Overhead
Overhead is time or work spent supporting a computation rather than performing the particular useful operation being counted. Its meaning depends on the measurement boundary.
How it works
Host submission, synchronization, allocation and runtime compilation can matter around a kernel. Some of that work overlaps device execution; adding every duration together would double-count overlapping intervals.
Work through an example
Launching many tiny kernels can expose repeated submission gaps. Fusion or CUDA graph replay may help, but compare complete application time and preserve the required output computation.
Where the term can mislead
Fixed overhead is not a universal device constant. Measure the actual process, driver, launch path and workload.
Connect the idea
Take it into practice
Continue with the existing concept, apply it to an interview problem, or follow a course lesson. Access requirements are shown before you open a page.
New to GPUs? Start with the free execution-model concept.
Sources and scope
Examples here are teaching calculations, not measured GPU performance. Consult the linked documentation for the exact device, instruction and tool version. Source review: 25 October 2026.