Eun Kyung Lee

Pol G. Recasens, Ferran Agullo, Yue Zhu et al.

Large language models have been widely adopted across different tasks, but their auto-regressive generation nature often leads to inefficient resource utilization during inference. While batching is commonly used to increase throughput, performance gains plateau beyond a certain batch size, especially with smaller models, a phenomenon that existing literature typically explains as a shift to the compute-bound regime. In this paper, through an in-depth GPU-level analysis, we reveal that large-batch inference remains memory-bound, with most GPU compute capabilities underutilized due to DRAM bandwidth saturation as the primary bottleneck. To address this, we propose a Batching Configuration Advisor (BCA) that optimizes memory allocation, reducing GPU memory requirements with minimal impact on throughput. The freed memory and underutilized GPU compute capabilities can then be leveraged by concurrent workloads. Specifically, we use model replication to improve serving throughput and GPU utilization. Our findings challenge conventional assumptions about LLM inference, offering new insights and practical strategies for improving resource utilization, particularly for smaller language models. The code is publicly available at https://github.com/FerranAgulloLopez/vLLMBatchingMemoryGap.

Kyungmi Lee, Zhiye Song, Eun Kyung Lee et al.

As AI workloads drive increases in datacenter power consumption, accurate GPU power estimation is critical for proactive power management. However, existing power models face a scalability bottleneck not in the modeling techniques themselves, but in obtaining the hardware utilization inputs they require. Conventional approaches rely on either costly simulation or hardware profiling, which makes them impractical when rapid predictions are required. This work presents EnergAIzer, which addresses this scalability bottleneck by developing a lightweight solution to predict utilization inputs, reducing the estimation walltime from hours to seconds. Our key insight is that kernels in AI workloads commonly employ optimizations that create structured patterns, which analytically determine memory traffic and execution timeline. We construct a performance model using these patterns as an analytical scaffold for empirical data fitting, which also naturally exposes module-level utilization. This predicted utilization is then fed into our power model to estimate dynamic power consumption. EnergAIzer achieves 8% power errors on NVIDIA Ampere GPUs, competitive with traditional power models with elaborate cycle-level simulation or hardware profiling. We demonstrate EnergAIzer's exploration capabilities for frequency scaling and architectural configurations, including forecasting the power of NVIDIA H100 with just 7% error. In summary, EnergAIzer provides fast and accurate power prediction for AI workloads, paving the way for power-aware design explorations.

Zhiye Song, Kyungmi Lee, Eun Kyung Lee et al.

We present EnergyLens, an end-to-end framework for energy-aware large language model (LLM) inference optimization. As LLMs scale, predicting and reducing their energy footprint has become critical for sustainability and datacenter operations, yet existing approaches either require production-level code and expensive profiling or fail to accurately capture multi-GPU energy behavior. As a result, practitioners lack tools for deciding which optimizations to prioritize and for selecting among existing deployment configurations when exhaustive profiling is impractical. EnergyLens addresses this gap with an intuitive einsum-based interface that captures LLM specifications including fusion, parallelism, and compute-communication overlap, combined with load-imbalance-aware MoE modeling and an empirically driven communication energy model for multi-GPU settings. We validate EnergyLens on Llama3 and Qwen3-MoE across tensor-parallel and expert-parallel configurations, achieving mean absolute percentage errors (MAPEs) between 9.25% and 13.19% for multi-GPU prefill and decode energy, and 12.97% across SM allocations for Megatron-style overlap. Our energy-driven exploration reveals up to 1.47x and 52.9x energy variation across configurations in prefill and decode efficiency and motivates distributed serving. We further show that compute-communication overlap is difficult to optimize with intuition alone, but EnergyLens correctly identifies Pareto-optimal overlap configurations.

Pol G. Recasens, Yue Zhu, Chen Wang et al.

Large language models (LLMs) have revolutionized the state-of-the-art of many different natural language processing tasks. Although serving LLMs is computationally and memory demanding, the rise of Small Language Models (SLMs) offers new opportunities for resource-constrained users, who now are able to serve small models with cutting-edge performance. In this paper, we present a set of experiments designed to benchmark SLM inference at performance and energy levels. Our analysis provides a new perspective in serving, highlighting that the small memory footprint of SLMs allows for reaching the Pareto-optimal throughput within the resource capacity of a single accelerator. In this regard, we present an initial set of findings demonstrating how model replication can effectively improve resource utilization for serving SLMs.

Yue Zhu, Hao Yu, Chen Wang et al.

The increasing adoption of large language models (LLMs) with extended context windows necessitates efficient Key-Value Cache (KVC) management to optimize inference performance. Inference workloads like Retrieval-Augmented Generation (RAG) and agents exhibit high cache reusability, making efficient caching critical to reducing redundancy and improving speed. We analyze real-world KVC access patterns using publicly available traces and evaluate commercial key-value stores like Redis and state-of-the-art RDMA-based systems (CHIME [1] and Sherman [2]) for KVC metadata management. Our work demonstrates the lack of tailored storage solution for KVC prefilling, underscores the need for an efficient distributed caching system with optimized metadata management for LLM workloads, and provides insights into designing improved KVC management systems for scalable, low-latency inference.

Hongpeng Guo, Haotian Gu, Xiaoyang Wang et al.

Federated learning (FL) is a machine learning paradigm that allows multiple clients to collaboratively train a shared model while keeping their data on-premise. However, the straggler issue, due to slow clients, often hinders the efficiency and scalability of FL. This paper presents FedCore, an algorithm that innovatively tackles the straggler problem via the decentralized selection of coresets, representative subsets of a dataset. Contrary to existing centralized coreset methods, FedCore creates coresets directly on each client in a distributed manner, ensuring privacy preservation in FL. FedCore translates the coreset optimization problem into a more tractable k-medoids clustering problem and operates distributedly on each client. Theoretical analysis confirms FedCore's convergence, and practical evaluations demonstrate an 8x reduction in FL training time, without compromising model accuracy. Our extensive evaluations also show that FedCore generalizes well to existing FL frameworks.

Sunyanan Choochotkaew, Chen Wang, Huamin Chen et al.

Estimating power consumption in modern Cloud environments is essential for carbon quantification toward green computing. Specifically, it is important to properly account for the power consumed by each of the running applications, which are packaged as containers. This paper examines multiple challenges associated with this goal. The first challenge is that multiple customers are sharing the same hardware platform (multi-tenancy), where information on the physical servers is mostly obscured. The second challenge is the overhead in power consumption that the Cloud platform control plane induces. This paper addresses these challenges and introduces a novel pipeline framework for power model training. This allows versatile power consumption approximation of individual containers on the basis of available performance counters and other metrics. The proposed model utilizes machine learning techniques to predict the power consumed by the control plane and associated processes, and uses it for isolating the power consumed by the user containers, from the server power consumption. To determine how well the prediction results in an isolation, we introduce a metric termed isolation goodness. Applying the proposed power model does not require online power measurements, nor does it need information on the physical servers, configuration, or information on other tenants sharing the same machine. The results of cross-workload, cross-platform experiments demonstrated the higher accuracy of the proposed model when predicting power consumption of unseen containers on unknown platforms, including on virtual machines.