PowerInfer: Fast Large Language Model Serving with a Consumer-grade GPU
Open accessThe DOI matches its Crossref record; 14 of the 14 references it cites were found in a registry or an index.
Abstract
This paper introduces PowerInfer, a high-speed Large Language Model (LLM) inference engine on a personal computer (PC) equipped with a single consumer-grade GPU. The key principle underlying the design of PowerInfer is exploiting the high locality inherent in LLM inference, characterized by a power-law distribution in neuron activation. This distribution indicates that a small subset of neurons, termed hot neurons, are consistently activated across inputs, while the majority, cold neurons, vary based on specific inputs. PowerInfer exploits such an insight to design a GPU-CPU hybrid inference engine: hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons are computed on the CPU, thus significantly reducing GPU memory demands and CPU-GPU data transfers. PowerInfer further integrates adaptive predictors and neuron-aware sparse operators, optimizing the efficiency of neuron activation and computational sparsity. The evaluation shows that PowerInfer significantly outperforms llama.cpp by up to 11.69× while retaining model accuracy across various LLMs (including OPT-175B) on a single NVIDIA RTX 4090 GPU. For the OPT-30B model, PowerInfer achieves performance comparable to that of a high-end server-grade A100 GPU, reaching 82% of its token generation rate on a single consumer-grade RTX 4090 GPU.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Han, S., Mao, H., & Dally, W. J. (2015) · arXiv · doi:10.48550/arxiv.1510.00149 · Cited by 4,662
LLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602
Searching for activation functions
Ramachandran, P., Zoph, B., & Le, Q. V. (2017) · arXiv · doi:10.48550/arxiv.1710.05941 · Cited by 2,335 · cited 1×
“If activated neurons reside in CPU memory, computing them on the CPU is faster than transferring them to the GPU, especially with the small numberof activated neurons and the small batch sizes typical in local deployments.”
(2011) · doi:10.1145/1966445.1966473 · Cited by 1,870
Efficient Memory Management for Large Language Model Serving with PagedAttention
Kwon, W., Li, Z., Zhuang, S., Sheng, Y., Zheng, L., Yu, C. H., Gonzalez, J., Zhang, H., & Stoica, I. (2023) · Proceedings of the 29th Symposium on Operating Systems Principles · doi:10.1145/3600006.3613165 · Cited by 1,817
Efficient Memory Management for Large Language Model Serving with PagedAttention
Kwon, W., Li, Z., Zhuang, S., Sheng, Y., Zheng, L., Yu, C. H., Gonzalez, J. E., Zhang, H., & Stoica, I. (2023) · arXiv · doi:10.48550/arxiv.2309.06180 · Cited by 1,802
WinoGrande: an adversarial winograd schema challenge at scale
Sakaguchi, K., Bras, R. L., Bhagavatula, C., & Choi, Y. (2021) · Communications of the ACM · doi:10.1145/3474381 · Cited by 356
DeepSpeed- Inference: Enabling Efficient Inference of Transformer Models at Unprecedented Scale
Aminabadi, R. Y., Rajbhandari, S., Awan, A. A., Li, C., Li, D., Zheng, E., Ruwase, O., Smith, S., Zhang, M., Rasley, J., et al. (2022) · SC22: International Conference for High Performance Computing, Networking, Storage and Analysis · doi:10.1109/sc41404.2022.00051 · Cited by 297 · cited 1×
“Model offloading, which partitions the model between GPU and CPU at the Transformer layer level [3,14,37].”
GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
Frantar, E., Ashkboos, S., Hoefler, T., & Alistarh, D. (2022) · arXiv · doi:10.48550/arxiv.2210.17323 · Cited by 197
SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot
Frantar, E., & Alistarh, D. (2023) · arXiv · doi:10.48550/arxiv.2301.00774 · Cited by 131
Cited by
A Review on Edge Large Language Models: Design, Execution, and Applications
Zheng, Y., Chen, Y., Qian, B., Shi, X., Shu, Y., & Chen, J. (2025) · ACM Computing Surveys · doi:10.1145/3719664 · Cited by 169
HybridFlow: A Flexible and Efficient RLHF Framework
Sheng, G., Zhang, C., Ye, Z., Wu, X., Zhang, W., Zhang, R., Peng, Y., Lin, H., & Wu, C. (2025) · Proceedings of the Twentieth European Conference on Computer Systems · doi:10.1145/3689031.3696075 · Cited by 74
HybridFlow: A Flexible and Efficient RLHF Framework
Sheng, G., Zhang, C., Ye, Z., Wu, X., Zhang, W., Zhang, R., Peng, Y., Lin, H., & Wu, C. (2024) · arXiv · doi:10.48550/arxiv.2409.19256 · Cited by 74
A Survey: Collaborative Hardware and Software Design in the Era of Large Language Models
Guo, C., Cheng, F., Du, Z., Kiessling, J., Ku, J., Li, S., Li, Z., Ma, M., Molom-Ochir, T., Morris, B., et al. (2025) · IEEE Circuits and Systems Magazine · doi:10.1109/mcas.2024.3476008 · Cited by 32
“However, driven by the growing need for latency-insensitive tasks, some studies [79], [80], [91], [92], [93] explore high-throughput LLM inference using limited resources, such as a single GPU, edge devices, and mobile devices.”
POD-A ttention : Unlocking Full Prefill-Decode Overlap for Faster LLM Inference
Kamath, A. K., Prabhu, R., Mohan, J., Peter, S., Ramjee, R., & Panwar, A. (2025) · Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2 · doi:10.1145/3676641.3715996 · Cited by 26
Xiao, C., Cai, J., Zhao, W., Lin, B., Zeng, G., Zhou, J., Zheng, Z., Han, X., Liu, Z., & Sun, M. (2025) · Nature Machine Intelligence · doi:10.1038/s42256-025-01137-0 · Cited by 24
Minimizing Response Latency in LLM-Based Agent Systems: A Comprehensive Survey
Park, G., Lee, S., & Park, Y. (2026) · IEEE Access · doi:10.1109/access.2026.3664226 · Cited by 4
Metadata
- Authors
- Song, Y., Mi, Z., Xie, H., & Chen, H.
- Venue
- Proceedings of the ACM SIGOPS 30th Symposium on Operating Systems Principles
- Publisher
- ACM
- Published
- 2024-11-04
- Pages
- 590-606
- OpenAlex
- W4404401018
- Cited by
- 133
- References
- 14
- FWCI
- 42.20
- Type
- conference paper
- Language
- en
- Open access
- bronze
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.