惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
博客园 - Franky
T
Tailwind CSS Blog
Microsoft Azure Blog
Microsoft Azure Blog
The Cloudflare Blog
博客园 - 叶小钗
N
Netflix TechBlog - Medium
罗磊的独立博客
量子位
MyScale Blog
MyScale Blog
A
About on SuperTechFans
Blog — PlanetScale
Blog — PlanetScale
V
Visual Studio Blog
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
GbyAI
GbyAI
B
Blog
腾讯CDC
爱范儿
爱范儿
Recent Announcements
Recent Announcements
有赞技术团队
有赞技术团队
F
Fortinet All Blogs
雷峰网
雷峰网
G
Google Developers Blog
Google DeepMind News
Google DeepMind News

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
8 FPS Real‑Time Video on Consumer GPU
Papers Mache · 2026-06-16 · via DEV Community

Papers Mache

MoVerse delivers 360° walkthrough video at roughly 8 FPS on a single RTX 4090, proving that interactive generative video no longer requires a multi‑GPU cluster. The system expands a narrow‑field input into a full panorama, lifts it onto a persistent 3D Gaussian scaffold, and streams the result through a causal autoregressive renderer (the distilled student model). All of this runs on desktop‑class hardware, not on a laptop or server farm.

Before this contribution, high‑fidelity video synthesis depended on either diffusion pipelines that rendered frames offline with dozens of GPUs or explicit 3D representations that struggled to exceed 1–2 FPS on a single card. Researchers typically traded visual quality for speed, and real‑time roaming was considered out of reach for consumer GPUs. Consequently, AR/VR prototypes had to fall back to pre‑recorded assets or low‑resolution sprites.

In our deployment configuration, this causal renderer reaches 8 FPS end‑to‑end roaming of the scene on a single NVIDIA RTX 4090 GPU. The authors emphasize that the same hardware can handle the full pipeline—from panorama diffusion to scaffold rendering—without bottlenecking the user‑controlled camera path [1].

The key to that throughput is the decoupling of world construction from observation rendering: a gravity‑aligned 360° panorama is first completed with topology‑aware diffusion, then a geometry‑aware residual predictor populates a dense Gaussian scaffold that can be rasterized in a single pass. Because the scaffold is explicit, the downstream autoregressive student only needs to refine already rendered views, keeping the per‑frame latency bounded.

The approach still caps the interactive experience at 8 FPS, well short of the 30 FPS horizon typically required for fluid VR head‑mounted displays. Moreover, the method assumes a single narrow‑field input and a static scene geometry, leaving open how it would cope with dynamic objects or longer trajectories.

If these numbers hold, benchmark suites for real‑time generative video should add a consumer‑GPU track that measures end‑to‑end FPS on a single RTX 4090. Downstream developers can now prototype free‑roaming AR experiences without procuring a GPU cluster, shifting the cost curve from capital‑intensive farms to a single high‑end desktop.

References

  1. MoVerse: Real-Time Video World Modeling with Panoramic Gaussian Scaffold