Buried in AS SSD Benchmark’s View menu is a switch that rewrites every speed in the results grid: MB/s or IOPS. Same test, same run, two different languages — and endless forum confusion when one person’s screenshot is in one unit and another’s is in the other. This page explains what each unit actually measures, how to convert between them (with a worked example built from a published review result), and which view to use for which job. If you have not run the test yet, it is on our download page.

The two units, defined plainly

MB/s (megabytes per second) is throughput: how much data moved per second. It answers “how much?” and it scales with block size — streaming a 1 GB file produces a huge MB/s figure because each request carries a large payload. IOPS (input/output operations per second) is a count of requests completed per second, regardless of how much data each carried. It answers “how many?” and it exposes the per-request cost that throughput hides: a drive completing 100,000 tiny requests per second is working very differently from one completing a few hundred enormous ones, even if both show the same MB/s. Neither unit is more correct; they describe different properties of the same behaviour, which is why the tool offers both.

How the conversion works

The two units are linked by block size, through one identity: throughput = IOPS × block size. For the 4K rows, the block size is 4 KB, so each IOPS is worth 4 KB/s of throughput. That gives the practical conversions for any 4K figure:

IOPS = (MB/s × 1,024) ÷ 4  and  MB/s = (IOPS × 4) ÷ 1,024

A worked example from a published result: in its Phison E12 reference-drive review, Legit Reviews reported the 4K-64Thrd row in IOPS — 441,299 IOPS read and 427,221 IOPS write. Convert the read figure: 441,299 × 4 KB = 1,765,196 KB/s, which is approximately 1,724 MB/s in the MB/s view. Run the same review’s single-thread 4K read of 57 MB/s the other way — 57 × 1,024 ÷ 4 ≈ 14,592 IOPS — and you can see why reviewers quote IOPS for random tests: “fourteen and a half thousand operations per second” communicates the request rate that “57 MB/s” conceals. Note that the sequential row does not convert this way in any useful sense, because the Seq test streams a 1 GB file rather than issuing 4 KB requests; IOPS is a meaningful lens for the 4K and 4K-64Thrd rows, not for Seq.

When to use the MB/s view

MB/s is the default view and the right default for most people. It matches the units on drive packaging and in most reviews — including every figure in our archive of real drive results — so comparisons line up without arithmetic. It is also the only sensible unit for the sequential row and for the Copy-Benchmark, where the payload per request is large and variable. If you are checking your drive against a published review, as our scoring guide recommends, stay in MB/s: that is the view the review almost certainly used.

When to use the IOPS view

Switch to IOPS when the request rate itself is the question. Drive manufacturers quote random performance in IOPS on specification sheets, so the IOPS view is the one that lets you compare a measured 4K-64Thrd result against the number on the box — the Phison E12 review above did exactly this, noting that its measured 441,299 IOPS read sat below the drive’s rated figure of up to 600,000 IOPS. IOPS is also the natural unit for server and database work, where the workload is defined by how many small operations arrive per second, and for spotting queue-depth behaviour: the jump in IOPS between the 4K row (one thread) and the 4K-64Thrd row (64 threads) shows how much parallelism the drive can turn into completed requests.

Three traps the switch creates

Trap one: comparing across units. A screenshot in IOPS pasted next to one in MB/s makes the IOPS drive look thousands of times faster. Before comparing any two results, check the unit in the grid header. Trap two: assuming the score changes. It does not — the read, write and total scores are calculated from the underlying measurements (see the score formula), and switching the display changes only how the rows are labelled. Trap three: converting the Seq row. Dividing a sequential MB/s figure by 4 KB produces a meaningless “IOPS” number, because the sequential test never issued 4 KB requests. Convert only within the rows the conversion describes.

The short version

MB/s tells you how much data moves; IOPS tells you how many requests complete. Use MB/s for everyday checks and review comparisons, IOPS for spec-sheet checks and request-rate workloads, convert 4K figures with the block-size identity above, and never compare two screenshots until you have confirmed they are showing the same unit. For the rest of the results grid, see how to read AS SSD scores.

Sharing a result? State the unit in words

Most unit confusion starts when a bare screenshot travels without its header. When you post a result — in a forum, a support thread or a marketplace listing — write the unit into your text (“4K read 54 MB/s” or “4K-64Thrd 441,299 IOPS”), and if you switched the view before taking the screenshot, say so. It takes one line, and it saves everyone who reads your result from silently comparing it against numbers in the other unit — the single most common misreading of this benchmark’s output.

