# Getting DreamX under the real-time budget on one GPU

> Closing the gap to continuous playout took a resident engine, a compiled path, and two traps avoided.

Source: https://windflow.online/blog/dreamx-real-time-budget · Published 2026-07-19 · Windflow Engineering

A 5B world model generated blocks slower than continuous playout allows.

The path down ran through a resident streaming engine, two-step denoising, compiled DiT and VAE, and FlashAttention-3.

Two traps sat on that path: torch.compile's recompile limit silently reverting a long-lived worker to eager, and an overlap that cannot help when both halves are throughput-bound.