// AI news · 2026-11-10 · NVIDIA
NVIDIA Launches Open-Source Snapshot for GPU Workload Restore
NVIDIA introduces Snapshot, an open-source checkpoint and restore system for GPU workloads on Kubernetes.
Published 2026-11-10 23:00 (Beijing time)Source NVIDIAAuto-summarised, fact-checked against source
01Summary
It reduces inference replica startup time from minutes to seconds, achieving 8x to 36x speedups over cold starts. The system supports multi-GPU models with up to 1 TB checkpoints and outperforms cold starts in all tested configurations, with GPU state recovery taking 18–29 seconds on B200 GPUs.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- Reduces GPU workload startup from minutes to seconds, 8x–36x faster than cold start.
- Supports multi-GPU models with up to 1 TB checkpoints using tensor and expert parallelism.
- Checkpoint creation takes 15 seconds to 5.5 minutes; restore time beats cold start in all tested cases.
- Validated on vLLM, SGLang, and TensorRT LLM with real-world inference workloads.
03Source
Start and Restore GPU Workloads in Seconds with NVIDIA Snapshot
NVIDIA · published 2026-11-10 23:00 Beijing time
05Related news
- Same source · NVIDIA Launches Two Open Source Projects for GPU Kubernetes Management2026-11-10 · NVIDIA
- Same source · NVIDIA Launches Alpamayo 2 Edge for Onboard AV Reasoning2026-11-11 · NVIDIA
- Same topic · AMD Launches 6th Gen EPYC Processors for AI and Enterprise Computing2026-11-11 · AMD
- Same topic · Jensen Huang's Spy Allegation Raises AI Security Concerns2026-11-11 · The Verge AI
- Same day · Reddit Launches AI-Powered Product Reviews2026-11-10 · TechCrunch
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All stories of 2026-11-10 · Archive by date · All AI news · Today