# S3: Increasing GPU Utilization during Generative Inference for Higher Throughput

arXiv: https://arxiv.org/abs/2306.06000 · HTML: https://mlsanity.com/paper/2306.06000/

## Key facts

- Cited by 81 papers in the map (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 28 of its 48 references that are in the map.
- First posted on arXiv 9 Jun 2023.

## Placed in

- Efficient LLM inference and serving / [KV cache optimization](https://mlsanity.com/inference/kv-cache-optimization/)

## Builds on (most cited first, 28 of 28)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [Longformer: The Long-Document Transformer](https://mlsanity.com/paper/2004.05150/) (2020)
- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [Big Bird: Transformers for Longer Sequences](https://mlsanity.com/paper/2007.14062/) (2020)
- [Rethinking Attention with Performers](https://mlsanity.com/paper/2009.14794/) (2020)
- [Compressive Transformers for Long-Range Sequence Modelling](https://mlsanity.com/paper/1911.05507/) (2019)
- [Fast Transformer Decoding: One Write-Head is All You Need](https://mlsanity.com/paper/1911.02150/) (2019)
- [GPT-NeoX-20B: An Open-Source Autoregressive Language Model](https://mlsanity.com/paper/2204.06745/) (2022)
- [FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU](https://mlsanity.com/paper/2303.06865/) (2023)
- [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022)
- [DeepSpeed Inference: Enabling Efficient Inference of Transformer Models at Unprecedented Scale](https://mlsanity.com/paper/2207.00032/) (2022)
- [Post-Training Quantization for Vision Transformer](https://mlsanity.com/paper/2106.14156/) (2021)
- [I-BERT: Integer-only BERT Quantization](https://mlsanity.com/paper/2101.01321/) (2021)
- [SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning](https://mlsanity.com/paper/2012.09852/) (2020)
- [MLPerf Inference Benchmark](https://mlsanity.com/paper/1911.02549/) (2019)
- [Clipper: A Low-Latency Online Prediction Serving System](https://mlsanity.com/paper/1612.03079/) (2016)
- [Can Foundation Models Wrangle Your Data?](https://mlsanity.com/paper/2205.09911/) (2022)
- [Fast Distributed Inference Serving for Large Language Models](https://mlsanity.com/paper/2305.05920/) (2023)
- [Serving DNNs like Clockwork: Performance Predictability from the Bottom Up](https://mlsanity.com/paper/2006.02464/) (2020)
- [TensorFlow-Serving: Flexible, High-Performance ML Serving](https://mlsanity.com/paper/1712.06139/) (2017)
- [TurboTransformers: An Efficient GPU Serving System For Transformer Models](https://mlsanity.com/paper/2010.05680/) (2020)
- [DeepRecSys: A System for Optimizing End-To-End At-scale Neural Recommendation Inference](https://mlsanity.com/paper/2001.02772/) (2020)
- [Sub-Linear Memory: How to Make Performers SLiM](https://mlsanity.com/paper/2012.11346/) (2020)
- [DLHub: Model and Data Serving for Science](https://mlsanity.com/paper/1811.11213/) (2018)
- [MP-Rec: Hardware-Software Co-Design to Enable Multi-Path Recommendation](https://mlsanity.com/paper/2302.10872/) (2023)

