KB vs KiB: Why Your Disk Shows Less Space

The Missing Disk Space Problem

You buy a 1 TB hard drive, plug it in, and your operating system reports 931 GB of available space. Where did the other 69 GB go? Is the manufacturer cheating? Not exactly. The gap is caused by two legitimate but incompatible definitions of what "kilo," "mega," "giga," and "tera" mean when applied to bytes. Understanding this distinction is essential for any developer who works with storage, networking, or data engineering.

Two Systems, One Set of Prefixes

The problem dates back to the earliest days of computing. In the metric system (SI), "kilo" means exactly 1,000. But memory chips are addressed using binary, and the nearest power of 2 to 1,000 is 1,024 (2¹⁰). Early computer scientists co-opted the metric prefixes to mean powers of 1,024, creating a persistent ambiguity.

To resolve this, the International Electrotechnical Commission (IEC) introduced new prefixes in 1998 specifically for binary multiples: kibi (KiB), mebi (MiB), gibi (GiB), and tebi (TiB). The idea was simple: SI prefixes (KB, MB, GB) would strictly mean powers of 1,000, and IEC prefixes (KiB, MiB, GiB) would mean powers of 1,024.

Why the Difference Matters

At the kilobyte level, the 2.4% difference between 1,000 and 1,024 is negligible. But the gap compounds with each tier because each tier multiplies by a different factor (1,000 vs 1,024). By the terabyte tier:

• 1 TB (SI) = 1,000,000,000,000 bytes
• 1 TiB (IEC) = 1,099,511,627,776 bytes
• Difference: 9.95%

This means a 1 TB drive (1,000,000,000,000 bytes) contains only 0.909 TiB, which most operating systems round to 931 GiB. The "missing" 69 GB was never there; it is a consequence of two different definitions of the same prefix.

Who Uses Which System?

SI (Powers of 1,000)

• Hard drive and SSD manufacturers: All storage vendors advertise capacity in SI units. A "500 GB" drive contains exactly 500,000,000,000 bytes.
• Network speeds: ISPs advertise in bits per second using SI prefixes. "100 Mbps" means 100,000,000 bits per second, not 100 × 2²⁰.
• macOS (since Snow Leopard): Apple switched to SI units in 2009. A file of 1,000,000 bytes displays as "1 MB."
• iOS, Android: Both mobile operating systems use SI units.
• Scientific computing: SI units align with the metric system used in physics and engineering.

IEC / Binary (Powers of 1,024)

• Windows: Still labels sizes as "KB," "MB," "GB" but actually calculates using powers of 1,024. This is technically incorrect per the IEC standard, but Microsoft has maintained this convention for backwards compatibility.
• Linux: Varies by tool. ls -lh uses binary (1 KiB = 1,024 bytes) with SI labels. ls --si uses true SI. df -h uses binary, df -H uses SI.
• RAM: Memory modules are genuinely binary. A "16 GB" DIMM contains 16 × 2³⁰ = 17,179,869,184 bytes. RAM is one of the few contexts where the binary interpretation is physically correct.
• Cloud storage pricing: AWS, GCP, and Azure define GiB as 2³⁰ bytes for billing purposes but sometimes use GB in marketing materials.

The Practical Impact for Developers

Storage Provisioning

When you provision a 100 GB EBS volume on AWS, you get 100 GiB (107,374,182,400 bytes), not 100 GB (100,000,000,000 bytes). But when a user uploads a file and you check file.size in JavaScript, you get the exact byte count. Displaying that as "1.5 GB" or "1.5 GiB" depends on your audience: end users expect SI (like their phone shows), while sysadmins expect binary (like their server reports). Use our Byte Calculator to see conversions in both systems simultaneously.

Configuration Files

Many configuration settings accept size values, and the interpretation varies wildly. Nginx uses lowercase m for megabytes (10⁶) and uppercase M for mebibytes (2²⁰). Kubernetes uses M for SI megabytes and Mi for mebibytes. PostgreSQL uses powers of 1,024 for all memory-related settings. Java's -Xmx flag interprets m/M as binary mebibytes. Always check the documentation for the specific tool.

Logging and Monitoring

When your monitoring dashboard shows "disk usage: 45.2 GB," which definition is it using? Prometheus and Grafana default to SI (base 10) for display but collect raw bytes. Datadog uses binary. The difference at the gigabyte level is roughly 7%, which can easily trigger or mask alerts set near capacity boundaries.

How to Convert Between SI and IEC

The key insight is that both systems ultimately measure the same thing: raw bytes. To convert, go through bytes as the common unit:

1. Convert the source to raw bytes (multiply by the appropriate power of 1,000 or 1,024)
2. Convert raw bytes to the target unit (divide by the appropriate power)

For example, to convert 500 GB (SI) to GiB (IEC):

500 GB = 500 × 10^9 = 500,000,000,000 bytes
500,000,000,000 / 2^30 = 500,000,000,000 / 1,073,741,824
                       ≈ 465.66 GiB

This is exactly what your operating system does when it reports a 500 GB drive as having 465 GB (using binary definitions with SI labels). The SI vs IEC tab in our calculator shows this comparison for any value instantly.

Best Practices

• Always specify the unit system in technical documentation. Write "500 GiB" or "500 GB (SI)," not just "500 GB."
• Store raw bytes internally and format for display at the presentation layer.
• Use IEC units (KiB, MiB, GiB) when precision matters, especially in infrastructure, capacity planning, and SLAs.
• Use SI units (KB, MB, GB) for user-facing displays where familiarity is more important than precision.
• Test with values near tier boundaries (e.g., 999 MB vs 1 GB) to catch off-by-one display bugs.

Try It Yourself

Enter any file size into our Byte Calculator and see both SI and IEC conversions side by side. The SI vs IEC tab highlights the exact percentage difference for any value, and the Transfer tab shows how long a file of that size would take to download at various speeds.