**We got DeepSeek-V4-Pro serving in 20 seconds**

> DeepSeek-V4-Pro on eight H200s cold-starts in nearly half an hour. Inferize brings the same warmed engine back in 20 seconds. Inferize: Highly Optimized, Elastic AI Inference Inferize is building highly optimized, elastic inference for AI workloads. Ridiculously fast, efficient LLM serving that scales with demand. Join the waitlist to be first to know when we launch. Inferize on X (@InferizeAI)

**来源信息**
- **来源**：Hacker News：AI 热帖
- **分类**：ai-models
- **发布时间**：2026-06-26 04:49（北京时间）
- **原文**：[打开原文](https://inferize.ai/blog/restoring-live-multi-gpu-llms-in-seconds)