Event ·Multi-source

NVIDIA ModelExpress将DeepSeek-V4 Pro启动时间降至2分钟内

NVIDIA AI使用ModelExpress(MX)将DeepSeek-V4 Pro的启动时间从8分钟缩短至不到2分钟，靠的是GPU-to-GPU RDMA快速传输权重。MX是NVIDIA Dynamo中的权重分发与缓存管理服务，它复用内核缓存，并通过NIXL让推理worker直接从其他GPU获取更新权重，避免集中广播。这一方法同样能加速推理和强化学习后训练。

Current read

NVIDIA把大模型冷启动快了4倍，DeepSeek-V4 Pro启动只要2分钟，靠ModelExpress和RDMA，推理和训练都能提速。

6 sources58° AI heatLast updated 2026/7/24 19:01:40

Evidence

6 sources

Primary sourceNVIDIA AI

2026/7/24 19:01:40View source

Related source 1arXiv: DeepSeek

2026/7/26 16:25:45View source

Related source 2Simon Willison’s Weblog

2026/7/22 23:01:00View source

Related source 3vLLM

2026/7/23 06:20:46View source

Related source 4LMSYS Org (SGLang)

2026/7/23 14:37:47View source

Related source 5AI Will

2026/7/26 10:00:30View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story