Kareus: Joint Reduction of Dynamic and Static Energy in Large Model Training
Not on the map. Why?
Abstract (arXiv)
The computing demand of AI is growing at an unprecedented rate, but energy supply is not keeping pace. As a result, energy has become an expensive and contended resource that requires explicit management and optimization. Although recent works have made significant progress in large model training optimization, they focus on optimizing either dynamic or static energy consumption. We find that fine-grained kernel scheduling and frequency scaling jointly and interdependently impact both dynamic and static energy consumption. Based on this finding, we design Kareus, a training system that pushes the time-energy tradeoff frontier by optimizing both aspects. Kareus decomposes the intractable joint optimization problem into local, partition-based subproblems. It then uses a multi-pass multi-objective optimization algorithm to find execution schedules that push the time-energy tradeoff frontier. Compared to the state of the art, Kareus reduces training energy by up to 28.3% at the same training time, or reduces training time by up to 27.5% at the same energy consumption.
Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Builds on 30 of its 57 references that are in the map. First posted on arXiv 25 Jan 2026.
Builds on
30 of 57 references.
References · 30
- 2017Attention Is All You Needcited by 60,361
- 2024The Llama 3 Herd of Modelscited by 14,715
- 2025Qwen3 Technical Reportcited by 8,820
- 2016XGBoost: A Scalable Tree Boosting Systemcited by 5,166
- 2018A Tutorial on Bayesian Optimizationcited by 744
- 2021Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LMcited by 632
- 2021Carbon Emissions and Large Neural Network Trainingcited by 567
- 2022Reducing Activation Recomputation in Large Transformer Modelscited by 302
- 2018Learning to Optimize Tensor Programscited by 140
- 2024MegaScale: Scaling Large Language Model Training to More Than 10,000 GPUscited by 112
- 2024DynamoLLM: Designing LLM Inference Clusters for Performance and Energy Efficiencycited by 112
- 2024NanoFlow: Towards Optimal Large Language Model Serving Throughputcited by 90
18 more
- 2022Zeus: Understanding and Optimizing GPU Energy Consumption of DNN Trainingcited by 79
- 2024FLUX: Fast Software-based Communication Overlap On GPUs Through Kernel Fusioncited by 57
- 2021Breaking the Computation and Communication Abstraction Barrier in Distributed Machine Learning Workloadscited by 40
- 2025The ML.ENERGY Benchmark: Toward Automated Inference Energy Measurement and Optimizationcited by 33
- 2025TAPAS: Thermal- and Power-Aware Scheduling for LLM Inference in Cloud Platformscited by 29
- 2025Triton-distributed: Programming Overlapping Kernels on Distributed AI Systems with the Triton Compilercited by 25
- 2025TileLink: Generating Efficient Compute-Communication Overlapping Kernels using Tile-Centric Primitivescited by 24
- 2023Reducing Energy Bloat in Large Model Trainingcited by 22
- 2022Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learningcited by 20
- 2024T3: Transparent Tracking & Triggering for Fine-grained Overlap of Compute & Collectivescited by 16
- 2025TokenWeave: Efficient Compute-Communication Overlap for Distributed LLM Inferencecited by 14
- 2022BaCO: A Fast and Portable Bayesian Compiler Optimization Frameworkcited by 11
- 2024PartIR: Composing SPMD Partitioning Strategies for Machine Learningcited by 10
- 2025MSCCL++: Rethinking GPU Communication Abstractions for AI Inferencecited by 10
- 2024Domino: Eliminating Communication in LLM Training via Generic Tensor Slicing and Overlappingcited by 9
- 2025Efficient Distributed MLLM Training with Cornstarchcited by 4
- 2025TetriServe: Efficiently Serving Mixed DiT Workloadscited by 4
- 2026FlipFlop: A Static Analysis-based Energy Optimization Framework for GPU Kernelscited by 2
Built on by
Citing papers per year: 2026: 3.
3 papers.