Conversion from Petaoctets per second to Tebioctets per second

Information about the Petaoctet per second (Po/s)

The petaoctet per second is a unit of data transfer rate equal to a thousand teraoctets per second, or eight petabits per second. Its symbol is Po/s. It describes the total internal bandwidth of the largest computing machines rather than any single connection, and it exists as an aggregate rather than as a rate anything can sustain on its own.

The clearest case is the interconnect of an exascale supercomputer. Such a machine has tens of thousands of nodes, each linked to the network at hundreds of gigaoctets per second, and the sum across the whole fabric reaches petaoctets per second. That figure is the reason such machines can run a single calculation spread over the whole system rather than many small independent ones.

Memory bandwidth adds up the same way. A machine with ten thousand accelerators, each with several teraoctets per second of local memory bandwidth, has tens of petaoctets per second in total. Whether that total means anything depends entirely on the calculation: a problem that can be divided so each node works mostly on its own data can use it, and a problem that cannot, cannot.

Data centre networks reach this range too. The aggregate capacity of the switching fabric inside a very large facility, counting every link between every rack, is measured in petaoctets per second. The design goal is that any server can reach any other at close to full speed, which requires far more internal capacity than the facility's external connections.

For scale, a petaoctet per second would transfer the entire contents of a large national archive in a second, or fill every hard drive manufactured in a day within about a minute. No storage system can supply data at this rate, and none can absorb it; the number describes movement inside a machine, between memory and processors, where nothing is being stored at all.

The unit also serves in optical research, where a single fibre carrying a petabit per second is 125 teraoctets per second, and an experimental system aggregating several such fibres approaches the petaoctet. Those figures come from laboratories rather than from anything deployed.

One petaoctet per second equals 1,000 teraoctets per second, 8 petabits per second, or about 0.8882 pebioctets per second.

Information about the Tebioctet per second (Tio/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.