AI engineering glossary · Inference serving optimisation
What is warp?
Warp: A group of 32 GPU threads that execute the same instruction together in lockstep. If threads in a warp take different branches, the paths run one after another, so kernels are written to keep warps uniform.
Warp: an example
A kernel launched with 256 threads per block contains 8 such groups.
Related terms
More Inference serving optimisation terms
- Weight-only quantisation
- Zero-point
- Acceptance rate
- Activation quantisation
- All-reduce
- Arithmetic intensity
- Asymmetric quantisation
- Autoscaling
Where you learn it
Warp is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.