IOPS and disk throughput

Type ms, block KB, and parallelism. 0.1 ms gives 10000 IOPS. 5 ms gives 200 IOPS. 1 ms gives 1000 IOPS. Typed IOPS, not a disk bench.

This is ⌊1000 / ms⌋ from typed latency. Parallelism does not scale the primary. Not a vendor spec. File time sits on file download time.

Input data

Enter average latency in ms — not a vendor “5000 IOPS” spec. The calculator derives IOPS (⌊1000 / ms⌋) and throughput from latency and block size.

Quick examples:

Results

Enter data and click Calculate.

How it works

IOPS and disk throughput in this calculator take the typed latency. 0.1 ms gives 10000 IOPS. 5 ms gives 200 IOPS. 1 ms gives 1000 IOPS. The primary formula is floor of 1000 / ms. The calculator does not run fio and does not read a disk spec.

Field iops-ms is milliseconds. Field iops-blocksize is KB. Field iops-parallel is in the fill. 1000 / 0.1 = 10000. 1000 / 5 = 200. 1000 / 1 = 1000. Parallelism 32 at 5 ms does not change 200 on the extras card.

10000 IOPS is not a CrystalDiskMark result. 200 IOPS does not come from a vendor QD curve. 1000 IOPS does not know a read/write mix. You type the ms. This is not a bench.

File download time next door divides a file by Mbps. RAID sizes capacity. Here 0.1 ms stays 10000 IOPS, the latency model alone.

Type 0.1, block 4, and parallelism 1, then Calculate. The result is 10000 IOPS. 5 ms, block 8, and 32 still give 200 IOPS on the primary.

0.1 ms gives 10000 IOPS. 5 ms gives 200 IOPS. 1 ms gives 1000 IOPS. Another latency changes 10000.

Formula

IOPS = ⌊1000 / ms⌋. MB/s = IOPS × blockKB / 1024. The result is unscaled by queue depth.

How to use

  1. Type 0.1 ms, block 4 KB, and parallelism 1.
  2. Click Calculate. The result is 10000 IOPS.
  3. 5 ms gives 200 IOPS. 1 ms gives 1000 IOPS.
  4. Typed latency, not a disk bench.
  5. The next card times a file from Mbps.

0.1 ms gives 10000 IOPS

IOPS = ⌊1000 / ms⌋. 0.1 ms gives 10000 IOPS. Not a disk bench.

IOPS
A result from latency. 0.1 ms leaves 10000 IOPS. Not fio.
disk
Typed ms, not a measurement. 5 ms gives 200 IOPS. Not vendor QD.
throughput
IOPS times block. 1 ms gives 1000 IOPS. Not a spec.

Examples

Example 1

  • 0.1 ms
  • block 4 KB
  • parallel 1

10000 IOPS

What IOPS at 0.1 ms? 10000 IOPS. Typed latency, not a bench.

Example 2

  • 5 ms
  • block 8 KB
  • parallel 32

200 IOPS

What about 5 ms and parallelism 32? 200 IOPS. The result without a queue multiplier.

Example 3

  • 1 ms
  • block 64 KB
  • parallel 1

1000 IOPS

What about 1 ms? 1000 IOPS.

Related calculators

File Download Time Calculator

File Download Time Calculator: size, unit, speed, and Mbps.

Use calculator →

RAID Capacity

RAID Capacity Calculator: disk count, size, and level.

Use calculator →

Cache Hit Ratio

Cache Hit Ratio Calculator: hits and misses.

Use calculator →

Data Transfer Time

Data Transfer Time Calculator: MB, Mbps, and efficiency.

Use calculator →

All IT calculators

Common questions

What IOPS at 0.1 ms?

10000 IOPS. 1000 / 0.1. Not a disk bench.

What about 5 ms and parallelism 32?

200 IOPS. The result is ⌊1000 / 5⌋, not × 32.

What about 1 ms?

1000 IOPS. 1000 / 1.

Is this a fio or CrystalDiskMark run?

No. You type ms. The calculator does not spin a disk.

Why iops-parallel?

It is in the fill as a note. The result stays IOPS without a queue scale.

Is 10000 IOPS an NVMe spec?

No. It is the 0.1 ms you type. The vendor sheet is outside this calculator.

Does a comma in 0,1 work?

Yes. 0,1 and 0.1 give 10000 IOPS.

How is this different from file download time?

That card turns 700 MB and 25 Mbps into 3.9 min. Here 0.1 ms gives 10000 IOPS.

Does zero ms count?

No. Without a positive latency there is no divide.

Knowledge sources

The calculator counts bits, bytes or throughput from your numbers. Below are SI and bit definitions (NIST).

Page updated in 2026.

IOPS vs throughput (MB/s)

IOPS is operations per second. Throughput is bytes per second (MB/s). This calculator derives IOPS from latency (⌊1000 / ms⌋), then MB/s from IOPS × block size. High IOPS does not always mean high MB/s — with small blocks you can have many operations and little data moved.

How IOPS × block size becomes MB/s

Simple relationship: throughput (KB/s) ≈ IOPS × block size (KB), then ÷ 1024 → MB/s. Intuition when you already know IOPS: 10,000 IOPS × 4 KB ≈ 40,000 KB/s ≈ 39 MB/s. The same IOPS at 64 KB blocks is ~640 MB/s — if the drive and controller can sustain it.

Optional parallelism (queue depth) multiplies effective IOPS in a simplified model — real SSDs/NVMe scale well; HDDs much less so.

Random vs sequential workloads

  • Random small blocks (4–8 KB) — typical for databases and indexes; IOPS and latency matter most.
  • Sequential large blocks (64 KB–1 MB+) — backups, media, scans; MB/s matters most.
  • Small random I/O hurts HDDs (seek) and stresses SSDs under heavy write; cloud IOPS limits are often separate for read/write.

HDD, SSD, NVMe and cloud IOPS limits

TypeRandom 4K IOPS (approx.)Typical latency
7200 rpm HDD~75–100~5–10 ms
SATA SSD~10,000–90,000~0.1–1 ms
NVMe SSD~100,000–1,000,000+~0.02–0.1 ms

In the cloud (EBS, Managed Disks, Persistent Disk) you often get a volume IOPS and MB/s cap — design databases against those limits, not against a local drive’s brochure IOPS.

Examples

  • 0.1 ms, 4 KB block → 10,000 IOPS ≈ 39 MB/s (same as 10k IOPS × 4 KB).
  • 2 ms, 64 KB block → 500 IOPS ≈ 31 MB/s (same as 500 IOPS × 64 KB).
  • Same IOPS, different block: 10,000 IOPS × 4 KB ≈ 39 MB/s; × 16 KB ≈ 156 MB/s — larger blocks raise MB/s when the workload is sequential.