NVIDIA ModelExpress Reduces DeepSeek Startup Time to Under 2 Minutes

NVIDIA's ModelExpress technology successfully reduced the startup time for the DeepSeek-V4 Pro model from 8 minutes to under 2 minutes. This was achieved by transferring model weights directly into VRAM via GPU-to-GPU RDMA.

2026-07-25 ~ 2026-07-25 · 2 related posts

• NVIDIA says ModelExpress cuts DeepSeek-V4 Pro startup from 8 minutes to under 2 — NVIDIAAI · 2026-07-25

1 near-duplicate retellings: NVIDIAAI