How SLC Cache Recovery Shapes SSD Write Performance

An SSD can appear exceptionally fast when it is fresh, lightly loaded, and receiving a short burst of data. A large game install, 4K video export, or disk clone tells a different story. Once the drive’s fast pseudo-SLC buffer fills, sustained write speed can fall sharply, sometimes from several gigabytes per second to a few hundred megabytes per second.

The behaviour is easy to misunderstand because the drive has not necessarily developed a fault. Its controller is changing how NAND flash is used, moving data between temporary high-speed cells and their long-term TLC or QLC arrangement. That background work competes with incoming writes and determines how quickly the drive can recover its advertised performance.

For PC builders, the important question is not simply how fast an SSD writes at the beginning of a benchmark. It is how long the cache lasts, how quickly it replenishes, and what happens when the drive is already warm or nearly full. Those details matter when choosing storage for a gaming system, workstation, NAS, or content-production machine.

Hardware Hounds readers can follow current component releases through the site’s hardware news, but specifications alone rarely explain real-world storage behaviour. A proper test must examine the entire transfer, the idle time afterwards, and the next write operation that follows.

The role of pseudo-SLC cache

Most modern consumer SSDs use NAND flash in TLC or QLC modes. TLC stores three bits per cell, while QLC stores four. Higher bit density reduces cost per terabyte, but writing and verifying several voltage states takes more time than using a cell as if it stored a single bit.

Pseudo-SLC cache temporarily treats a portion of that NAND as single-level cells. The controller can write data quickly with fewer voltage states to manage. When the host sends a short burst, this produces impressive benchmark numbers and responsive file transfers. The cache may be fixed, dynamically allocated, or a mixture of both.

Dynamic cache capacity changes according to free space. A mostly empty drive may reserve a large portion of its NAND for fast writes, while a nearly full drive has far less room to do so. This is why the same SSD can show different results after a fresh Windows installation, months of game downloads, or a drive filled to 90 per cent.

Why large transfers cause a slowdown

During a long sequential write, the SLC region eventually fills. The controller must then write incoming data directly into TLC or QLC cells, or first fold existing SLC data into its native format to free cache space. That folding process is commonly called garbage collection or cache recovery, although the exact algorithms vary between manufacturers.

The transition can be abrupt. A drive rated for 7,000 MB/s in a short burst might settle near 1,000 MB/s, and some QLC models can fall below that level under sustained workloads. The result depends on NAND type, controller design, firmware, temperature, available spare area, and whether the transfer is sequential or made up of smaller random writes.

A 1 TB SSD may also deliver a different experience from its 2 TB sibling. Larger capacities generally contain more NAND dies working in parallel, which can increase throughput and provide a larger effective cache. Capacity is therefore linked to performance, rather than being just a measure of how many games or video files fit on the drive.

What happens after the transfer ends

When the operating system stops sending data, the SSD can use idle time to move pseudo-SLC data into its normal TLC or QLC layout. This process releases cache space for the next burst. Recovery may begin immediately, but it is not always aggressive because the controller must balance background work against power consumption, temperature, and future requests.

A short pause may restore only part of the cache. A few minutes of desktop activity can leave the drive in a very different state from a completely idle hour. If the computer enters sleep mode, loses power, or starts another demanding task, the controller may postpone some housekeeping. Drives also differ in whether they continue background folding while the system is idle.

This is particularly relevant to a workstation that alternates between large projects. Exporting a video, copying camera footage, and then immediately opening another project can create a repeated write-and-recovery cycle. The second transfer may be slower even though the first benchmark looked excellent. The drive is still processing the first workload internally.

For an enthusiast in Perth or Adelaide, high room temperatures can make this pattern more severe. A hot NVMe drive has less thermal headroom, so throttling can arrive during the same period that cache recovery is competing for controller time. A motherboard heatsink helps, but airflow around the M.2 slot still matters.

How to test sustained write behaviour

A useful SSD review should include a transfer large enough to exceed the advertised cache. A 10 GB benchmark is convenient, but it may measure only the fast front portion of a drive. Writing several hundred gigabytes, or filling a substantial share of the available space, reveals the transition to native TLC or QLC performance.

Test conditions need to be recorded carefully. The drive should be tested when empty and when partially full, with its temperature monitored throughout. Reviewers should also report the speed before the transfer, the lowest sustained rate, the time spent recovering, and the result of a second transfer after a defined idle period.

Synthetic tools such as fio, Iometer, and large-file copy tests can expose different parts of the design. A single huge sequential file stresses bandwidth, while many smaller files add metadata and random-write overhead. Real applications may produce a mixture of both, so one chart should not be treated as a complete description of the drive.

Queue depth matters as well. A high queue-depth benchmark can keep the controller busy enough to hide the user-visible pauses that appear during ordinary desktop work. Conversely, a low queue depth may understate the drive’s capability during professional workloads. Good testing explains the workload rather than presenting one peak figure as universal.

Controller design and drive capacity matter

A DRAM-equipped SSD can maintain a mapping table in dedicated memory, reducing the need to fetch address information from NAND. DRAM-less models often rely on Host Memory Buffer, which borrows a small amount of system RAM. HMB can work well for everyday use, but sustained writes, low free space, and heavy background activity may expose differences between designs.

NAND quality also shapes recovery. TLC generally offers more consistent sustained performance than QLC, although a well-designed QLC drive can be perfectly suitable for a game library or media archive. Firmware decides how much NAND is reserved as over-provisioning, when folding begins, and how aggressively the controller prioritises cache space.

Drive fullness is one of the easiest variables to overlook. Leaving 10 to 20 per cent free gives the controller more working room for garbage collection and wear levelling. A full drive can have a smaller dynamic cache, slower recovery, and higher write amplification. The effect is often more noticeable after large transfers than during casual browsing or gaming.

Australian pricing makes this comparison practical. A 2 TB TLC model from retailers such as Scorptec, Centre Com, or Umart may cost more than a high-capacity QLC alternative, but the extra expense can be justified for frequent editing or large backups. GST-inclusive pricing and local warranty support also make advertised overseas bargains less attractive than they first appear.

Choosing an SSD for real workloads

For gaming, the cache usually has enough time to handle ordinary updates and installations, particularly on a spacious drive with plenty of free capacity. Game loading is dominated by reads, so a dramatic post-cache write drop may have little effect during play. Large Steam or Xbox app installs can still expose the difference between a premium TLC model and an entry-level QLC drive.

Content creators, developers, and anyone moving virtual machines should focus on sustained performance and recovery behaviour. A drive that writes at 2,000 MB/s for an entire project may be more useful than one that reaches 7,000 MB/s for the first 30 seconds. The same principle applies to local backups, where a long transfer can run well beyond the cache window.

A sensible test routine is to copy a large folder, leave the system idle for five minutes, and repeat the copy. Repeat it again after the drive is nearly full and while monitoring temperature. This shows whether the SSD recovers quickly, whether its dynamic cache shrinks, and whether thermal throttling changes the result.

When comparing products, look beyond a single peak sequential-write figure. Check the NAND type, controller, DRAM or HMB design, rated endurance, capacity-specific reviews, and warranty terms. Broader market perspective can help explain why NAND pricing and capacity deals fluctuate, but hands-on sustained testing remains the best guide to how an SSD will behave in a particular PC.