This page is an archive, not an estimate. Every number in the table below was recorded by a named review site testing a specific drive with AS SSD Benchmark, and every row links to the original review so you can check the figure yourself. None of these results was measured by this site — they are published review data, collected so you can compare your own run against a real result for the same drive instead of a guessed “typical” range. If your drive is not listed, the method in what is a good AS SSD score shows how to find its review. You can produce your own result with the tool from our download page.

The results table

Speeds are MB/s, shown as read / write. A dash means the review’s text did not state that row — we reproduce exactly what each source published, nothing more.

How the sources tested

Legit Reviews states the AS SSD version in each of its reviews, and the versions vary because the reviews span years: the HyperX Predator and Samsung SM951 results were produced with 1.7.4739.38088, the Crucial BX200 with 1.8.5636.37293, and the Intel 660p, Kingston A1000 and KC1000, Intel Optane 800P and Phison E12 results with 2.0.6485.19676; the WD Blue SN500 review used 2.0.6821.41776. RelaxedTech’s results come from a single comparison review of the Samsung 870 EVO, in which the 970 EVO Plus, Crucial P1, 870 EVO, 860 EVO, MX500 and TR200 were all tested on one bench — those six rows are directly comparable with each other. KitGuru’s Samsung 970 EVO Plus 1 TB review (AS SSD 2.0) is a useful extra data point in a different format: it reports a read score of 2,133 and a write score of 2,258 for the 1 TB model, against the 500 GB model’s total of 4,931 in the table — a reminder that capacity changes results within the same product family.

What the table shows

Three patterns are worth reading out of the data. First, the interface ceiling is visible. The five SATA drives’ totals sit between 823 and 1,242 and their sequential reads between 508 and 558 MB/s — the whole class pressed against the same link limit, as SATA vs NVMe scores explains. Second, sequential speed does not rank the drives. The Intel Optane 800P has the lowest sequential write figure of any NVMe drive in the table (611.58 MB/s) yet a total of 3,309, because its 4K read of 152.75 MB/s is nearly three times the next best in that column — the score formula rewards the 4K row, as Sequential vs 4K discusses. Third, version discipline matters. The two oldest results (HyperX Predator, SM951) were produced under AS SSD 1.7; their totals are historical records, not numbers to rank against the 2.0 rows beneath them.

Extra figures published by the same reviews

Some reviews published rows beyond the table’s columns. Legit Reviews reported the Phison E12 reference drive’s 4K-64Thrd figures in IOPS — 441,299 IOPS read and 427,221 IOPS write (the IOPS view is explained in MB/s vs IOPS explained) — and Copy-Benchmark ranges for several drives, for example 635 MB/s (Program) to 1,783 MB/s (ISO) for the Intel 660p and 486 / 323 / 407 MB/s (ISO / Program / Game) for the Crucial BX200; see Copy-Benchmark explained for how to read those. The compression behaviour behind some of these results is covered in Compression-Benchmark explained.

Using this archive honestly

Compare your result only with the row for your exact drive and capacity, note the version that produced it, and match your test size to the 1 GB default most reviews use. If your numbers land close to the published row, your drive is performing as reviewed; if one row is far off, that row — not the total — tells you where to investigate. And if you are choosing a new drive on benchmark evidence, weigh the 4K column at least as heavily as the sequential column; our best NVMe SSDs guide builds on these same published results. This archive will grow as we verify further published reviews; every addition must pass the same test as the rows above — a named source, a specific drive, and numbers we have read in the original review.

The last row of the AS SSD Benchmark grid, Acc.time, is the easiest to overlook and the only one where a smaller number wins. It reports access time — latency — in milliseconds, separately for reading and writing. Throughput rows tell you how fast data flows once the drive gets going; access time tells you how long the drive takes to get going at all. This page explains how each of the two figures is measured, why they differ, and how to judge the numbers on your own result. Run the benchmark from our download page to follow along.

How the read access time is measured

Read access time in AS SSD is determined across the entire capacity of the drive — a full-stroke measurement, as the tool’s official description (carried by TweakTown’s review archive and the major download sites) puts it. The benchmark requests data from positions spanning the whole drive rather than from one convenient corner, so the figure represents the average time to begin reading from anywhere, not from the fastest neighbourhood. That makes the read figure a conservative, whole-drive latency measure: a drive cannot flatter it by being quick only in one region.

