energy-efficiency

This paper proposes a carbon-aware fine-tuning objective that balances task accuracy and inference CO2 emissions, testing it on Gemma-2 2B, Llama-3.1 8B, and Qwen-2.5 14B across MMLU subjects, and finds model- and task-dependent break-even regions.

MIT Technology Review reports on a wave of startups pursuing post-transformer architectures for LLMs, as the dominant model family faces growing costs, energy use, and context-length limits. Companies like Subquadratic aim to build the next generation of AI.

Jeff Geerling's benchmarks show Intel's Core 5 320 in a Dell XPS 13 matching or beating Apple Silicon on performance per watt, suggesting x86 can compete with ARM on efficiency.

The paper compares five parameter-efficient fine-tuning methods on four small language models for on-device personalization, finding LoRA+ best for energy efficiency and QLoRA best for memory-limited deployment.

This paper empirically studies how prompt wording affects energy consumption for on-device LLMs, showing that keyword choices can significantly impact decoding length and total energy, suggesting prompt engineering as a lightweight energy optimization lever.

The author cleaned his solar panels and observed a 2-5% increase in power output, making cleaning worthwhile but with diminishing returns over time.

IonQ's research shows that quantum fine-tuning using trapped-ion systems can significantly reduce energy consumption compared to classical computing, offering a near-term bridge to quantum utility as AI workloads strain power grids.

The article presents a custom Rust + CUDA attention-transformer engine achieving 0.63 J/token energy efficiency on H100 with bit-exact determinism, surpassing typical models, and plans open-source release after patent.

A Wired article recommending four best home air conditioners for extreme heat, including the Midea U-Shaped unit and Zafro Lullaby Duo Portable AC, highlighting features like quiet operation, smart controls, and energy efficiency.

This paper presents a multi-objective Bayesian optimization approach to automate weight selection in reinforcement learning for energy-aware control, demonstrating superior sample efficiency over grid search on a physical Quanser Aero 2 testbed.

China is deploying evaporative cooling infrastructure using mist nozzles in public spaces across multiple cities, which is more energy-efficient than air conditioning, while Europe relies on traditional air conditioners.

The Johnson Thermoelectric Energy Converter (JTEC) is a solid-state heat engine that converts thermal energy into electricity using hydrogen in an electrochemical cycle, claiming up to 60% efficiency.

The article discusses NSRAM, a new artificial neuron on a silicon chip that aims to dramatically improve energy efficiency in AI by mimicking biological neurons, addressing the high power consumption of GPUs in data centers.

A new report finds that installing a heat pump can boost a home's resale value by 0.6% to 1%, recouping up to 25% of the upfront cost.

Introduces EVLA, a framework that enhances vision-language driving assistants with real-time awareness of electrified powertrain states, enabling energy-optimal and physically grounded decisions.

KernelPro is a closed-loop multi-agent system that uses LLMs and micro-profiling tools to automatically optimize GPU kernel code, achieving geomean speedups of 2.42×/4.69×/5.30× on KernelBench and demonstrating a measured 11.6% energy reduction at matched speed.

Unconventional AI introduces Un-0, the first large-scale generative model built on physics as a compute primitive, using coupled oscillators to generate images with competitive quality while promising dramatically improved energy efficiency.

The paper discusses the small scaling exponents of large language models, arguing that they indicate an unsustainable regime in terms of energy resources. It also examines the 'pedestal effect' and draws analogies with fluid turbulence to comment on data smoothness.

NVIDIA's Rubin generation AI servers achieve 100% liquid cooling with a 45°C coolant temperature, drastically reducing energy and water consumption in data centers.

Researchers propose using sound waves and phi-bits to create neuromorphic devices that better mimic biological neurons, potentially enabling faster, more energy-efficient computing for pattern recognition and data analysis.