Your SSD's Hidden Enemy: Why Random 4K IOPS Determines How Fast Your PC Feels

Your SSD's Hidden Enemy: Why Random 4K IOPS Determines How Fast Your PC Feels

Your SSD's Hidden Enemy: Why Random 4K IOPS Determines How Fast Your PC Feels

Your PC has been getting slower. Not in an obvious way — not crash-slower, not blue-screen-slower. Just that half-second pause before an app opens that wasn’t there last year. The slight stutter when Windows Explorer loads a folder full of files. Games that feel like they’re thinking before they let you in.

So you do what any reasonable person does. You run a health check. The drive comes back at 99% health. You check the read speeds in a benchmark tool — they look exactly like the numbers on the box. Everything seems completely fine.

And yet something is clearly not the same as the day you built this PC.

Here is the problem: you checked the wrong number. The speeds that show up in most drive health tools — and the ones printed largest on the box — describe something your PC almost never actually does. The metric that governs everything you just described has a different name, and most people have never once thought about it.

Why the Number on the Box Is Not the Number That Matters

Storage marketing has one job, and that job is to make the biggest possible number look like the most important one. For SSDs, that number is sequential read speed — how fast the drive can transfer a single large file from start to finish. It is what fills the banner on the product listing, and it is what most benchmark tools show by default.

The problem is that your PC almost never reads one large file from start to finish. When Windows boots, it is not reading one big file — it is reading hundreds of tiny ones simultaneously, scattered across the drive in no particular order. When you open Photoshop, your browser, a game, or a code editor, the same thing is happening. Small, scattered, random reads and writes in rapid bursts. The metric that measures this is called IOPS — input/output operations per second — specifically random 4K IOPS, which counts how many of those small 4-kilobyte operations the drive can handle every second.

Random IOPS is what your PC actually runs on for about 90% of normal use, and it degrades in ways sequential speed does not. A drive can show a perfectly healthy sequential speed while its random IOPS have quietly fallen to a fraction of where they started. The health check says 99%. The benchmark says fast. But you are living in the gap between what the marketing measures and what your daily experience runs on.

The Car Park Analogy

Imagine a multi-storey car park. When it is half empty, you pull in and find a spot in seconds — easy, no stress, in and out. That is a drive with plenty of free space handling your IOPS requests cleanly.

Now imagine the same car park at 95% capacity on a Saturday afternoon. You pull in and the attendant has to jog between levels, shift three cars to free up one space, bring someone else around from the third floor, and radio ahead before your car can park. Every transaction takes dramatically longer — not because the car park got physically smaller, but because the system has no slack left to work with.

Your SSD controller does exactly this when the drive is nearly full. Before it can write new data, it has to read what is there, consolidate it, erase the block, and rewrite it somewhere else just to make room. That overhead eats directly into IOPS. The drive is not broken. It is just out of breathing room. And the slower it gets at handling all those small random operations, the more you feel it every single time you click something.

The IOPS Risk Tiers: What Your Usage Profile Actually Means

Different users hit the IOPS wall for different reasons at different speeds. Here is an honest breakdown:

Who You Are	Biggest IOPS Risk	What to Watch	How to Protect It	addlink Pick
Casual user (docs, browsing, video)	Drive quietly filling up over years without noticing	Fill level — check every few months	Keep 20% free, set a reminder	S20 / S70 Lite
Gamer (installs, patches, game libraries)	Heat from GPU proximity with no heatsink, drive filling from installs	Drive temperature under load	Heatsink + case airflow, uninstall games you are not playing	S93 / A93
Content creator (video, rendering)	Sustained large writes exhausting the write cache mid-project	Drive temp during long renders	Batch big transfers, let the drive idle between sessions	S95
AI / developer (models, datasets)	Constant heavy random writes plus datasets eating free space fast	IOPS under sustained workload	Generous free space buffer, tiered storage setup	G55H

One thing the table cannot show: fill level and heat often compound each other. A hot drive with no slack to work with is not twice as slow — it is noticeably worse than either problem alone, because the controller is throttling its speed at the same moment it is working hardest to find free space.

What the Spec Sheet Is Actually Telling You (And What It Is Hiding)

Here is the part that nobody explains when you are buying a drive. The IOPS number on the spec sheet — the one that says something like “up to 1,000,000 IOPS” — is measured at queue depth 32. That means 32 requests stacked up and waiting simultaneously.

Your desktop PC, doing normal things, runs at queue depth 1 to 4. Almost never higher. The difference between what happens at QD1 and QD32 on the same drive is enormous — often 10 to 30 times. A drive rated for 500,000 IOPS at QD32 might deliver 17,000 IOPS at QD1, which is the depth your OS actually uses when you open an app or boot Windows.

This is also why NVMe drives feel noticeably snappier than SATA SSDs even when their sequential speeds are similar. It is not about bandwidth. NVMe communicates directly with your CPU over PCIe, with up to 65,000 parallel command queues. SATA uses a protocol designed for spinning hard drives: one queue, 32 commands deep. NVMe drives have around 10 to 20 microseconds of latency per operation; SATA drives run closer to 100 microseconds. At QD1 random reads — the exact workload that determines how instant your PC feels — that latency difference is what you are actually experiencing every time you click.

The metric worth looking at when choosing or evaluating a drive is QD1 4K random read performance, not the headline number. If a reviewer or benchmarking tool gives you that figure, it tells you far more about daily responsiveness than any MB/s claim.

Two Habits That Are Silently Costing You Performance Right Now

Most people assume an SSD either works or it does not. The reality is that it slides along a performance curve based on two things you control every day.

The first is how full you let it get. The controller on every consumer SSD reserves a portion of the drive as a working area for wear levelling, garbage collection, and the write cache that absorbs incoming data quickly before it gets sorted into its final home. When the drive fills past around 80 to 90%, that working area shrinks. The controller starts doing significantly more work per operation. Random write IOPS in particular can fall to a fraction of what the drive was doing when it had room. The fix is not complicated: treat roughly 80% used as your personal ceiling. That last 20% is not wasted space — it is what keeps everything else fast.

The second is heat, which most people never think about until it is already causing problems. An M.2 NVMe drive that hits sustained temperatures above around 75 to 80°C will throttle its own performance to protect itself. This is normal and intentional — the drive is protecting your data. But a drive that spends its working life running hot throttles more frequently, and that shows up as random slowdowns exactly when you are pushing it hardest. Drives mounted under a GPU without a heatsink, in cases with poor airflow, or in compact builds where M.2 slots get little ventilation are the most common culprits. Adding a heatsink — even a basic one — typically drops operating temperatures by 10 to 20 degrees, which is often the difference between a drive that throttles constantly and one that does not throttle at all.

FAQ

1. My drive health says 99% and the speeds look normal. Why does my PC still feel slower than before?

Because drive health percentages and sequential speed tests do not measure what you are noticing. They check for physical NAND wear and large-block transfer speed, respectively — neither of which captures random 4K IOPS at low queue depth. To get a useful reading, run CrystalDiskMark and look at the Q1T1 (queue depth 1, thread 1) random read and write rows. Those numbers describe your actual daily experience. If they are significantly lower than your drive’s rated specs, fill level and temperature are the first two places to look.

2. Does it make any difference whether I have Windows or macOS?

One thing does matter: TRIM support. TRIM is a command that tells your SSD which data blocks have been deleted, so the controller can prepare them for reuse in advance rather than scrambling to clean them up mid-write. On Windows 10 and 11, TRIM is enabled by default for NVMe drives. You can verify it by opening a command prompt as administrator and running fsutil behavior query DisableDeleteNotify — a result of 0 means TRIM is on. On macOS, TRIM is enabled by default for internal SSDs. If TRIM is somehow disabled, writes slow down and random IOPS suffers noticeably because the controller is constantly trying to clean up blocks it had no warning about.

3. If I clear space on a nearly-full drive, does performance come back?

Yes, meaningfully. The IOPS degradation from fill level is largely reversible — once the controller has breathing room again, it can rebuild its cache and garbage-collect properly. You will likely notice the improvement within a day of regular use. The one thing that does not fully recover is write-amplification damage accumulated over a long time, but for most users over a normal drive lifespan, clearing space will return noticeably better random write performance fairly quickly.

4. What is the actual difference in feel between a SATA SSD and an NVMe drive day to day?

The gap is real but often misunderstood. Sequential speeds — the big MB/s numbers — are where NVMe looks dramatically faster on paper, but that rarely translates directly to a faster-feeling desktop because most tasks are not moving large files. What you actually feel is the latency difference: NVMe drives respond to each operation in around 10 to 20 microseconds; SATA drives take around 100 microseconds per operation because of the protocol overhead. At queue depth 1, across thousands of tiny operations per minute, that five to ten times difference in response time adds up. Boot times, app launches, and general system snap are noticeably crisper on NVMe — not because it moves data faster in bulk, but because each individual response arrives sooner.

5. My drive runs hot under load. Is a heatsink actually worth it, or is it just cosmetic?

It is functional. Testing consistently shows that bare M.2 Gen4 and Gen5 drives — especially in slots near a hot GPU — can hit 85 to 95°C under sustained load within a minute or two, at which point the controller starts throttling performance to protect the hardware. A heatsink with proper thermal contact typically holds the same drive under 70°C, at which point throttling does not engage at all. It is not a visible improvement in normal use, but it becomes obvious the moment you are doing something that pushes the drive hard — a large file transfer, a game install, a render export — and you stop seeing the slowdowns that were quietly interrupting it before. The addlink G55H ships with a heatsink for this exact reason: at Gen5 speeds and sustained AI workloads, thermal management is not optional.

The Verdict: So What Should You Actually Do?

Here it is in plain terms:

If your PC has gradually gotten less snappy and you have not changed anything: Check your drive’s fill level first. If it is above 80%, that is almost certainly your answer. Clear space, let the drive idle for a few hours, and retest your QD1 random performance in CrystalDiskMark. For a replacement when the time comes, the addlink S20 or S70 Lite handles every light daily use case without the overhead you will never need.

If you are a gamer with a drive that is always hovering near full between game installs: You are hitting both problems at once — fill level eating into your write cache, and heat from the GPU slot compounding it during play. The addlink S93 and A93 are built for exactly this kind of bursty gaming workload, and their slim single-sided design fits cleanly into slots where thermals are already tight.

If you are a content creator whose drive slows down mid-project on long renders or exports: The issue is write cache exhaustion — the drive absorbs writes quickly at first, then slows when the buffer fills. The addlink S95 gives you the capacity to keep meaningful free space even with large project files, and the headroom to absorb big sustained writes without collapsing mid-session.

If you are running AI workloads or writing large datasets regularly: Your use case will genuinely stress a drive’s random write IOPS under sustained load, and heat will be a consistent factor. The addlink G55H — Gen5, heatsink included — is built for this kind of workload. Keep at least 20% of it free, and pair it with a second drive for overflow. That combination will not bottleneck what you are doing.

You can explore the full addlink SSD lineup at addlink.com.tw/m-2-pcie-ssd. Before anything else though, open CrystalDiskMark and run the Q1T1 test on whatever drive you are using right now. The number you see is the truest single measure of how fast your PC actually feels — and it might surprise you.

Amazon Store