{"type":"analysis","id":"analysis:research-orbax-can-checkpoint-as-fast-as-storage-finishes-maxtext-ships-that-as-an-opt","slug":"research-orbax-can-checkpoint-as-fast-as-storage-finishes-maxtext-ships-that-as-an-opt","title":"Orbax can checkpoint as fast as storage finishes; MaxText ships that as an opt-in, not a measured goodput default","description":"Continuous checkpointing is a real Orbax save policy, wired into MaxText behind enable_continuous_checkpointing. Google’s March 2026 blog claims smaller P50 intervals and MTBF-based resource savings, but the published HTML does not give extractable numeric goodput, and no independent measurement was found.","published_at":"2026-11-08T02:00:00.000Z","updated_at":"2026-11-08T02:00:00.000Z","record_date":"2026-11-08","topics":["google","research"],"source_urls":["https://orbax.readthedocs.io/en/latest/api_reference/checkpoint.v1.training.save_decision_policies.html","https://raw.githubusercontent.com/google/orbax/main/checkpoint/CHANGELOG.md","https://raw.githubusercontent.com/AI-Hypercomputer/maxtext/main/src/maxtext/common/checkpoint_context.py","https://raw.githubusercontent.com/AI-Hypercomputer/maxtext/main/src/maxtext/configs/base.yml","https://developers.googleblog.com/en/boost-training-goodput-how-continuous-checkpointing-optimizes-reliability-in-orbax-and-maxtext/","https://orbax.readthedocs.io/en/latest/guides/checkpoint/v1/async_checkpointing.html","https://cloud.google.com/blog/products/ai-machine-learning/goodput-metric-as-measure-of-ml-productivity","https://developers.googleblog.com/en/we-terminated-a-tpu-mid-training-and-it-recovered-in-seconds-introduction-to-elastic-training-with-maxtext/","https://maxtext.readthedocs.io/en/latest/guides/checkpointing_solutions/gcs_checkpointing.html","https://raw.githubusercontent.com/AI-Hypercomputer/maxtext/main/docs/reference/core_concepts/checkpoints.md","https://arxiv.org/html/2605.23066v2","https://arxiv.org/abs/2605.23066","https://orbax.readthedocs.io/en/latest/guides/checkpoint/v1/maximizing_performance.html","https://github.com/google/orbax/issues/1624","https://github.com/AI-Hypercomputer/maxtext/pull/4381"],"language":"en","routes":{"html":"https://newruntime.com/analysis/research-orbax-can-checkpoint-as-fast-as-storage-finishes-maxtext-ships-that-as-an-opt/","markdown":"https://newruntime.com/analysis/research-orbax-can-checkpoint-as-fast-as-storage-finishes-maxtext-ships-that-as-an-opt.md","json":"https://newruntime.com/analysis/research-orbax-can-checkpoint-as-fast-as-storage-finishes-maxtext-ships-that-as-an-opt.json"},"author":{"role":"author","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"editorial_provenance":{"desk":"authorial","target_ref":"deep_research:5a33e6a7-8f90-47c8-b992-6c469d451430","authorial_article_id":"5a33e6a7-8f90-47c8-b992-6c469d451430"},"next_reads":[{"type":"topic","path":"/topics/google/","url":"https://newruntime.com/topics/google/","title":"Google","reason":"Explore the google topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/google.md","json_url":"https://newruntime.com/topics/google.json"},{"type":"collection","path":"/analysis/","url":"https://newruntime.com/analysis/","title":"Analysis","reason":"Open the latest authorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/analysis.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}