
Endurance: DWPD and TBW, and How to Read Them
Manufacturers express an SSD's endurance with two equivalent metrics. DWPD (Drive Writes Per Day) states how many times per day you can write the drive's full capacity throughout the entire warranty period. TBW (Terabytes Written) expresses the same thing as a total volume of written data. The relationship is direct: TBW = DWPD × capacity × 365 × warranty years. A 7.68 TB SSD rated at 1 DWPD over 5 years supports roughly 14,000 TBW.
Two important nuances when interpreting them. First, DWPD and TBW are not a hard limit: they are the threshold the warranty covers; the drive does not die the moment you exceed it, but the manufacturer stops answering for it. Second, capacity works in favour of large drives: a 15.36 TB QLC drive rated at just 0.3 DWPD still accepts more than 4.6 TB of writes per day — enough for many backup repositories and capacity tiers. Before ruling QLC out on endurance grounds, work out how much you actually write each day.
Over-provisioning (reserved space the controller uses for wear leveling and garbage collection) also matters, as does the write pattern: large sequential writes wear the NAND far less than small random ones, because they generate much less internal write amplification.
SLC Cache and the Sustained-Write "Cliff"
Programming a QLC cell with 16 voltage levels is a slow, delicate process. To hide it, almost every QLC SSD (and many TLC drives) reserves a portion of its NAND to operate in pseudo-SLC mode: 1 bit per cell, extremely fast writes. Write bursts land first in that SLC cache, and the controller migrates them to the QLC cells in the background.
The problem shows up with sustained writes: once the cache is exhausted, the drive starts programming QLC cells directly and speed drops abruptly — from several GB/s down to a few hundred MB/s, in some models below an HDD. This is the so-called write cliff. That is why, on an enterprise SSD, you should look at the sustained write figure after cache, not the marketing peak of the first burst.
The other side of the coin is reading: reading a cell means sensing its voltage, a process that barely gets more expensive with more bits per cell. In practice, TLC and QLC read performance is very similar, both sequential and random. This asymmetry is exactly why QLC shines in read-intensive workloads: if you write little and read a lot, QLC's weakness almost never comes into play.
Comparison Table: SLC vs MLC vs TLC vs QLC
The following table summarises the trade-off between density, endurance and cost for each NAND flash type (P/E cycles are indicative orders of magnitude; they vary with 3D NAND generation and manufacturer):
| NAND type | Bits per cell | Approx. P/E cycles | Relative cost per GB | Typical use |
|---|---|---|---|---|
| SLC | 1 (2 levels) | ~100,000 | Very high | Industrial, internal cache inside other drives |
| MLC | 2 (4 levels) | ~10,000 | High | Legacy; nearly gone from enterprise catalogues |
| TLC | 3 (8 levels) | ~3,000–5,000 | Medium | Current standard: mixed use, databases, virtualization |
| QLC | 4 (16 levels) | ~1,000–1,500 | Low | Capacity tier: object storage, CDN, archive |
When to Choose QLC
QLC is the right choice when the goal is to maximise flash capacity per euro and the workload is mostly reads. Since it reads almost as fast as TLC, it replaces HDDs at an advantage in scenarios where flash used to be unaffordable:
- check_circle Capacity tier: the cold or warm layer of a tiered architecture, with NVMe TLC as the hot tier and QLC absorbing the volume.
- check_circle Object storage: objects are written once and read many times — the ideal pattern for QLC. It is the foundation of high-density S3 storage platforms.
- check_circle CDN and static content: video on demand, images, software downloads; massive reads with occasional writes.
- check_circle Active archive: historical data that must remain queryable at flash latency (analytics over old data, document repositories, regulatory compliance).
- check_circle Backup repositories with frequent reads: when restores, verifications and recovery tests are routine, QLC's fast reads shorten the RTO without paying TLC prices.
When to Choose TLC
TLC remains the general-purpose standard in servers: more P/E cycles, stable sustained writes and a much broader enterprise catalogue. It is the safe choice whenever writes carry weight in the workload:
- check_circle Mixed use: any workload where reads and writes combine unpredictably, from application servers to message queues.