How the write access time is measured

Write access time is measured using the 1 GB test file — the same default test size the main synthetic tests use. It captures how long the drive takes to begin and complete write accesses against that file. Because the two directions are measured differently — full-capacity reads versus a 1 GB write pattern — the read and write figures are not two readings of one quantity. Treat them as two separate diagnostics, each meaningful on its own terms, and expect them to differ.

Why write access time is usually the larger number

On most results you will ever see, the write figure exceeds the read figure, sometimes by a wide margin. The reason is structural. Reading flash is a lookup: the controller translates an address and returns the stored charge state. Writing flash is a programme operation, potentially preceded by block management work inside the drive, and on a busy or nearly full drive it can involve the controller shuffling existing data out of the way first. Caching hides much of this on a healthy drive with free space, which is why the write figure is also a useful canary: when it balloons relative to the same drive’s earlier results, the drive’s free space, cache behaviour or background state has changed. A sourced example below shows a completely typical gap.

A real example, with its source

In its review of the Lexar NQ100 240 GB, a DRAM-less SATA drive, Assistouest Informatique published a full AS SSD row set: sequential speeds of 510 MB/s read and 323 MB/s write, 4K results of 18 MB/s read and 57 MB/s write, 4K-64Thrd results of 53 MB/s read and 132 MB/s write — and access times of 0.05 ms read and 0.64 ms write. Two lessons sit in that pair of numbers. First, the write access time is more than twelve times the read figure on an ordinary, healthy budget drive; a gap like that is normal, not a fault. Second, even the “slow” figure here is a fraction of a millisecond — the review itself notes these access times as excellent next to a mechanical drive, which is the context that matters: access time is where the felt difference between flash and spinning media lives, far more than in any throughput row.

How access time relates to the other rows

Access time and the 4K row are two views of the same underlying quality. A drive that begins each small access quickly posts both a low Acc.time and a strong 4K result; a drive with poor small-access latency is weak in both. Access time is not, however, an input to the total score — the simplified score formula is built from the Seq, 4K and 4K-64Thrd rows only — so treat Acc.time as diagnostic evidence rather than a scoring component. It is also unit-sensitive: AS SSD always shows this row in milliseconds, even when you switch the speed rows to IOPS with the View menu (see MB/s vs IOPS explained), because latency is a time, not a rate.

Reading your own Acc.time result

Judge the row in three passes. Direction: lower is better, in both columns — the only row in the grid where that is true. Proportion: compare the read/write gap with published results for your drive or its class rather than expecting the two to match; a wide gap is typical, and the Lexar example above shows how wide “typical” can be. Change over time: the most valuable use of this row is longitudinal — re-run the benchmark periodically under the same conditions (same version, same test size, similar free space) and watch for drift, especially in the write figure, which reacts first to a filling drive. If a single row has to summarise why an SSD feels instant, it is this one — and now you know exactly how both halves of it are measured. For the rest of the grid, see how to read AS SSD scores.

A note on comparing access times with reviews

One frustration to expect: many reviews print the speed rows and the total but leave access times inside their screenshot rather than their text, so a like-for-like published Acc.time for your exact drive can be hard to find. That is a gap in the reviews, not in your result. The Lexar example above is valuable precisely because the reviewer stated both figures in words. Where no sourced figure exists for your drive, fall back on the longitudinal method — your own earlier runs, under identical settings, are the most honest baseline an access time can have.

Every drive advertises its sequential speed, and AS SSD Benchmark prints it in the first row — so it is natural to treat the Seq row as the verdict. It is not. The sequential test streams one large file from end to end, a pattern your computer almost never produces outside of copying a movie or an ISO image. The 4K rows measure something much closer to your actual day. This page explains the difference, why the benchmark’s own scoring sides with 4K, and shows a published head-to-head in which the drive with the lower sequential speed won the total score comfortably. To see your own rows, run the tool from our download page.

What sequential speed measures — and when it matters

The Seq test reads and writes a 1 GB file in one continuous stream. Nothing interrupts it, nothing competes with it, and the drive can run its flash channels flat out. That is a real capability, and it governs real tasks: copying large files, capturing video, unpacking big archives, installing games from a local image. If your work is moving large media files, the Seq row deserves your attention. But notice how narrow that list is. Booting Windows, launching applications, loading game levels, compiling code, browsing — none of these stream one file end to end. They issue thousands of small, scattered requests, which is a different test entirely.