## Built on by (most cited first, 30 of 81)

- [KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache](https://mlsanity.com/paper/2402.02750/) (2024)
- [A Survey on Efficient Inference for Large Language Models](https://mlsanity.com/paper/2404.14294/) (2024)
- [Efficient Large Language Models: A Survey](https://mlsanity.com/paper/2312.03863/) (2023)
- [DynamoLLM: Designing LLM Inference Clusters for Performance and Energy Efficiency](https://mlsanity.com/paper/2408.00741/) (2024)
- [Towards Efficient Generative Large Language Model Serving: A Survey from Algorithms to Systems](https://mlsanity.com/paper/2312.15234/) (2023)
- [Efficient LLM Scheduling by Learning to Rank](https://mlsanity.com/paper/2408.15792/) (2024)
- [Shortened LLaMA: Depth Pruning for Large Language Models with Comparison of Retraining Methods](https://mlsanity.com/paper/2402.02834/) (2024)
- [ChunkAttention: Efficient Self-Attention with Prefix-Aware KV Cache and Two-Phase Partition](https://mlsanity.com/paper/2402.15220/) (2024)
- [Efficient Interactive LLM Serving with Proxy Model-based Sequence Length Prediction](https://mlsanity.com/paper/2404.08509/) (2024)
- [MegaAgent: A Large-Scale Autonomous LLM-based Multi-Agent System Without Predefined SOPs](https://mlsanity.com/paper/2408.09955/) (2024)
- [LLM Inference Serving: Survey of Recent Advances and Opportunities](https://mlsanity.com/paper/2407.12391/) (2024)
- [Don't Stop Me Now: Embedding Based Scheduling for LLMs](https://mlsanity.com/paper/2410.01035/) (2024)
- [OmniRouter: Budget and Performance Controllable Multi-LLM Routing](https://mlsanity.com/paper/2502.20576/) (2025)
- [Mind the Memory Gap: Unveiling GPU Bottlenecks in Large-Batch LLM Inference](https://mlsanity.com/paper/2503.08311/) (2025)
- [M'elange: Cost Efficient Large Language Model Serving by Exploiting GPU Heterogeneity](https://mlsanity.com/paper/2404.14527/) (2024)
- [How Far Are We From AGI: Are LLMs All We Need?](https://mlsanity.com/paper/2405.10313/) (2024)
- [A Survey: Collaborative Hardware and Software Design in the Era of Large Language Models](https://mlsanity.com/paper/2410.07265/) (2024)
- [Prefill-Decode Aggregation or Disaggregation? Unifying Both for Goodput-Optimized LLM Serving](https://mlsanity.com/paper/2508.01989/) (2025)
- [SLO-aware GPU Frequency Scaling for Energy Efficient LLM Inference Serving](https://mlsanity.com/paper/2408.05235/) (2024)
- [AsymKV: Enabling 1-Bit Quantization of KV Cache with Layer-Wise Asymmetric Quantization Configurations](https://mlsanity.com/paper/2410.13212/) (2024)
- [Online Scheduling for LLM Inference with KV Cache Constraints](https://mlsanity.com/paper/2502.07115/) (2025)
- [JITServe: SLO-aware LLM Serving with Imprecise Request Information](https://mlsanity.com/paper/2504.20068/) (2025)
- [Taming the Titans: A Survey of Efficient LLM Inference Serving](https://mlsanity.com/paper/2504.19720/) (2025)
- [Slice-Level Scheduling for High Throughput and Load Balanced LLM Serving](https://mlsanity.com/paper/2406.13511/) (2024)
- [MoQAE: Mixed-Precision Quantization for Long-Context LLM Inference via Mixture of Quantization-Aware Experts](https://mlsanity.com/paper/2506.07533/) (2025)
- [TPI-LLM: Serving 70B-scale LLMs Efficiently on Low-resource Edge Devices](https://mlsanity.com/paper/2410.00531/) (2024)
- [Fast Inference for Augmented Large Language Models](https://mlsanity.com/paper/2410.18248/) (2024)
- [HyGen: Efficient LLM Serving via Elastic Online-Offline Request Co-location](https://mlsanity.com/paper/2501.14808/) (2025)
- [Multi-Bin Batching for Increasing LLM Inference Throughput](https://mlsanity.com/paper/2412.04504/) (2024)
- [ELIS: Efficient LLM Iterative Scheduling System with Response Length Predictor](https://mlsanity.com/paper/2505.09142/) (2025)

## Abstract (arXiv)

Generating texts with a large language model (LLM) consumes massive amounts of memory. Apart from the already-large model parameters, the key/value (KV) cache that holds information about previous tokens in a sequence can grow to be even larger than the model itself. This problem is exacerbated in one of the current LLM serving frameworks which reserves the maximum sequence length of memory for the KV cache to guarantee generating a complete sequence as they do not know the output sequence length. This restricts us to use a smaller batch size leading to lower GPU utilization and above all, lower throughput. We argue that designing a system with a priori knowledge of the output sequence can mitigate this problem. To this end, we propose S3, which predicts the output sequence length, schedules generation queries based on the prediction to increase device resource utilization and throughput, and handle mispredictions. Our proposed method achieves 6.49× throughput over those systems that assume the worst case for the output sequence length.