IPv4 vs IPv6 Subnet Arithmetic and Address Conventions

Count remaining bits in the correct family

IPv4 has 32 address bits; IPv6 has 128. Raw count is 2^(width − prefix). IPv4 /24 therefore has 256 addresses, while IPv6 /64 has 2^64 addresses. A /64 is invalid in IPv4 rather than an extremely small IPv4 network.

Apply family-specific conventions

A traditional IPv4 broadcast subnet often reserves the first network label and final broadcast address, producing 254 traditional hosts in a /24. The /31 point-to-point exception and /32 host range need separate treatment. IPv6 has no IPv4-style broadcast address, so subtracting two is not its raw counting rule.

Normalize notation without changing width

IPv4 is commonly written as four decimal octets. IPv6 uses hexadecimal groups and permits zero compression with ::. Compression shortens the text but not the underlying 128-bit value. A calculator should normalize the containing prefix by clearing host bits in either family.

Keep IPv6 counts exact

IPv6 ranges exceed the exact-integer capacity of ordinary floating-point numbers. Use exact integer values for counts and boundaries, and keep them as strings when passing through a serialization layer. A rounded count may look reasonable while losing low-order bits needed for a boundary check.

Separate arithmetic from network design

Range counts do not determine active devices, safe allocation density, firewall scope or routing policy. Documentation prefixes such as 192.0.2.0/24 and 2001:db8::/32 are useful examples, not production allocation suggestions. Check standards and your operational context before configuring a network.

Apply the same prefix logic with different widths

IPv4 192.0.2.130/25 clears seven host bits to network 192.0.2.128 and ends at 192.0.2.255. It has 128 raw addresses. IPv6 2001:db8::5/126 clears two host bits to 2001:db8::4 and ends at ::7, with four raw addresses. A shorter text label does not imply a shorter bit width.

The address family determines valid prefix bounds. IPv4 /33 and IPv6 /129 are invalid. A normalized network should preserve the fixed prefix bits and clear only host bits; keeping the original input allows an audit of which endpoint was examined.

Count child prefixes without host-reservation subtraction

A /24 IPv4 block contains 2^(28 − 24) = sixteen /28 blocks. A /48 IPv6 block contains 2^(64 − 48) = 65,536 /64 blocks. These are counts of prefixes, not device counts. Subtracting two from each child block does not change how many child blocks fit inside the parent.

Arithmetic partitioning assumes alignment. Choosing an arbitrary start inside a block does not create a valid prefix boundary of any desired size. The calculator can derive the containing boundary; an operational allocation still needs route and addressing-policy review.

Examine tiny ranges without a universal “usable” rule

Traditional IPv4 /30 reserves the network and broadcast labels, leaving two usual hosts. RFC 3021 permits both /31 addresses on a point-to-point link. A /32 is one address. IPv6 /127 has two raw addresses and /128 one; their deployment roles should not be inferred by copying IPv4 subtraction.

The label “usable hosts” must therefore declare its convention. An exact total is a binary range property. Reserved roles, local policy and protocol-specific behavior determine practical use. Advertising every raw address as freely deployable would overstate what the arithmetic proves.

Preserve strings at serialization boundaries

A /64 count of 18,446,744,073,709,551,616 is greater than the ordinary safe floating-point integer range. Exact internal arithmetic can still be spoiled if a client converts that output string to a Number. Keep range counts and large integer boundaries as text or an explicit exact-integer type. Prefer documentation address ranges for reproducible examples, and never treat an example prefix as an allocation grant or a configuration instruction.