NVIDIA says ModelExpress cuts DeepSeek-V4 Pro startup from 8 minutes to under 2
NVIDIAAI · x · 2026-07-25
NVIDIA says it cut DeepSeek-V4 Pro startup time from 8 minutes to under 2 minutes by moving model weights over the fastest path to GPU memory using GPU-to-GPU RDMA.
- The company says the result comes from NVIDIA ModelExpress (MX), a weight distribution and cache management service inside NVIDIA Dynamo.
- MX reuses kernel caches and lets inference workers fetch updated weights directly from other GPUs via NIXL, avoiding centralized broadcasts.
- NVIDIA says the same approach also speeds up inference and RL post-training by keeping weight movement off the critical path.
Related event: NVIDIA Slashes DeepSeek-V4 Pro Startup Time to Under 2 Minutes(4 posts)→
More from Infra
- Anthropic ARR goes vertical after price cuts: tokens up 4x, $100B+ ARR back in play — GavinSBaker · 2026-10-26
- Alex Banks argues power caps GPU supply, real risk is demand cooling first — thealexbanks · 2026-10-26
- Analyst: Ant has the most fungible AI workload of any model provider today — BenBajarin · 2026-10-26
- Anthropic S-1: training infra edge cut per-task cost 40% and doubled revenue growth — BenBajarin · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- US states halt new data centers as 72% of Americans oppose local builds — YvesMulkers · 2026-10-26