Why Copying One Giant File Is a Terrible SSD Speed Test
Copying one giant file is a terrible way to test your SSD’s speed because it only measures the drive’s fastest possible sequential write, which is the easiest workload for an SSD and the least representative of real-world use. It also ignores the cache behavior, controller limitations, and mixed read/write patterns that dominate everyday performance. To get a meaningful number, you need a benchmark that tests sustained throughput, random access, and queue depths.
What a Single-File Copy Actually Measures
When you copy a single large file (say, a 20 GB video), the operating system issues a long stream of sequential write commands. The SSD answers each one as fast as the NAND flash can accept data, which is exactly what the drive’s marketing spec sheet calls “sequential write speed.” That number is the absolute best-case throughput, achieved only when the drive is writing to empty or recently erased blocks in a linear fashion.
In practice, that scenario is rare. Most everyday workloads involve small files, random reads and writes, and a mix of both. A single-file copy also benefits from the drive’s SLC cache (a fast pseudo-single-level-cell buffer) that absorbs the first several gigabytes of writes at peak speed. Once the cache fills, the drive must flush data to slower TLC or QLC NAND, and the speed often drops by 50% or more.
Why the Cache Makes the Test Misleading
Modern SSDs use a small portion of the NAND as a write cache, typically running in SLC mode. This cache can be as small as a few gigabytes on budget drives or as large as 200 GB on high-end models. When you copy a single file, the first portion (often 5–20 GB) is written at full speed, and the test ends before the cache fills. The result you see is the cache speed, not the sustained speed of the drive.
To expose the real sustained performance, you need to copy a file larger than the cache, or run a benchmark that writes continuously for several minutes. Tools like CrystalDiskMark’s “All 0xFF” or “Random 4K Q32T16” tests, or AS SSD’s “Seq” and “4K” tests, are designed to bypass or exhaust the cache. A single-file copy rarely does that unless the file is enormous.
Real-World Workloads Are Random and Mixed
Your operating system, applications, and games do not read and write one giant file. They access hundreds of small files, update metadata, and interleave reads and writes. These are random I/O operations, measured in IOPS (input/output operations per second), not megabytes per second. A drive that copies a 20 GB movie at 2 GB/s might still feel slow when launching an app because it only manages 30 MB/s on random 4K reads.
Furthermore, a single-file copy is a pure write test. It tells you nothing about read speed, mixed read/write performance, or latency under load. An SSD that excels at sequential writes can be mediocre at random reads, which is what matters for boot times, game loading, and database queries.
What a Proper SSD Benchmark Tests
A reliable SSD benchmark should include at least these four tests:
- Sequential read/write – the same as a file copy, but with the cache controlled or exhausted.
- Random 4K read/write – measures IOPS, which correlates with real-world responsiveness.
- Mixed read/write – simulates a busy system where reads and writes happen simultaneously.
- Sustained write – writes continuously for 10+ minutes to show the post-cache speed.
Tools like CrystalDiskMark, AS SSD, and ATTO Disk Benchmark offer these tests with configurable file sizes and queue depths. For a quick check, set the test file size to at least 16 GB to exceed the cache on most drives. For a thorough test, run the full suite and compare the 4K random numbers, not just the sequential ones.
How to Choose a Benchmark Tool
When picking a benchmark, consider what you want to learn:
- If you want a quick, comparable number, use CrystalDiskMark with the default settings but a 16 GB test file.
- If you want to see sustained performance, use AS SSD’s “Seq” test or run a custom test with a large file size.
- If you want real-world simulation, use PCMark 10’s storage test, which replays actual application traces.
Always run the benchmark on an empty or nearly empty drive, and close background apps. The drive’s free space affects performance because SSDs slow down when they are more than 70% full. Also, ensure the drive is connected to a native NVMe port (not a SATA adapter) and that the motherboard’s BIOS is set to AHCI or NVMe mode, not IDE.
Common Mistakes When Testing SSDs
Even with a good benchmark, people make errors that invalidate the results:
- Testing a nearly full drive – performance drops significantly as free space shrinks.
- Using a file that is smaller than the cache – you measure cache speed, not sustained speed.
- Copying between two drives – the bottleneck is often the source drive or the file system, not the target.
- Running while the OS is doing background tasks – Windows Update, indexing, or antivirus scans skew results.
- Comparing numbers across different benchmark tools – each tool uses different file sizes and queue depths, so results are not directly comparable.
Frequently Asked Questions
Can I use a file copy to compare two SSDs?
Only if you copy a file larger than the largest cache (e.g., 50 GB) from a fast source like RAM disk, and you repeat the test several times to average results. Even then, the copy speed is affected by the source drive, the file system, and the OS cache, so it is not a controlled comparison. A proper benchmark is more reliable.
Why does my SSD slow down after copying 10 GB?
That is the SLC cache filling up. Once the cache is exhausted, the drive writes directly to the slower NAND, and the speed drops to the sustained rate. This is normal and expected, especially on QLC-based drives. The size of the cache varies by drive model and firmware.
Is a higher sequential speed always better?
No. Sequential speed matters for tasks like video editing or transferring large files, but most everyday use (booting, opening apps, gaming) depends more on random 4K read speed and latency. A drive with 3 GB/s sequential but poor 4K random can feel slower than one with 2 GB/s sequential but excellent random performance.
What is a good SSD speed for a typical user?
For an NVMe SSD, sequential read speeds above 3 GB/s and random 4K read IOPS above 50,000 are considered good. For SATA SSDs, sequential speeds around 550 MB/s and random 4K IOPS around 80,000 are typical. But real-world feel is more about latency and random performance than peak sequential numbers.
Should I test my SSD while it is empty or full?
Test it empty for a best-case number, and test it at your normal usage level (e.g., 70% full) to see what you actually get. SSDs slow down as they fill because the controller has to do more garbage collection. Both numbers are useful, but the full-drive test is more representative of daily use.
Does the file system affect the copy speed?
Yes. NTFS, APFS, and ext4 have different metadata overhead and allocation strategies. Also, copying from an NTFS drive to an exFAT drive can be slower due to different cluster sizes. For a fair test, use the same file system on both source and destination, and format the target drive with a large cluster size if you plan to store big files.
Related
Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.