What the 4K rows measure

The 4K row reads and writes randomly placed 4 KB blocks, one at a time — the access pattern of an operating system pulling in libraries, registry entries, fonts, thumbnails and application data from all over the drive. The 4K-64Thrd row runs the same pattern with 64 threads queued at once, modelling a busy system with many programs active together. Both rows are bottlenecked by latency and by how efficiently the controller handles small requests, not by raw link bandwidth. That is why their numbers look tiny next to the Seq row, and why they vary so much between drives that share identical sequential ratings. Our guide to reading the results grid covers each row in context.

The scoring formula already picked a side

You do not have to take our word for which row matters — the benchmark’s author encoded the answer in the score formula. In the simplified published form, sequential speed is multiplied by 0.1 before it counts, while the 4K and 4K-64Thrd results count at full value. A drive must therefore be ten times faster sequentially to gain what a modest 4K improvement delivers. The total score is, by design, mostly a random-performance score with a sequential garnish. When two drives’ totals surprise you, the explanation is almost always sitting in their 4K rows.

Proof from published results: the slower-Seq drive that won

Legit Reviews’ archive contains a near-perfect natural experiment. The Intel Optane SSD 800P 118 GB (AS SSD 2.0.6485.19676) posted sequential speeds of 1,335.90 MB/s read and 611.58 MB/s write — unremarkable numbers, lower in both directions than many ordinary drives. But its 4K results were 152.75 MB/s read and 129.01 MB/s write, far beyond what NAND-based drives typically manage in that row, and its total came to 3,309. Compare the Samsung SM951 512 GB (tested with AS SSD 1.7.4739.38088, so its total is not on the same version scale — compare its rows, not its score): sequential speeds of 1,988 MB/s read and 1,510 MB/s write, roughly 49% faster reading and two and a half times faster writing than the Optane drive, but a 4K read of just 45 MB/s against the Optane’s 152.75 MB/s. The Optane drive was built for low-latency small-block work, and in the row that models daily use it was more than three times faster — while losing the headline row badly. Anyone who had ranked these two drives by their Seq numbers alone would have ranked them backwards for desktop responsiveness.

How to weight the rows for your own use

Match the row to the workload. General desktop, office and gaming: read the 4K row first, then 4K-64Thrd, and treat Seq as background information. Video editing, large backups, file-server duty: the Seq row and the Copy-Benchmark’s ISO scenario (see Copy-Benchmark explained) become genuinely important, because sustained streaming is your actual workload. Mixed or unknown use: the total score is a reasonable shorthand precisely because it is 4K-weighted — but glance at the rows anyway, since a total can hide one collapsing row behind two strong ones. And in every case, compare only against results from the same AS SSD version and test size, ideally a professional review of your exact drive, as set out in what is a good AS SSD score. The Seq row is the loudest number on the screen. The 4K rows are the honest ones.

One more place the same lesson hides

The Copy-Benchmark tells the same story in plain language. Its ISO scenario — two large files — is a sequential workload, and it is always the fastest of the three. Its Program scenario — a folder full of small files — is a random workload wearing a copy dialog, and it is always the slowest, on every drive, by a wide margin. Legit Reviews’ figures for the Crucial BX200 make the point concretely: 486 MB/s in the ISO test against 323 MB/s in the Program test, from the same drive in the same minute. Nothing about the drive changed between those two runs; only the shape of the work changed. Your operating system spends its day shaped like the Program test, not the ISO test — which is the whole argument of this page in a single pair of numbers.

Run AS SSD Benchmark on a SATA SSD and on an NVMe SSD and the results look like different sports: the sequential row multiplies several times over, the 4K-64Thrd row explodes, yet the humble single-thread 4K row barely moves. None of that is a benchmarking error. Each row is faithfully reporting a different structural difference between the two interfaces. This page explains, row by row, where the gap comes from — and backs it with published results measured on a single test bench, so the comparison is honest. If you want to reproduce any of this, the tool is on our download page.

The sequential gap is a link-speed gap

SATA III carries data over a 6 Gb/s link that was designed for mechanical drives. However fast the flash behind it becomes, a SATA drive’s sequential throughput stops at what that link can carry — in practice, review results for good SATA drives cluster right up against the link’s ceiling. NVMe drives attach directly to PCIe lanes, and each PCIe generation roughly doubles the available bandwidth, so the sequential ceiling is far higher and keeps moving. AS SSD’s Seq row measures a single 1 GB file streamed end to end, which is precisely the workload that hits the link ceiling and nothing else. That is why the Seq row shows the largest SATA-to-NVMe gap of the whole grid: it is mostly measuring the cable, not the flash.

