If you’ve ever taken a look at a solid state drive, you’ve probably seen a variety of performance figures. But what’s the meaning behind them? How important are they?
In How to Choose the Right Storage for Your Gaming PC, we covered the most common types of storage: Standard hard disk drives (HDDs) and common types of solid state drives (SSDs), reviewing how they best fit in a gaming PC. Today, we discuss where speed comes in.
Random and Sequential: What’s the deal?
Both on specifications pages and in general, there are two main ways that a storage drive’s speed of accessing data will be measured: Randomly, and sequentially.
Simply put, sequential access is when data is either read or written in a regular, orderly back-to-back fashion. The data is all right next to each other (not necessarily physically, but let’s not get into that).
On the other hand, random access is, well, when data isn’t accessed in a regular, back-to-back fashion—even if it’s just one block of data off.
Picture yourself grabbing some bottles off of a shelf. If you do it sequentially, you’re taking everything off in a row, without skipping. If you do it randomly, you might reach into different rows and at different points in the rows. Simple enough, right?
Units of Speed: IOPS, MB/s, QD? Oh my!
But that’s not all the details you will encounter. Oftentimes, you see speeds listed as not just the usual MB/s we see in many places in the world of technology—but also as IOPS, sometimes associated with a vague “Up to [whatever number] IOPS!” marketing claim.
MB/s, or megabytes per second, is a measure of throughput. That is, it straightforwardly tells us the amount of data moved each second. Simple enough, though it doesn’t tell us anything about latency (the time it took for any individual reads or writes to be completed after they were requested). Thankfully, outside of a few very-low-end SSDs in momentarily-high-use situations, you likely won’t experience any noticeable latency with an SSD.
IOPS, or Input Output Operations per Second, however, are a measure of the number of individual reads and writes (altogether) in a second. But a measure of IOPS tells us nothing about performance, really, unless it’s associated with more information. The IOPS multiplied by the size of the data gives us throughput. For example, 500 IOPS at 2 MB per operation would be 500 MB/s.
On a specification page, IOPS are sometimes used in reference to a certain data size (usually 4KB of data) being read or written, which does provide more information; but ultimately, those IOPS figures will never be reached in a realistic scenario—each access takes time to complete (latency), and that itself will also vary.
Alongside random read/write figures, you may see QD followed by a number, usually 1 or 32. QD stands for queue depth. This is the number of uncompleted, queued reads and writes that the device can handle.
Picture a line/queue at a grocery store. SATA SSDs, as well as old PCIe SSDs (from before NVMe saw wide implementation), can only have one queue of 32 at most, but NVMe SSDs can have a whopping 65,536 simultaneous queues, each with up to 65,536 queued accesses. So, a SATA SSD is like a typical discount retail store that only ever has one person working the registers for some incomprehensible reason, while NVMe SSDs are like an ideal store’s checkout section from a dream: practically infinite staffed cash registers. That said, it is doubtful that someone who won’t be hammering their SSD hard (and trust us, if you’re reading this, it’s extremely unlikely you’ll be pushing your SSD as hard as we mean) will ever go past a QD of about 6, even on a SATA SSD.
Going Beyond the Specifications
In truth, the specs tell us—well, not much of anything, really. From a technical standpoint, it is impossible for them to do so. This is because SSD performance varies depending on: the size of the data being read or written, the randomness of the data being read or written, the mixture of reads and writes, the queue depth, and how much the data can be compressed (if applicable to the SSD).
In specification lists, these factors are either really polarized or completely unknown. But in real-world applications, these will all vary constantly. Throw in multitasking and the waters get even more muddy. It takes rigorous testing that typical users (and most hardware reviewers, for that matter) are not capable of to reveal the impacts of these factors across different SSDs.
4KB random read/write is theoretically the most relevant, as 4KB (almost all random) accesses are the most common in your typical operating system like Windows. In my experience, they tend to make up around 47% of all accesses to the SSD while doing things like web browsing or using Discord, though some people find larger numbers closer to 60%. So if you’re going to focus on one of the data points provided about the drives you’re considering for your next system, that might be one to watch. However, the same ideas above still apply here, making the specifications somewhat irrelevant in practical terms.
So How Much Does All of This Really Matter Then?
