Bits vs. bytes
Network links, NICs and ISP plans are rated in bits per second (lowercase b: Mbps, Gbps). File sizes and most download meters use bytes per second (uppercase B: MB/s). One byte is 8 bits:
So 500 Mbps is 62.5 MB/s, and a 1 GbE port can move at most 125 MB/s before overhead.
Decimal vs. binary prefixes
In networking, prefixes are decimal (SI): 1 Mbps = 1,000,000 bit/s and 1 Gbps = 109 bit/s. Binary prefixes (IEC) are powers of 1024: 1 MiB = 1,048,576 bytes. Many operating systems and tools (Windows Explorer, rsync, some browsers) display MiB/s but label it MB/s, which makes the same transfer look about 4.9% slower.
Conversion table
| Link | MB/s | MiB/s | GB per hour |
|---|---|---|---|
| 100 Mbps | 12.5 | 11.9 | 45 |
| 500 Mbps | 62.5 | 59.6 | 225 |
| 1 Gbps | 125 | 119 | 450 |
| 2.5 Gbps | 312.5 | 298 | 1,125 |
| 10 Gbps | 1,250 | 1,192 | 4,500 |
| 25 Gbps | 3,125 | 2,980 | 11,250 |
| 100 Gbps | 12,500 | 11,921 | 45,000 |
These are line-rate values. Ethernet, IP and TCP headers take roughly 5% at a 1500-byte MTU, so a healthy 1 GbE transfer usually shows 110–118 MB/s.
Worked example
An ISP sells 500 Mbps. A browser shows 58 MB/s during a download. 58 × 8 = 464 Mbps, which is 93% of the plan: normal once protocol overhead is accounted for. If the browser actually means MiB/s, it is 58 × 8.389 = 487 Mbps.
Reading tool output correctly
| Tool | Default unit |
|---|---|
iperf3 | Mbits/sec, Gbits/sec (decimal bits); -f M for MBytes |
show interface (Cisco IOS) | bits/sec (5-minute average) |
ip -s link, /proc/net/dev | raw byte counters |
nload, bmon | configurable; check the header |
MikroTik Winbox / /interface monitor-traffic | bps with SI prefixes |
Storage and bus speeds
The same conversion explains common bottlenecks. A SATA III SSD tops out around 550 MB/s (4.4 Gbps), so it cannot saturate a 10 GbE link on its own; a single 7,200 rpm hard disk delivers roughly 150–250 MB/s sequential, which is more than 1 GbE but far below 10 GbE. USB 3.0 (5 Gbps signaling) carries about 400 MB/s of real data after encoding and protocol overhead.
When comparing measurements, always convert to the same unit and the same prefix system first; most "the link is slow" tickets turn out to be a bits/bytes or MB/MiB mix-up.