| Tebioctets per second (Tio/s) | Gibibits per second (Gibit/s) |
|---|---|
| 1 Tebioctet per second | 8192 Gibit/s |
| 2 Tebioctets per second | 16384 Gibit/s |
| 3 Tebioctets per second | 24576 Gibit/s |
| 4 Tebioctets per second | 32768 Gibit/s |
| 5 Tebioctets per second | 40960 Gibit/s |
| 10 Tebioctets per second | 81920 Gibit/s |
| 20 Tebioctets per second | 163840 Gibit/s |
| 25 Tebioctets per second | 204800 Gibit/s |
| 50 Tebioctets per second | 409600 Gibit/s |
| 100 Tebioctets per second | 819200 Gibit/s |
| Reference | Tebioctets per second (Tio/s) | Gibibits per second (Gibit/s) |
|---|---|---|
| A dial-up modem | 0.00000000636646 Tio/s | 0.0000521541 Gibit/s |
| Typical home broadband | 0.0000113687 Tio/s | 0.0931323 Gibit/s |
| Gigabit Ethernet | 0.000113687 Tio/s | 0.931323 Gibit/s |
| Streaming a 4K film | 0.00000284217 Tio/s | 0.0232831 Gibit/s |
The tebioctet per second is a unit of data transfer rate equal to 1,024 gibioctets per second, or two to the fortieth power octets per second. Its symbol is Tio/s. It is the binary counterpart of the teraoctet per second, and the two differ by 10 per cent.
Nothing outside a large machine moves data this quickly. The rate describes the memory bandwidth of an accelerator with stacked memory, the internal fabric of a high-end processor package, or the aggregate throughput of a parallel filesystem spread across thousands of drives. All of these are built from binary structures, and their totals are binary quantities divided by time.
High-performance computing is where the unit is written most often. A supercomputer's storage system is specified by how many tebioctets per second it can deliver to the compute nodes, because that number determines how quickly a simulation can save its state and resume. A machine that computes quickly but writes slowly spends its time waiting.
The ten per cent difference from the decimal unit is significant in that context. A filesystem procured to deliver 10 teraoctets per second and one delivering 10 tebioctets per second differ by a whole teraoctet per second, which in a facility of that size represents a substantial fraction of the hardware budget.
In bits a tebioctet per second is 8 tebibits per second, and in decimal terms about 1.1 teraoctets per second. Expressing the same rate four different ways is routine at this level, because the storage industry, the memory industry, the network industry and the standards bodies each prefer a different one.
For everyday comparison, a tebioctet per second would fill a large consumer hard drive in about twenty seconds. No external interface carries this; the figure describes movement between components inside a single system, where the wires are short and there are very many of them running in parallel.
Graphics processors have brought the rate within reach of a single component. A stack of high-bandwidth memory bonded directly to the processor die delivers well over a tebioctet per second to the chip that uses it, and a card carrying several such stacks passes a few. That bandwidth, rather than raw arithmetic speed, is what limits the training of large models: the arithmetic units sit idle unless the memory can keep them fed. The same reasoning explains why supercomputer designers spend as much effort on the paths between memory and processor as on the processors themselves, and why the rate is quoted in binary units when the memory it describes is addressed in powers of two.
One tebioctet per second equals 1,024 gibioctets per second, 1,099,511,627,776 octets per second, or about 1.100 teraoctets per second.
The gibibit per second is a unit of data transfer rate equal to 1,073,741,824 bits per second. Its symbol is Gibit/s. It is the binary counterpart of the gigabit per second, and the two now differ by 7.4 per cent, which is enough to matter in any engineering specification.
The unit belongs to the inside of a machine rather than to the network. Memory buses, processor interconnects and the links between chips on the same board all move a power-of-two number of bits per clock cycle, so their throughput is naturally expressed with a binary prefix. A bus sixty-four bits wide clocked at a given frequency delivers a rate that is a binary multiple of that frequency.
Networking, by contrast, is decimal all the way down. Gigabit Ethernet carries exactly one thousand million bits per second, not 1,073,741,824, and the symbol rate on the wire is chosen to make that so. Confusing the two overstates a link's capacity by seven per cent, which in a capacity plan is the difference between adequate and insufficient.
In octets a gibibit per second is 134,217,728, or 128 mebioctets per second. That is close to the throughput of a fast mechanical hard drive and well below a modern solid-state drive, so it sits at the point where storage and internal buses meet and where matching their rates becomes a design question.
Benchmark tools are the commonest place to see the unit written correctly. A memory bandwidth test that allocates buffers in powers of two and measures how long they take to traverse naturally reports in gibibits or gibioctets per second, and a well-written tool says so explicitly rather than rounding to the decimal unit.
The reason to insist on the distinction here rather than lower down the scale is arithmetic. At the kibibit the gap was 2.4 per cent and could be ignored; here it is nearly a thirteenth, and it grows by a further 2.4 per cent at every step above. Getting into the habit at this level costs nothing and avoids compounding errors later.
One gibibit per second equals 1,073,741,824 bits per second, 134,217,728 octets per second, or about 1.074 gigabits per second.