Compute bound
A workload is compute-bound when the processor is the limiting factor — you could feed it data faster and it'd keep up.
Think of it like
A fast eater at an all-you-can-eat buffet — never waiting for food.
Example
Matrix multiplication on large matrices is usually compute-bound; the CPU/GPU does useful work the whole time.
How it actually works
Compute-bound is the happy case — your hardware is busy. Memory-bound is the sad case — your hardware is waiting. You're compute-bound if adding more memory bandwidth doesn't help. Identifying which you are is step one of optimization.
For product teams
Compute-bound workloads scale well; memory-bound ones need different strategies (quantization, smaller batches).
For engineers
Profile with tools like nsys or roofline analysis to confirm you're compute-bound before optimizing compute.
Related
- Memory bound — When memory is the bottleneck instead.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome