Search for: All records

GPU power consumption has become a central systems challenge as modern AI datacenters increasingly rely on accelerators whose power footprints reach unprecedented scales. A single NVIDIA B200 already draws around 1000 W, making large multi-GPU nodes among the most power-hungry computing platforms ever deployed. Despite this, GPUs still expose only a single, device-wide DVFS domain. This coarse control is increasingly mismatched to modern AI inference, where power demand is highly non-uniform across models, phases, and workloads. LLMs alternate between compute-bound prefill and memory-bound decode phases, models exhibit widely varying kernel behavior, and emerging agentic pipelines chain together models with sharply different computational profiles. As GPUs shift toward multi-die designs and multi-model stacking becomes essential for utilization, a single global frequency increasingly leads to unnecessary and wasteful high-power operation. We introduce PowerWeave, the first spatial DVFS mechanism for GPUs, built around an OS-level power-management control plane. PowerWeave dynamically learns per-stream frequencyscaling behavior from kernel latencies and uses a global governor to react to request-rate changes, tail-latency behavior, and SLO slack. This design enables adaptive, fine-grained, kernel-aware power management that preserves SLOs while improving energy efficiency across diverse ML workloads. We build PowerWeave in Rust as a fully transparent powergovernor layer atop the GPU driver and evaluate it across diverse LLM scenarios and agentic workloads on B200 GPUs. PowerWeave reduces energy consumption by 28% on average in both disaggregated-prefill and multitenant settings, achieving up to an 8× improvement over device-wide DVFS in disaggregated prefill. For agentic workflows, PowerWeave reduces energy consumption by 19% on average without compromising throughput. PowerWeave eliminates thermal throttling across all evaluated workloads while maintaining zero SLO violations. Finally, our hardware analysis shows that spatial DVFS is practical: even perSM clock domains add less than 0.5% GPU die area overhead. 

In “Post Reinforcement Learning Inference,” Vasilis Syrgkanis and Ruohan Zhan develop a new inferential framework for data collected via reinforcement learning (RL) algorithms, the adaptive systems that update strategies as outcomes unfold. Traditional statistical methods fail in this setting because adaptivity induces time-varying variance and dependence across samples. Syrgkanis and Zhan propose an adaptively weighted generalized method of moments (AW-GMM) estimator that stabilizes this variance through data-dependent weights. They prove that the weighted estimator achieves consistency and asymptotic normality, enabling valid hypothesis testing and confidence intervals for policy values and dynamic treatment effects. Their method provides a unified approach for structural estimation and inference under nonstationary, adaptively generated sequence data, with applications to dynamic off-policy evaluation and personalized decision systems. 

We develop a general theory of omitted variable bias for a wide range of common causal parameters, including average treatment effects, average causal derivatives, and policy effects from covariate shifts. We show how plausibility judgments on the maximum explanatory power of omitted variables are sufficient to bound the bias, facilitating sensitivity analysis in otherwise complex models. Finally, we provide statistical inference methods that can leverage modern machine learning algorithms for estimation. These results allow empirical researchers to perform sensitivity analyses in a flexible class of machine-learned causal models using very simple tools. Empirical examples demonstrate the utility of our approach. 

We study the problem of designing procurement auctions for the strategic uncapacitated facility location problem: a company needs to procure a set of facility locations in order to serve its customers and each facility location is owned by a strategic agent. Each owner has a private cost for providing access to their facility (e.g., renting it or selling it to the company) and needs to be compensated accordingly. The goal is to design truthful auctions that decide which facilities the company should procure and how much to pay the corresponding owners, aiming to minimize the total cost, i.e., the monetary cost paid to the owners and the connection cost suffered by the customers (their distance to the nearest facility). We evaluate the performance of these auctions using the frugality ratio. We first analyze the performance of the classic VCG auction in this context and prove that its frugality ratio is exactly 3. We then leverage the learning-augmented framework and design auctions that are augmented with predictions regarding the owners’ private costs. Specifically, we propose a family of learning-augmented auctions that achieve significant payment reductions when the predictions are accurate, leading to much better frugality ratios. At the same time, we demonstrate that these auctions remain robust even if the predictions are arbitrarily inaccurate, and maintain reasonable frugality ratios even under adversarially chosen predictions. We finally provide a family of “error-tolerant” auctions that maintain improved frugality ratios even if the predictions are only approximately accurate, and we provide upper bounds on their frugality ratio as a function of the prediction error. 

Abstract In a variety of applications, including nonparametric instrumental variable (NPIV) analysis, proximal causal inference under unmeasured confounding, and analysis of missing-not-at-random data with shadow variables, we are interested in inference on a continuous linear functional (e.g. average causal effects) of nuisance functions (e.g. NPIV regression) defined by conditional moment restrictions. These nuisance functions are often weakly identified, meaning the moment restrictions are ill-posed and may admit multiple solutions. This paper proposes a novel condition for the functional to be strongly identified (amenable to n) rate asymptotically normal estimation) even when the nuisance function remains weakly identified. The condition implies the existence of debiasing nuisance functions. We propose penalized minimax estimators for both the primary and debiasing nuisance functions. These estimators accommodate flexible function classes and, crucially, converge to fixed limits determined by the penalization, irrespective of the nuisances' identifiability. We use these penalized estimators to construct a debiased functional estimator and prove its asymptotic normality under general high-level conditions, leading to valid confidence intervals. Our method is illustrated in partially linear proximal causal inference and instrumental variable regression problems.