Intel / article

Real Time Video At Batch Size One

Source: runway.com
Author: Runway editorial sitemap
Date: Published Nov 5, 2026

Why it matters

Shows how a 57B autoregressive video model was cut from eight to four H100s per session for real-time use, by fixing cache reprocessing and fusing all-reduce into FP8 matmul. Applicable to anyone serving large diffusion transformers at low latency.

Terms in this piece · Glossary

transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.

Read the source runway.com