AI Infra Interviews logo
Reading performance · free reference

Scoreboard stall

A scoreboard stall occurs when a warp cannot issue its next instruction because a tracked dependency has not completed.

Long scoreboardShort scoreboard

How it works

Nsight Compute associates long-scoreboard waits with L1TEX operations such as local or global memory accesses, and short-scoreboard waits with relevant MIO operations such as shared-memory accesses. Samples commonly point to the waiting consumer instruction; inspect its producers to understand the delay.

One illustrative scheduler opportunity

Four resident warps. Two ready. One selected.

All four warps remain resident. C and D are eligible at this opportunity, and the scheduler selects C. This teaching snapshot does not specify a device’s total scheduler count or issue width. Occupancy alone cannot distinguish these states.

Work through an example

An add consuming a loaded register can collect long-scoreboard samples even though the add itself performs no memory access. Follow that register back to the load and examine access patterns and independent work.

Where the term can mislead

A long-scoreboard percentage does not prove HBM bandwidth saturation. Cache behavior, dependent accesses and insufficient concurrency can produce waits too.

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.

  1. NVIDIA Nsight Compute Profiling Guide ↗