For the average PC builder in today’s market? Not that much.
As long as you are already using at least a decent modern SATA SSD, you won’t see an appreciable benefit by upgrading. Really, as long as the specs don’t show anything that is suspiciously low for an SSD compared to others you’re considering, it’s not worth focusing on the performance specs. Yes, I really mean what I’m saying—even when comparing with an NVMe SSD capable of gigabytes per second speeds.
Wait, NVMe provides far better performance, so shouldn’t the discrepancy be big enough to make a difference?
Boot times won’t see a large benefit. This has been shown time and time again across independent benchmarks as well as reviews, such as those on The Tech Report and TechPowerUp. Except in the rare off-case, your choice of motherboard is likely to make a bigger difference, as this German review shows in the bottom chart.
Game loading times? Not that either, in the overwhelming majority of cases, unless you’re loading something like a lot of very-high-resolution textures. The bottleneck is largely the CPU once you have at least a modern SATA SSD (since the CPU has to decompress the game’s files, load objects, and set up the scene). There may be a few games that benefit, though nowhere near the extent of benefit seen in upgrading from a hard disk drive to a SATA SSD.
Video editing? Not even that sees significant benefit, unless editing in, and exporting to, large raw formats without enough RAM when using the SSD as a scratch drive. Even high-quality compression formats often used in professional, high-end applications can’t tax a decent SATA SSD.
Photo editing? Perhaps if you work with very large images… once again, without enough RAM while using the SSD as a scratch drive.
As you can probably figure already, NVMe’s usefulness for the average user is quite limited. Such SSDs are mostly useful for the enterprise sector. Like mentioned in our article, How to Choose the Right Storage for Your Gaming PC, “Ordinarily, the extremely high read/write speeds of an NVMe drive are really only going to be useful if transferring large files such as video, audio, or images.” Modern software simply doesn’t access storage in such a way that makes NVMe SSDs particularly beneficial—it just doesn’t happen to a large or frequent enough extent.
And if even NVMe SSDs don’t provide an appreciable benefit, you certainly won’t gain much from a high-end SATA SSD compared to any SSD that isn’t bottom of the barrel (and perhaps even then).
That said, perhaps the most important thing to consider for most regular users is sustained write performance. That is, write speeds when writing for potentially many gigabytes without stopping. You see, many SSDs use a buffer, since the memory chips that store the data long-term are relatively slow. Everything being written on to the SSD passes gets put in this buffer, which is meant to be much faster than the main storage chips, before being stored. This way, you see a high level of performance most of the time. But, once this buffer—if present—fills up, performance will almost always drop. This may happen when, for example, copying or decompressing large amounts of data, or when installing a game or other large program. Some SSDs, especially ones that are low-end and/or low-capacity, may drop to HDD-level speeds; some may be in-between HDD speeds and their typical speeds; and some may never experience any significant drop at all. The point at which performance drops will also vary depending on how much of the SSD’s capacity is filled. Unfortunately, there is no indication of this aspect of performance from manufacturers, by typical benchmarking software, or even in some professional reviews.
Solid state drive performance specs and common benchmarking software don’t provide performance figures that are relevant to the real-world. This is because, as stated earlier, “SSD performance varies depending on: the size of the data being read or written, the randomness of the data being read or written, the mixture of reads and writes, the queue depth, and how much the data can be compressed (if applicable to the SSD). In specification lists, these factors are either really polarized or completely unknown. But in real-world applications, these will all vary constantly.”
Sustained writes are perhaps the most relevant aspect of SSD performance for most users day-to-day, as you may see a sudden drop in performance when doing something that writes a lot of data to the SSD. For this, TechPowerUp reviews make for good multi-faceted references, and KitGuru’s “Real Life File Testing” also provides an interesting real-world look at copying large numbers of files.
We recommend not comparing SSDs based on their provided performance specifications, as they don’t translate to real-world performance. The aforementioned TechPowerUp reviews provide a decent array of real-world tests, which are unfortunately few-and-far-between in the world of SSDs. But, for most people, any commonly-suggested SSD from a reputable manufacturer should do the job just fine without ever needing to look at a bunch of tests.
If you have any questions for us, or any comments, feel free to use the comments section below!