# On the Computational Inefficiency of Large Batch Sizes for Stochastic Gradient Descent

arXiv: https://arxiv.org/abs/1811.12941 · HTML: https://mlsanity.com/paper/1811.12941/

## Key facts

- Cited by 25 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 30 Nov 2018.

## Built on by (most cited first, 25 of 25)

- [Zeus: Understanding and Optimizing GPU Energy Consumption of DNN Training](https://mlsanity.com/paper/2208.06102/) (2022)
- [On Large-Cohort Training for Federated Learning](https://mlsanity.com/paper/2106.07820/) (2021)
- [How Does Critical Batch Size Scale in Pre-training?](https://mlsanity.com/paper/2410.21676/) (2024)
- [Stochastic Training is Not Necessary for Generalization](https://mlsanity.com/paper/2109.14119/) (2021)
- [Power Lines: Scaling Laws for Weight Decay and Batch Size in LLM Pre-training](https://mlsanity.com/paper/2505.13738/) (2025)
- [Small batch deep reinforcement learning](https://mlsanity.com/paper/2310.03882/) (2023)
- [Batch size-invariance for policy optimization](https://mlsanity.com/paper/2110.00641/) (2021)
- [On Efficient Training of Large-Scale Deep Learning Models: A Literature Review](https://mlsanity.com/paper/2304.03589/) (2023)
- [Reducing Energy Bloat in Large Model Training](https://mlsanity.com/paper/2312.06902/) (2023)
- [Globally Convergent Multilevel Training of Deep Residual Networks](https://mlsanity.com/paper/2107.07572/) (2021)
- [Modeling the Machine Learning Multiverse](https://mlsanity.com/paper/2206.05985/) (2022)
- [Layered gradient accumulation and modular pipeline parallelism: fast and efficient training of large language models](https://mlsanity.com/paper/2106.02679/) (2021)
- [Batch size invariant Adam](https://mlsanity.com/paper/2402.18824/) (2024)
- [ILASR: Privacy-Preserving Incremental Learning for Automatic Speech Recognition at Production Scale](https://mlsanity.com/paper/2207.09078/) (2022)
- [Beware of the Batch Size: Hyperparameter Bias in Evaluating LoRA](https://mlsanity.com/paper/2602.09492/) (2026)
- [Parallel Split Learning with Global Sampling](https://mlsanity.com/paper/2407.15738/) (2024)
- [Shift-Curvature, SGD, and Generalization](https://mlsanity.com/paper/2108.09507/) (2021)
- [Scaling the Wild: Decentralizing Hogwild!-style Shared-memory SGD](https://mlsanity.com/paper/2203.06638/) (2022)
- [Towards joint scaling laws with optimal batch size schedules](https://mlsanity.com/paper/2607.27731/) (2026)
- [Non-Asymptotic Analysis of Online Multiplicative Stochastic Gradient Descent](https://mlsanity.com/paper/2112.07110/) (2021)
- [Learning Deep Optimal Embeddings with Sinkhorn Divergences](https://mlsanity.com/paper/2209.06469/) (2022)
- [Improving Neural Network Training using Dynamic Learning Rate Schedule for PINNs and Image Classification](https://mlsanity.com/paper/2507.21749/) (2025)
- [On Using Large-Batches in Federated Learning](https://mlsanity.com/paper/2509.10537/) (2025)
- [Tula: Optimizing Time, Cost, and Generalization in Distributed Large-Batch Training](https://mlsanity.com/paper/2603.18112/) (2026)
- [Perfect Parallelization in Mini-Batch SGD with Classical Momentum Acceleration](https://mlsanity.com/paper/2605.18609/) (2026)

## Abstract (arXiv)

Increasing the mini-batch size for stochastic gradient descent offers significant opportunities to reduce wall-clock training time, but there are a variety of theoretical and systems challenges that impede the widespread success of this technique. We investigate these issues, with an emphasis on time to convergence and total computational cost, through an extensive empirical analysis of network training across several architectures and problem domains, including image classification, image segmentation, and language modeling. Although it is common practice to increase the batch size in order to fully exploit available computational resources, we find a substantially more nuanced picture. Our main finding is that across a wide range of network architectures and problem domains, increasing the batch size beyond a certain point yields no decrease in wall-clock time to convergence for either train or test loss. This batch size is usually substantially below the capacity of current systems. We show that popular training strategies for large batch size optimization begin to fail before we can populate all available compute resources, and we show that the point at which these methods break down depends more on attributes like model architecture and data complexity than it does directly on the size of the dataset.