| Tebibits per second (Tibit/s) | Teraoctets per second (To/s) |
|---|---|
| 1 Tebibit per second | 0.137438953472 To/s |
| 2 Tebibits per second | 0.274877906944 To/s |
| 3 Tebibits per second | 0.412316860416 To/s |
| 4 Tebibits per second | 0.549755813888 To/s |
| 5 Tebibits per second | 0.68719476736 To/s |
| 10 Tebibits per second | 1.37438953472 To/s |
| 20 Tebibits per second | 2.74877906944 To/s |
| 25 Tebibits per second | 3.4359738368 To/s |
| 50 Tebibits per second | 6.8719476736 To/s |
| 100 Tebibits per second | 13.7438953472 To/s |
| Reference | Tebibits per second (Tibit/s) | Teraoctets per second (To/s) |
|---|---|---|
| A dial-up modem | 0.0000000509317 Tibit/s | 0.000000007 To/s |
| Typical home broadband | 0.0000909495 Tibit/s | 0.0000125 To/s |
| Gigabit Ethernet | 0.000909495 Tibit/s | 0.000125 To/s |
| Streaming a 4K film | 0.0000227374 Tibit/s | 0.000003125 To/s |
The tebibit per second is a unit of data transfer rate equal to 1,099,511,627,776 bits per second, which is 1,024 gibibits per second. Its symbol is Tibit/s. It is the binary counterpart of the terabit per second, and the two differ by 10 per cent.
Ten per cent is the point at which the distinction becomes a matter of money rather than of pedantry. A supplier quoting a system at a hundred terabits per second and a customer measuring a hundred tebibits per second are not describing the same performance, and the difference is ten terabits — more than most organisations' entire external connectivity.
The rate belongs to the interior of very large machines. The aggregate memory bandwidth of a rack of accelerators, or the internal switching capacity of a large network chip, reaches this range, and both are built from power-of-two structures: memory channels of fixed binary width, switch ports in powers of two, buffers sized in binary. Expressing their totals with binary prefixes preserves the arithmetic that produced them.
In octets a tebibit per second is 137,438,953,472, or 128 gibioctets per second. That is more than any single storage device can supply and more than any external cable carries. It is a figure that describes something happening inside a cabinet, between chips connected by short traces on a board, where the physical distance is measured in centimetres.
Optical research also brushes this range. A laboratory demonstration carrying a petabit per second down one fibre is a thousand times higher, but individual wavelength channels and the electronics driving them work at tebibit-scale aggregates, and papers reporting them often state the binary figure because the underlying frame sizes are binary.
The habit of writing the lowercase i is worth keeping even where the reader is unlikely to check. A number written unambiguously can be converted correctly by anyone who reads it later; one written ambiguously cannot be repaired, and at ten per cent the ambiguity is no longer harmless.
One tebibit per second equals 1,024 gibibits per second, 137,438,953,472 octets per second, or about 1.100 terabits per second.
The teraoctet per second is a unit of data transfer rate equal to a thousand gigaoctets per second, or eight terabits per second. Its symbol is To/s. It describes memory bandwidth inside the fastest processors and the aggregate throughput of large computing systems, rather than any link between separate machines.
Graphics and accelerator chips are the clearest example. A modern accelerator uses stacked memory placed alongside the processor on the same package, and the bandwidth between them reaches several teraoctets per second. That figure is what allows thousands of arithmetic units to be kept busy at once, and it is now the property that most often determines how fast a machine learning workload runs.
The reason bandwidth rather than arithmetic has become the limiting factor is worth stating. Processors have grown far faster at computing than memory has at supplying data, so a modern chip can perform tens of operations for every octet it reads. Any calculation that touches memory more often than that is limited by the memory, and most real calculations are.
Supercomputer interconnects reach this range in aggregate. The network joining thousands of nodes carries teraoctets per second across the whole machine, though no single link does. The design problem is to arrange the topology so that any node can reach any other quickly, which is why these networks are built as multi-dimensional meshes and trees rather than as simple stars.
To make the number concrete, one teraoctet per second would fill a large consumer hard drive in twenty seconds and transfer the entire text of every book in a national library within a minute. Nothing in ordinary use approaches it, and no external cable of any kind carries it.
The unit also appears in descriptions of storage arrays. A large parallel filesystem, spread across thousands of drives, can deliver a few teraoctets per second in aggregate to a supercomputer, and that figure is quoted as a headline specification because it determines how quickly a simulation can be written out and read back.
One teraoctet per second equals 1,000 gigaoctets per second, 8 terabits per second, or about 0.9095 tebioctets per second.