The cleanest illustration comes from a single review bench. RelaxedTech tested both interfaces on the same system: the SATA Samsung 870 EVO 500 GB recorded sequential speeds of 558 MB/s read and 549 MB/s write, while the NVMe Samsung 970 EVO Plus 500 GB on the same bench recorded 3,503 MB/s read and 3,285 MB/s write — more than six times the read throughput, from the interface change alone.

The 4K-64Thrd gap is a queueing gap

The second structural difference is how commands queue. AHCI, the protocol SATA drives speak, was built around a single command queue suited to a spinning disk that services one request at a time. NVMe was designed for flash from the start, with deep parallel queues feeding many flash channels at once. AS SSD’s 4K-64Thrd test — random 4 KB blocks spread across 64 threads — exists to expose exactly this difference, and it doubles as the classic illustration of why Native Command Queuing matters: drives stuck in IDE mode, where NCQ is unavailable, fall flat in this row even on fast hardware. This row is also the biggest single contributor to the total score (see the score formula), which is why NVMe totals pull so far ahead: the scoring rewards the queue depth that NVMe was built to exploit.

The single-thread 4K gap is small — and that is the interesting part

Now look at the row that barely changes. Single-queue 4K random access is limited less by the link than by latency: one small request, issued, waited on, completed. Doubling link bandwidth does almost nothing for it, which is why a strong SATA drive’s 4K row can sit surprisingly close to an NVMe drive’s. A sourced example: Legit Reviews measured the entry-level NVMe Intel SSD 660p 1 TB at 54 MB/s 4K read, while the same outlet’s SATA-era results for good drives are in a similar neighbourhood — the interface revolution mostly passed this row by. This is also why moving from SATA to NVMe rarely feels six times faster in daily use, whatever the Seq row says: the workload your operating system generates looks much more like the 4K row than like a 1 GB file stream. We expand on that in Sequential vs 4K scores.

Same bench, two interfaces: the totals

RelaxedTech’s same-bench results pull the whole story together. The SATA drives in that roundup — Samsung 870 EVO 500 GB at a total of 1,242, Samsung 860 EVO 500 GB at 1,238 and Crucial MX500 500 GB at 1,231 — finished within a few points of each other, all pressed against the same interface ceiling. The NVMe drives on the same bench finished far ahead: the Crucial P1 500 GB at 2,691 and the Samsung 970 EVO Plus 500 GB at 4,931. Notice what the totals are really saying: within one interface, drives are separated by small margins, because the interface is the binding constraint; across interfaces, the totals separate by multiples, driven first by link bandwidth in the Seq row and then amplified by queue depth in the 4K-64Thrd row.

How to use this when comparing your own results

Three practical rules follow. First, never judge an NVMe drive by SATA expectations or the reverse — judge each drive against published results for its own model, as our scoring guide sets out. Second, if an NVMe drive returns SATA-shaped results — sequential speeds stuck at SATA-like levels and a flat 4K-64Thrd row — suspect the installation before the drive: it may be seated in a slower slot, running at reduced lane width, or using a legacy driver mode, all of which AS SSD’s top information strip helps reveal. Third, remember that the totals also embed the AS SSD version, so compare only within a version. The full set of sourced results behind this page is collected in AS SSD scores compared. Read that way — each drive against its own interface class, each row explained by the structure behind it — the SATA/NVMe gap stops looking like a verdict on your drive and starts looking like what it is: two different connection technologies, measured honestly by the same small tool.

It is the first question everyone asks after a run: is this score good? The honest answer is that there is no universal scale to check it against — and any page that hands you a tidy score range for an entire class of drives, without showing where the numbers came from, is guessing. A total score depends on the drive model, its capacity, the AS SSD version, the test size, the driver mode and the host system. This guide takes the only approach that actually works: compare your result, row by row, with a published professional review of the same drive. Run the test from our download page, then work through the method below.

