NVIDIA says Snapshot restores replicas with nearly 1 TB checkpoints in under 43 seconds
In NVIDIA's own B200 tests a restore beats a cold start by a wide margin, but the timings leave out container image pulls and need the same GPU model and driver.
Tag TensorRT LLM
NVIDIA measures FLUX 3 Klein image generation over three times faster on one B200
NVIDIA stacks NVFP4, sparse attention and step caching to get there, and tests on data center Blackwell parts with no GeForce numbers. Its image-drift score rises with each speedup.