Warpgroup
A warpgroup is a cooperating group of four consecutive warps, used by particular NVIDIA instructions such as Hopper warpgroup matrix multiply-accumulate.
How it works
The 128 participating threads must follow the instruction family’s collective and synchronization requirements. Asynchronous matrix operations also require correct fencing, commit and wait operations before dependent work consumes their results.
Work through an example
A producer warp can arrange a tile transfer while a group of consumer warps performs matrix work, provided the kernel respects the supported instruction and barrier protocol.
Where the term can mislead
A warpgroup is not a universal replacement for a warp or block. Newer matrix-instruction families can have different participation requirements.
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.