Step 1: Find a review of your exact drive

Search for your drive’s exact model and capacity plus “AS SSD” — capacity matters, because larger models often write faster than smaller ones in the same family, and reviewers test specific capacities. Review sites such as Legit Reviews, KitGuru and TweakTown have published AS SSD results for hundreds of drives over the years. You want a review that prints the sequential and 4K rows and the total, and ideally states the AS SSD version used. Our own archive in AS SSD scores compared collects verified examples with their sources, and it is a good starting point.

Step 2: Match the test conditions before comparing anything

Two results are only comparable when the conditions match. Check three things. Version: a score from AS SSD 1.7 and a score from 2.0 are not on the same scale. Test size: the 1 GB default is what most reviews use; a 10 GB run on your machine will not line up with a 1 GB review run. Display mode: make sure you are reading MB/s, not IOPS — switch with the View menu if needed, as explained in MB/s vs IOPS explained. If the conditions differ, the comparison is void; change your settings and re-run instead of explaining the gap away.

Step 3: Compare rows, not just the total

The total is a weighted summary that mostly reflects 4K performance (see the score formula), so two drives can reach similar totals by different routes. Compare your Seq and 4K rows with the review’s rows individually. Small run-to-run variation is normal — a few percent either way on any row means “in line with the review”. One row collapsing while the others hold up is a diagnostic clue: a weak 4K-64Thrd row points at driver mode, a weak sequential write row on a nearly full drive points at the drive’s write cache or free space, and a bad alignment flag in the top strip can depress everything at once.

Four real reference points, with sources

To show what “in line with a review” looks like, here are four published results — each one a specific drive, tested by a named site, with the version noted where the review states it. None of these is a range or an estimate; each is the result that outlet actually recorded.

Samsung 870 EVO 500 GB (SATA) — RelaxedTech recorded a total score of 1,242, with sequential speeds of 558 MB/s read and 549 MB/s write. If your 870 EVO 500 GB lands near those rows under the same settings, it is performing as reviewed.

Crucial BX200 480 GB (SATA) — Legit Reviews (AS SSD 1.8.5636.37293) recorded sequential speeds of 520 MB/s read and 461 MB/s write and a total of 1,041. Note the older version: compare your BX200 run against this only with the version difference in mind.

Intel SSD 660p 1 TB (NVMe) — Legit Reviews (AS SSD 2.0.6485.19676) recorded sequential speeds of 1,747.2 MB/s read and 1,681.9 MB/s write, 4K results of 54 MB/s read and 136 MB/s write, and a total of 2,433.

Samsung 970 EVO Plus 500 GB (NVMe) — RelaxedTech recorded sequential speeds of 3,503 MB/s read and 3,285 MB/s write and a total of 4,931 — on the same test bench where the SATA 870 EVO above scored 1,242, which is why same-bench review comparisons are so much more trustworthy than numbers gathered from different systems.

What a “bad” result usually means

When a drive scores far below its own review results under matched conditions, the cause is far more often the setup than the silicon. The usual suspects, in the order we check them: the drive running under a legacy IDE-mode driver instead of AHCI or NVMe; a misaligned partition (AS SSD prints the alignment in the top strip); the drive installed in a slower slot or port than it was reviewed in; a nearly full drive whose write cache has no room to work; and thermal throttling on an M.2 drive under a long test. Work through those before concluding the drive itself is at fault — our troubleshooting cluster covers each of them.

The bottom line

“Good” is not a number — it is a match. A good AS SSD score is one that lines up, row by row, with what a competent review measured from the same drive under the same version and test size. Anything else — forum screenshots without context, class-wide ranges with no sources, scores from a different benchmark version — will either worry you needlessly or reassure you wrongly. If you are shopping on the strength of benchmark evidence, start from drives with strong published results; our best NVMe SSDs guide collects current options, and how to read AS SSD scores will help you read the evidence behind any of them.

Every AS SSD Benchmark run ends with three numbers: a read score, a write score and a total. They look official enough that people quote them as if they were measured — but they are calculated, from the four test rows above them, using a formula the author made public. This page sets out the simplified formula exactly as secondary guides document it, explains why it is weighted the way it is, and covers the two comparison mistakes the formula makes dangerous. You can run the test yourself from our download page.

