inference/research
verifiedhigh · 0.90updated 2026-10-04

CUDA Graphs: Official Documentation

whitepaper

Fields

id
sources/src5-cuda-graphs-docs
Publisher
NVIDIA
Kind
whitepaper
Published
2025-01-01
Accessed
2026-10-04
notes
CUDA Graphs capture and replay of kernel sequences. Reduces CPU launch overhead. Key for inference serving. Used by vLLM, TensorRT-LLM. Supports multi-stream capture, graph update, conditional nodes.

Notes

CUDA Graphs capture and replay of kernel sequences. Reduces CPU launch overhead. Key for inference serving. Used by vLLM, TensorRT-LLM. Supports multi-stream capture, graph update, conditional nodes.

Cross-references

This record references no other records.