- check_circle Databases: OLTP with constant transactions, journals and redo logs writing continuously; the scenario where QLC's write cliff would be fatal.
- check_circle Virtualization and VDI: dozens of virtual machines generate a constant random-write pattern that demands endurance and predictable write latency.
- check_circle Write-intensive workloads: log and telemetry ingestion, data staging, write caching; here even 3 DWPD TLC or higher is the sensible pick.
How to Read It in Enterprise SSD Specs
Manufacturers segment their enterprise catalogues by write profile, and DWPD is the most reliable clue. Read-intensive (RI) drives are rated between 0.3 and 1 DWPD and can be TLC or QLC; mixed-use (MU) drives sit around 3 DWPD and are TLC with extra over-provisioning; and write-intensive (WI) drives, increasingly rare, reach 10 DWPD. Catalogues such as Samsung Semiconductor's SSD line-up show this segmentation by DWPD and NAND type very clearly.
Practical rule based on DWPD:
Divide the GB you write per day by the usable capacity of the drive. If the result is below 0.3 DWPD, QLC is a firm candidate and you will save on cost per TB. Between 0.3 and 1 DWPD, go for read-intensive TLC. Above 1 DWPD, pick 3 DWPD mixed-use TLC. Always add headroom for peaks and future growth.
Beyond DWPD, also check the sustained write speed after cache, steady-state write latency and power-loss protection (PLP), a must on any server drive. One piece of market context: AI-industry demand is pushing NAND flash prices upwards, hitting the high-end ranges hardest. In this environment, QLC has become the most realistic way to stretch the budget of capacity tiers without giving up flash.
EasyDataHost: Storage Sized to the Workload
At EasyDataHost we do not sell disks off a catalogue: we size storage around each customer's real read/write profile. Our storage servers combine technologies according to data temperature:
- arrow_right Enterprise NVMe TLC for production: databases, virtualization and mixed-use workloads on drives with verified endurance and sustained-write figures.
- arrow_right Capacity tiers for backup and archive: high-density repositories where cost per TB rules, with QLC flash or HDD depending on access pattern and target RTO.
- arrow_right S3 storage on Ceph: scalable object storage with an NVMe hot tier — the natural fit for the read-intensive patterns where QLC performs best.
- arrow_right Our own datacenter in Spain, with ISO 27001 certification, ENS compliance and 24/7 technical support that knows your configuration.
If you are weighing QLC against TLC for your next project, contact our team: we will analyse your daily write volume and propose the configuration with the best price/performance, with no obligation.
Frequently Asked Questions
Is it safe to use QLC SSDs in a production server?
Yes, as long as the workload is mostly reads. QLC reads at speeds very similar to TLC, so it works very well in capacity tiers, object storage, CDN, active archive and backup repositories with frequent restores. For databases, virtualization or write-heavy workloads, TLC remains the right choice.
What do DWPD and TBW mean on an enterprise SSD?
DWPD states how many times per day you can write the drive's full capacity during the warranty; TBW is the total volume of written data that warranty covers. They relate as: TBW = DWPD × capacity × 365 × warranty years. A 7.68 TB SSD at 1 DWPD over 5 years equals roughly 14,000 TBW.
Why do sustained writes drop so sharply on QLC SSDs?
Because write bursts land first in a very fast SLC cache. When a sustained write exhausts it, the drive starts programming QLC cells with 16 voltage levels directly and speed drops abruptly: the write cliff. Reads are barely affected, which is exactly why QLC shines in read-intensive workloads.
Conclusion
QLC and TLC are not competing for the same job: they are two different tools for two different problems. The decision comes down to four ideas:
- arrow_right More bits per cell = more capacity per euro and less endurance: QLC makes flash cheaper at the cost of P/E cycles and sustained writes.
- arrow_right Reading is the shared strength: TLC and QLC read at similar speeds, so in read-intensive workloads QLC delivers the best price per TB of flash.
- arrow_right TLC for production with writes: databases, virtualization and mixed use demand the P/E cycles and stable writes of TLC (1–3 DWPD).
- arrow_right DWPD is your compass: work out your real daily writes, compare them with the drive's capacity, and let the number — not the marketing — decide the NAND type.