How SLC Cache Exhaustion Impacts Storage Benchmarks

This article explores how Single-Level Cell (SLC) cache exhaustion fundamentally alters the benchmarking results of modern solid-state drives (SSDs). By examining what SLC caching is, why standard benchmarks often fail to expose its limits, and what happens to write speeds when the cache fills up, you will understand how to accurately interpret storage performance metrics and avoid misleading hardware assessments.

Understanding the SLC Cache

Modern consumer SSDs primarily use Triple-Level Cell (TLC) or Quad-Level Cell (QLC) NAND flash memory to store data efficiently and affordably. Because writing data directly to TLC and QLC memory is inherently slow, manufacturers program a portion of the drive to operate in Single-Level Cell (SLC) mode, which only stores one bit of data per cell instead of three or four. This SLC cache acts as a high-speed buffer, absorbing incoming data quickly and moving it to the slower, permanent storage areas later when the drive is idle.

The Flaw in Short-Duration Benchmarks

Most popular consumer benchmarking tools, such as CrystalDiskMark or AS SSD, use relatively small test files (typically 1GB to 16GB) by default. Because these small data payloads fit entirely within the drive's SLC cache, the benchmark only measures the speed of the high-speed buffer.

This results in benchmark scores that match the manufacturer's advertised maximum speeds. While these numbers look impressive, they only reflect burst performance and create a false impression of the drive's true sustained capabilities.

The Performance Cliff of Cache Exhaustion

SLC cache exhaustion occurs when a continuous data write operation exceeds the capacity of the SLC buffer. When a benchmark pushes past this limit, the test results change drastically.

Once the cache is full, the SSD is forced to do two things at once: write new incoming data directly to the slower native TLC or QLC memory, and simultaneously fold (compress and move) the data already in the SLC cache to free up space. This heavy dual workload causes write speeds to plummet instantly. In benchmark graphs, this is visible as a sharp drop-off, commonly referred to as a "performance cliff."

The Severity of the Speed Drop

The degree to which cache exhaustion alters benchmark results depends on the drive's underlying hardware and current capacity:

  • TLC vs. QLC NAND: On a high-end TLC drive, a sequential write speed of 7,000 MB/s might drop to 1,500 MB/s after cache exhaustion—still a very usable speed. However, on a budget QLC drive, the speed can drop from 5,000 MB/s down to 100 MB/s or less, making it slower than a traditional mechanical hard drive.
  • Drive Fill State: Most modern SSDs use a dynamic SLC cache, meaning the cache size shrinks as the drive fills up with permanent files. Benchmarking an empty drive yields a massive SLC cache and prolonged high speeds. Benchmarking a drive that is 80% full will result in almost immediate cache exhaustion, yielding drastically lower benchmark scores because there is very little free space left to act as an SLC buffer.
  • Latency Spikes: Aside from raw speed drops, cache exhaustion drastically increases drive latency. Benchmarks measuring access times and Input/Output Operations Per Second (IOPS) will show severe degradation, as the drive controller is overwhelmed with managing data relocation.

How to Benchmark for Cache Exhaustion

To get a complete picture of a storage drive's performance, benchmarking must go beyond the SLC cache. Hardware reviewers achieve this by running sustained sequential write tests that span the entire capacity of the drive, logging the exact gigabyte threshold where the cache runs out.

For everyday users using standard benchmarking software, increasing the test file size to 64GB or higher and increasing the number of test passes will usually force the drive into a state of cache exhaustion. This reveals the drive's true native write speed, providing an accurate representation of how the hardware will perform during heavy workloads like large file transfers, massive game installations, or 4K video rendering.