The simplified formula

Secondary guides that document the author’s published formula give it in this simplified form (see sources below):

Read score = Seq Read × 0.1 + 4K Read + 4K-64Thrd Read
Write score = Seq Write × 0.1 + 4K Write + 4K-64Thrd Write

The total score is a combination of the read and write scores. Treat this as what it is labelled: a simplified form of the calculation, as reported by secondary guides such as PC Hardware Pro’s AS SSD walkthrough and MundoBytes, both of which describe the same structure — sequential speed weighted at a tenth, the two 4K results added at full weight. We deliberately do not print a more elaborate version of the formula: more complex variants circulate online without a verifiable source, and a scoring formula you cannot trace is worse than a simplified one you can.

Why sequential speed counts for almost nothing

The 0.1 multiplier is the most important design decision in the formula. A drive that reads sequentially at 3,000 MB/s contributes only 300 points from that row, while its 4K and 4K-64Thrd rows contribute their full MB/s values — and on a fast drive the 4K-64Thrd row alone can be worth well over a thousand points. In other words, the total score is mostly a random-performance score. That is a deliberate editorial choice by the tool’s author: small-block random access is what an operating system does all day, so a benchmark score meant to rank drives by how responsive they feel should reward 4K behaviour and treat sequential bragging rights as a minor input. It also explains a pattern that confuses many first-time users — a drive with modest sequential numbers but strong 4K results can outscore a drive with spectacular sequential numbers and ordinary 4K results. We show a real example of exactly that in Sequential vs 4K scores.

A worked sanity check

You can use the simplified formula to sanity-check any published result. Take the Intel SSD 660p 1 TB as reviewed by Legit Reviews (AS SSD 2.0.6485.19676): sequential speeds of 1,747.2 MB/s read and 1,681.9 MB/s write, and 4K results of 54 MB/s read and 136 MB/s write, with a published total of 2,433 points. The sequential rows contribute roughly 175 points on the read side and 168 on the write side under the simplified formula — a small fraction of the total. The rest has to come from the 4K and 4K-64Thrd rows, which is precisely the weighting the formula describes. If a screenshot you find online shows a total that cannot be reconciled with its own rows at all, treat the screenshot — not the formula — with suspicion.

Why totals cannot be compared across versions

The results archive on this site spans AS SSD 1.7, 1.8 and 2.0, because that is what review sites used at the time — Legit Reviews, for example, tested the Samsung SM951 with version 1.7.4739.38088 and the Phison E12 reference drive with version 2.0.6485.19676. Test internals changed between those generations, so a total from a 1.7 run and a total from a 2.0 run are not points on one scale. Comparing them as if they were is the single most common scoring error on forums. When you line results up side by side — as we do in AS SSD scores compared — note the version printed with each result and only draw conclusions within the same version.

Why test size matters too

AS SSD lets you choose a test file of 1, 3, 5 or 10 GB; the default is 1 GB. The choice changes what the rows measure. A 1 GB run fits comfortably inside the fast cache area of many drives, while a 10 GB run pushes well past it and exposes sustained-write behaviour that the short test never touches. Two totals produced with different test sizes are therefore answers to different questions, even on the same drive, in the same version, on the same day. When you publish or compare a score, state the test size next to it — and if a comparison does not state one, assume the 1 GB default but do not build conclusions on the assumption.

What the formula does not include

Two parts of the tool never enter the score at all. The Copy-Benchmark (ISO, Program and Game scenarios) reports its own speeds and times, but contributes nothing to the read, write or total score. The Compression-Benchmark draws a graph of performance against data compressibility — vital for understanding controllers that accelerate compressible data — but likewise adds no points; we explain that graph in Compression-Benchmark explained. Access time is displayed in the grid but is not an input to the simplified formula either; it is diagnostic information, covered separately in access time explained. Finally, remember what the score is: a compact summary of one short synthetic workload, on one system, on one day. Used that way — and compared like-for-like, as our guide to judging a score recommends — it is a genuinely useful number. Used as a universal ranking across versions, test sizes and systems, it is numerology.

Sources: simplified formula as documented by PC Hardware Pro and MundoBytes; review results from Legit Reviews as linked above.

An AS SSD Benchmark result window packs a lot into a small grid: four test rows, two columns of numbers, three scores at the bottom, and a strip of drive information at the top. Read in the right order, it tells you not just how fast a drive is, but where it is fast, where it is slow, and whether something in your system is holding it back. This guide walks through every element of the results screen, what it measures, and which direction is better. If you have not run the test yet, grab the tool from our download page first — it is portable, so there is nothing to install.

