Kilobyte per second to Terabyte per second

A typical 4 MB MP3 file at 7 kB/s took about 9–10 minutes to download. Napster users in 1999 would queue up songs before bed and hope the phone line stayed connected overnight. A single disconnection meant starting over from scratch.

Lowercase "k" with uppercase "B" (kB/s) means 1,000 bytes per second (SI decimal). Uppercase "K" with uppercase "B" (KB/s) traditionally meant 1,024 bytes per second (binary). In practice, most software uses them interchangeably, and the difference is only 2.4%.

Apps display kB/s when transfer speeds are genuinely that slow — downloading over congested mobile networks, tethering in rural areas, or transferring tiny files where the connection never ramps up. It is also common in SSH/SCP transfers that display instantaneous speed during small file copies.

The first consumer ADSL plans offered 256 kbps downstream, delivering about 32 kB/s — roughly 4.5× faster than a 56k modem. A 512 kbps plan gave 64 kB/s. That first jump from 7 to 32 kB/s felt revolutionary, cutting a 10-minute download to about 2 minutes.

An SMS is limited to 140 bytes (160 characters in GSM-7 encoding), and the signalling channel transmits it almost instantly. But if you think of SMS throughput over a sustained period, the practical rate is about 0.1–0.5 kB/s because of the overhead between messages.

Large language models have billions of parameters that must be read from memory for every inference pass. An LLM with 70 billion parameters at 16-bit precision needs 140 GB of data read per forward pass. At 3 TB/s, the H100 can perform roughly 20 inference passes per second — bandwidth directly determines tokens-per-second output.

During LLM inference each token requires reading all model weights from memory. A 70-billion-parameter model at 16-bit precision means 140 GB read per forward pass. At 30 tokens per second, that is 4.2 TB/s of memory reads — right at the limit of an H100's HBM3. This is why AI inference is "memory-bound": the GPU's compute cores sit idle waiting for data. Quantising weights to 8-bit or 4-bit halves or quarters the bandwidth demand, directly increasing tokens per second.

The NVIDIA B200 GPU with HBM3e achieves approximately 8 TB/s of memory bandwidth as of 2025. Each generation roughly doubles bandwidth — from 2 TB/s (A100) to 3.35 TB/s (H100) to 4.8 TB/s (H200) to 8 TB/s (B200). The trajectory suggests 16+ TB/s within a few years.

About 16.7 minutes. A petabyte is 1,000 terabytes, so at 1 TB/s, the math is simple division. For context, the Library of Congress contains roughly 10–20 petabytes of data. Transferring it all at 1 TB/s would take about 3–6 hours.

Yes — petabytes per second (PB/s). Experimental optical interconnects and photonic computing architectures are pushing toward PB/s-class bandwidth. Some supercomputer storage systems already aggregate into the PB/s range when all nodes operate simultaneously. It is the next frontier for AI training clusters.