# Minibatch persistency, eight years later: what batch reuse costs in steps and joules, and what it saves in data

arXiv: https://arxiv.org/abs/2609.13922 · HTML: https://mlsanity.com/paper/2609.13922/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 14 of its 17 references that are in the map.
- First posted on arXiv 12 Sep 2026.

## Builds on (most cited first, 14 of 14)

- [The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale](https://mlsanity.com/paper/2406.17557/) (2024)
- [Scaling Data-Constrained Language Models](https://mlsanity.com/paper/2305.16264/) (2023)
- [Measuring the Effects of Data Parallelism on Neural Network Training](https://mlsanity.com/paper/1811.03600/) (2018)
- [Zeus: Understanding and Optimizing GPU Energy Consumption of DNN Training](https://mlsanity.com/paper/2208.06102/) (2022)
- [Benchmarking Neural Network Training Algorithms](https://mlsanity.com/paper/2306.07179/) (2023)
- [The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents](https://mlsanity.com/paper/2402.03220/) (2024)
- [No Train No Gain: Revisiting Efficient Training Algorithms For Transformer-based Language Models](https://mlsanity.com/paper/2307.06440/) (2023)
- [Repetita Iuvant: Data Repetition Allows SGD to Learn High-Dimensional Multi-Index Functions](https://mlsanity.com/paper/2405.15459/) (2024)
- [Reducing Energy Bloat in Large Model Training](https://mlsanity.com/paper/2312.06902/) (2023)
- [Faster Neural Network Training with Data Echoing](https://mlsanity.com/paper/1907.05550/) (2019)
- [tf.data service: A Case for Disaggregating ML Input Data Processing](https://mlsanity.com/paper/2210.14826/) (2022)
- [Improved Scaling Laws in Linear Regression via Data Reuse](https://mlsanity.com/paper/2506.08415/) (2025)
- [Stochastic Optimization with Laggard Data Pipelines](https://mlsanity.com/paper/2010.13639/) (2020)
- [Faster SGD training by minibatch persistency](https://mlsanity.com/paper/1806.07353/) (2018)

## Abstract (arXiv)

Minibatch persistency reuses data instead of reading it: rather than drawing a fresh minibatch at every optimizer step, it takes K consecutive steps on the same one. Absorbed into data echoing in 2019, it has carried one objection -- that reuse merely imitates a larger learning rate -- and no baseline tuned as carefully as the method itself. This paper runs the missing test. A pre-registered study trains a 49M-parameter Transformer on FineWeb-Edu at minibatch size B in {32, 128, 512}, 8 seeds per cell, tuning the learning rate separately for every batch size and every arm, against a reuse-free control that changes the sampling and nothing else. Each headline claim is a cost to reach a fixed loss, read on four axes: optimizer steps, fresh tokens, seconds, and joules at the socket. We then replicate on new seeds and a newer GPU generation, and put the three arms on one schedule in steps. The outcome of our study is that what minibatch reuse buys is neither speed nor energy but data, and only at large minibatch size: at B = 32 it reads more fresh tokens than the baseline, not fewer. On steps, seconds and joules it is at best free; and at B = 512, where it looks best, a registered control cannot separate the effect of reuse from the position on the learning-rate schedule at n = 8 seeds. The technique is therefore worth using where fresh data rather than compute is the binding cost: a corpus that runs out, a pipeline that pays per sample, a stream that cannot be rewound. Where the data can simply be read again, spaced epochs do as well or better.