Start at the top: the drive information strip

Before looking at any speed numbers, read the lines above the grid. AS SSD shows the drive model, firmware version, controller or driver in use, the partition offset, and the total capacity. Two things here matter more than any score. First, the controller line tells you which storage driver the drive is running under — a legacy IDE-mode driver (often flagged as pciide) cannot use Native Command Queuing and will drag the random results down. Second, the alignment value should read OK. A misaligned partition shows as BAD and forces the drive to do extra work on almost every access. A low score with a BAD flag in this strip is a configuration problem, not a slow drive.

The Seq row: sequential speed

The first row, Seq, measures how fast the drive reads and writes one large 1 GB file in a single continuous stream. This is the number that resembles the headline speed on the box, and it is the best case a drive can achieve: no searching, no queue management, just sustained throughput. Sequential speed is what dominates when you copy a single large video file or install image. Higher is better, and the read and write columns often differ — many drives read considerably faster than they write. What the Seq row does not tell you is how the drive feels in everyday use, because an operating system rarely reads one file end to end. That is what the next two rows are for.

The 4K row: single-thread random speed

The 4K row measures reading and writing randomly scattered 4 KB blocks, one request at a time. The numbers look shockingly small next to the Seq row — tens of MB/s instead of hundreds or thousands — and that is normal. Random access to tiny blocks is the hardest work a drive does, and it is also the closest match to what Windows, applications and games actually request all day: small pieces of data from all over the drive. Of all the rows, this one tracks the subjective snappiness of a system most closely. A drive can post a huge Seq number and still feel ordinary if its 4K row is weak. We dig into that relationship in Sequential vs 4K scores.

The 4K-64Thrd row: random speed under load

4K-64Thrd repeats the 4K test but spreads the requests across 64 simultaneous threads. This exposes how well the drive and its controller handle a deep queue of commands — the scenario Native Command Queuing was built for, and the scenario where drives running in IDE mode fall behind, because IDE cannot queue commands this way. Server-style and multitasking workloads live here: many programs reading and writing at once. On a healthy modern drive this row is many times higher than the single-thread 4K row. If your 4K-64Thrd result is barely above your 4K result, suspect the driver mode shown in the top strip before you blame the drive.

The Acc.time row: access time

Acc.time is the only row where lower is better. It reports average access time in milliseconds, separately for read and write. The read figure is measured across the full capacity of the drive (a full-stroke measurement), while the write figure is measured with the 1 GB test file. Access time is latency: how long the drive takes to start delivering data, before throughput even enters the picture. Small differences here are a large part of why an SSD feels instant compared with a mechanical drive. We cover this row in depth in access time explained.

The scores at the bottom

Below the grid, AS SSD prints a read score, a write score and a bold total score. These are not extra measurements — they are a weighted summary of the rows above, calculated with a published simplified formula in which the sequential result counts for very little and the 4K results dominate. That weighting is deliberate: the score is designed to reflect responsiveness, not headline throughput. Two warnings belong with every total score. First, it is only meaningful against results produced by the same AS SSD version and the same test size, because both change the arithmetic. Second, a total on its own hides the shape of the result — always glance back at the rows to see why a drive scored what it scored. The full breakdown is in our guide to the AS SSD score formula.

The Copy-Benchmark: a separate, practical test

Under the Tools menu, the Copy-Benchmark runs three copy scenarios using the operating system’s own cached copy functions: ISO (two large files), Program (a folder of many small files) and Game (a mixture of large and small files). Each reports a speed in MB/s and a duration in seconds. Because it uses the OS cache rather than bypassing it, the Copy test behaves more like a real file copy than the synthetic rows do — and the Program scenario, with its swarm of small files, is usually the slowest of the three, for the same 4K reasons as above. Our dedicated walkthrough is the Copy-Benchmark explained guide.

A 60-second reading routine

When a fresh result appears, read it in this order: (1) top strip — driver mode and alignment both OK? (2) Acc.time — fractions of a millisecond, with no wild read/write gap? (3) 4K row — does it look proportionate to drives of the same class you have seen reviewed? (4) 4K-64Thrd — a healthy multiple of the 4K row? (5) Seq — in line with the interface the drive is connected through? (6) total score — compared only against same-version, same-test-size results for the same or similar drives. That routine catches almost every misleading result before it sends you shopping for a drive you do not need. To see how your numbers stack up, compare them with the verified review results collected in AS SSD scores compared, and if you are viewing results in IOPS instead of MB/s, switch back with the View menu — the two displays are explained in MB/s vs IOPS explained.