Tag
该论文提出 HRF-Res，一种用谐振放电（HRF）神经元替代传统 LIF 神经元的液态状态机，使脉冲储层具备频率选择性动力学，在十个基准上达到脉冲储层方法的最先进精度，同时比等效非脉冲振荡储层能耗降低多达 19 倍。
The paper introduces Power-of-Two Leaky Integrate-and-Fire (PoT-LIF) neurons that emit signed, power-of-two amplitude spikes to enable event-driven spiking computation in Vision-Language-Action models without restructuring pretrained policies. Integrated into the π0.5 VLA projections, it matches or exceeds the continuous-activation baseline on LIBERO benchmarks while reducing modeled arithmetic energy by about 86.95%.
Researchers and companies are developing brain-inspired neuromorphic semiconductor chips aimed at making data centers smarter and more energy-efficient, reducing the power demands of modern AI and computing workloads.
多家企业正采用 NVIDIA DSX AI Factory 平台，将电网、设施与算力统一规划仿真运行，以突破 AI 数据中心面临的电力瓶颈。GE Vernova、Hitachi Energy、Schneider Electric、Siemens Energy 等公司与 NVIDIA 合作，推动电网感知的电源架构与数字孪生方案。
德国计划到2030年将数据中心规模翻倍、AI算力提升四倍，但其气候承诺与《能源效率法》中要求数据中心回收废热的规定产生冲突，政府最终在行业压力下弱化了该要求，凸显欧洲在AI扩张与减排目标之间的两难。
MIT Technology Review explainer on how brute-force data center scaling is hitting ceilings as classic scaling laws (Moore's, Dennard's, Koomey's, Kryder's, Keck's) slow, and how fundamental physical limits of computing still leave room for progress. Draws on Feynman's 1959 'There's Plenty of Room at the Bottom' lecture to frame the next frontier below today's engineered limits.
ATEST: Adaptive Resource-Aware Test-Time Scaling for LLM Inference 提出了一种资源感知的测试时缩放框架，联合考虑额外采样的预期价值与当前 GPU 成本，在 GPU 负载变化下动态决定采样何时停止，从而在满足能耗预算的同时保持准确率，并显著减少 GPU 时间与能耗。
Neuralwatt 分析一个月云端流量发现，agent harness 如何构建 prompt 决定了大量 GPU 计算量：缓存命中率差异导致未缓存输入在 1.6% 到 11.1% 之间，一次缓存失效大致会让请求能耗翻倍，且这些问题多数可通过避免编辑已发送内容来修复。
Artificial Analysis 的 AA-AgentPerf 基准通过回放真实编码 agent 轨迹测试推理硬件能效，结果显示 NVIDIA Vera Rubin 在几乎所有速度档位上均居前沿首位，领先于 DeepSeek V4 Flash/Pro、Kimi K3 和 GLM-5.2/5.3 等模型的测试配置。
Oxford researchers directly measure the inference energy efficiency of 86 open-weight LLM checkpoints across seven benchmarks and six hardware platforms, finding large energy disparities among similarly capable models and arguing benchmark scores should be reported alongside measured energy consumption.
Microsoft's expansion of its Azure Maia system to multi-gigawatt deployments distinguishes Maia 200 for production workloads from Maia 300 planned for 2027, prompting discussion about how to fairly compare inference efficiency across generations. The piece emphasizes that metrics like energy per accepted result, under explicit workload and quality checks, matter more than raw throughput.
GitHub 和耶鲁气候变化传播项目对 1,039 名 GitHub 用户的联合调查显示，80% 的开发者希望获得编写更节能代码的工具，71% 担忧 AI 系统的环境影响，75% 认为雇主应主动减少环境足迹。报告指出，开发者缺乏发现浪费、量化改进并推动优化的清晰方法。
Chip-Hexa 是一个开源的实验性处理器架构项目，探索用 16 个简单核心通过六边形片上网络互连，以在性能、能耗和实现成本之间取得平衡。项目目前处于早期研究阶段，包含架构规范和 Python 仿真模型，尚未经硅验证。
该论文提出 Spora，联合设计脉冲编码与注意力算子，通过 Unipolar Binary Spiking (UBS) 和 Bipolar Binary Spiking (BBS) 在少时间步下实现更高容量的连续语义表示，在 GLUE 基准上超越 SpikeLM。
MIT professor Christina Delimitrou 使用机器学习优化大型数据中心的资源管理与服务器架构，以提升能效、减少电力浪费，从而缓解数据中心扩张对电网和环境造成的压力。
UCLA researchers unveiled a hybrid optical-neural processor that detects deepfake videos with ~97.79% accuracy while analyzing 15-18 video streams in parallel, using light propagation to replace energy-hungry digital decoding. The system achieves 99.86% sensitivity and can be deepened with passive diffractive layers for higher accuracy at minimal energy cost.
GeoWind2Plan 是一个面向城市低空无人机的几何-风场-规划框架，利用局部几何条件神经算子在约 3 秒内预测任务相关的 3D 城市风场（CFD 需约 8 小时），并据此优化能量高效的 3D 路径与速度剖面，相比无风感知规划可减少 4.5%–12.7% 的能耗。
This paper investigates how shared spatial computation in neural operators can reduce energy consumption for virtual sensing applications, particularly in nuclear energy systems, while maintaining prediction accuracy.
Researchers developed a lab-scale device that captures carbon dioxide efficiently by pumping it across a battery. The technology aims to reduce capture costs significantly, with projections down to $92 per ton at scale.
WattMate is a Mac menu bar app that shows which app is currently draining your battery in watts and estimates the minutes saved if you quit it, then measures the actual change after quitting. It requires Apple silicon and macOS 14+.