{ "schema_version": "0.6.0", "paper_id": "energy-consumption-analysis-large-language-models-cpu-gpu", "page_url": "https://threadslab.org/research-publications/papers/energy-consumption-analysis-large-language-models-cpu-gpu/", "title": "Energy Consumption Analysis of Large Language Models Across CPU and GPU Using Diverse Metric Types", "title_variants": [], "authors": [ "Tong Zhang", "Leila Tahmooresnejad", "Naser Ezzati-Jivan" ], "author_details": [ { "name": "Tong Zhang", "orcid": null, "profile_url": "https://ieeexplore.ieee.org/author/720453590041487" }, { "name": "Leila Tahmooresnejad", "orcid": null, "profile_url": "https://ieeexplore.ieee.org/author/345895937078737" }, { "name": "Naser Ezzati-Jivan", "orcid": "https://orcid.org/0000-0003-1435-6297", "profile_url": "https://ieeexplore.ieee.org/author/37071999400" } ], "publication": { "year": 2025, "venue": "2025 IEEE International Conference on Collaborative Advances in Software and COmputiNg (CASCON)", "type": "conference paper", "publication_date": "2025-11-10", "online_date": null, "print_date": "2025-11-10", "volume": null, "issue": null, "pages": "301-310", "article_number": null, "publisher": "IEEE", "issn": [], "isbn": [], "crossref_type": "proceedings-article" }, "publication_type": "conference paper", "status": "published_with_public_abstract", "canonical_source_url": "https://doi.org/10.1109/CASCON66301.2025.00056", "source_record_id": "energy-consumption-analysis-of-large-language-models-across-cpu-and-gpu-using-diverse-metric-typ-ab9e5bb7ed", "identifiers": { "doi": "10.1109/CASCON66301.2025.00056" }, "abstract": "The paper presents a multi-level analysis of energy consumption for large language model workloads across CPU and GPU settings, combining diverse static and dynamic metrics to build and validate energy-prediction models across NLP tasks.", "abstract_source": "Public abstract/metadata mirror and IEEE bibliographic record, paraphrased; full text not obtained.", "abstract_available": true, "scholar_eligibility": { "eligible": false, "basis": "not-eligible", "note": "The page is a discovery record; it does not claim Google Scholar article-host eligibility." }, "description": "The paper presents a multi-level analysis of energy consumption for large language model workloads across CPU and GPU settings, combining diverse static and dynamic metrics to build and validate energy-prediction models across NLP tasks.", "evidence_level": "full-text-reviewed", "evidence": { "source_basis": "full-text-reviewed", "coverage": "material paper sections", "summary_origin": "AI-assisted catalog editorial summary", "review_status": "catalog-reviewed; paper-author approval pending", "verified_on": "2026-08-26", "sources": [ { "note": "Local PDF reviewed, including hardware, counters, task/model design, regression models, and MAPE results" }, { "note": "IEEE DOI record: https://doi.org/10.1109/CASCON66301.2025.00056" }, { "note": "IEEE Xplore record: https://ieeexplore.ieee.org/document/11344465/" } ] }, "summary": { "core_contribution": "The paper models LLM inference energy separately on CPU and GPU using hardware counters, device metrics, and task/model features, then compares classical and neural regressors across language tasks.", "problem": "Energy demand depends on the model, task, context, hardware, and execution device. Aggregate power alone does not explain these differences or provide a practical predictor for choosing a lower-energy inference configuration.", "method": "The experiments use an Intel Core i7-12700K, an NVIDIA RTX 3060 Ti, 32 GB system memory, and 8 GB GPU memory. perf, psutil, and pynvml collect 11 CPU/system metrics and GPU utilization, memory, temperature, and fan signals at the stated one-second sampling interval; clocks are locked. More than 50 Hugging Face models are tested across sentence similarity, summarization, question answering, and generation groups, with 20 random configurations per model and 10 repetitions. Linear/Ridge/LASSO/SGD, MLP, and random-forest models are evaluated with five- and ten-fold cross-validation and MAPE.", "findings": "The reported examples show higher total GPU energy than CPU energy over a 10-second window in some comparisons, but higher CPU energy per iteration; MiniLM is given as approximately 0.4 J on GPU versus 1.2 ± 0.1 J on CPU. GPT2-XL is among the high-demand cases. CPU instruction/cycle counters and GPU utilization/temperature are useful predictors, while GPU memory becomes more important for generation and long context. The reported cross-validation MAPE reaches 2.58% for a five-fold CPU Ridge model and 6.28% for a five-fold GPU linear model; ten-fold results include 2.16% CPU SGD and 4.59% GPU MLP.", "limitations": "The study is inference-only on one CPU/GPU platform with fixed clocks, one-second sampling, and no systematic DVFS, accelerator, batching, concurrency, or data-transfer analysis. NVML metrics are relatively high-level, and the transfer of the learned models to other hardware, runtimes, and serving policies is not established.", "future_work": "Evaluate heterogeneous accelerators and serving stacks, include batching/concurrency and host-device transfer, model DVFS and thermal effects, measure training and end-to-end service energy, and report leakage-safe cross-hardware generalization with uncertainty intervals." }, "tags": [ "energy-efficiency", "llm-efficiency", "performance-modeling", "machine-learning", "performance-analysis" ], "keywords": [ "LLM energy consumption", "CPU energy", "GPU energy", "green AI", "static metrics", "dynamic metrics", "NLP workloads", "energy prediction", "CASCON 2025" ], "versions": [ { "id": "published-version", "label": "Published version", "relation": "version-of-record", "title": "Energy Consumption Analysis of Large Language Models Across CPU and GPU Using Diverse Metric Types", "url": "https://doi.org/10.1109/CASCON66301.2025.00056", "pdf_url": null, "status": "published", "canonical_for_citation": true }, { "id": "ieee-record-and-public-abstract-metadata", "label": "IEEE record and public abstract metadata", "relation": "source-record", "title": "Energy Consumption Analysis of Large Language Models Across CPU and GPU Using Diverse Metric Types", "url": "https://ieeexplore.ieee.org/document/11344465/", "pdf_url": null, "status": "public_abstract", "canonical_for_citation": false } ], "access": { "status": "published_with_public_abstract", "note": "IEEE/Crossref identity and DOI are verified, and a public abstract-level record is available. The full paper was not obtained; exact instrumentation, hardware, model list, equations, workloads, and quantitative results are therefore not asserted. No PDF is redistributed here.", "license": null }, "resources": { "code": null, "data": null, "slides": null, "demo": null }, "citation_guidance": { "when_to_cite": "Cite this paper when modeling or comparing LLM inference energy on CPU and GPU using counters and device telemetry.", "points": [ "perf/psutil/pynvml features and task/model-aware regressors.", "The multi-task, multi-model evaluation with cross-validation and MAPE.", "The distinction between total-window energy and per-iteration energy.", "The fixed-hardware, inference-only boundary of the measurements." ], "canonical_version_id": "published-version" }, "provenance": { "metadata_verified_on": "2026-08-26", "metadata_source": [ "Local PDF reviewed, including hardware, counters, task/model design, regression models, and MAPE results", "IEEE DOI record: https://doi.org/10.1109/CASCON66301.2025.00056", "IEEE Xplore record: https://ieeexplore.ieee.org/document/11344465/" ], "summary_written_by": "AI-assisted", "summary_verified_by": "full-text-grounded catalog review; author approval pending", "linked_preprint_record": null, "author_order_note": null }, "batch": { "phase": 2, "batch_label": "expanded forty-paper release", "status": "included_in_expanded_catalog", "selected_at": "2026-08-28" } }