Grant Wilkins

Dillon Jensen, Obi Nnorom, Grant Wilkins et al.

Large-scale AI model training workloads use thousands of GPUs operating in tightly synchronized loops. During synchronous communication, start-up, shut-down, and checkpointing, GPU power consumption can swing from peak to idle within milliseconds. These large and rapid load swings endanger grid infrastructure as they induce steep power ramp rates, voltage and frequency shifts, and reactive power transients that can damage transformers, converters, and protection equipment. To solve this problem, we introduce EasyRider, a power architecture to mitigate power fluctuations at the rack level. EasyRider uses passive components and actively-controlled auxiliary energy storage to attenuate rack power swings. A software system continually monitors the energy storage system to maximize its lifetime in the presence of frequent charge/discharge cycles. EasyRider filters rack power variations to be within grid safety requirements without requiring software modifications to AI training frameworks or wasting energy. We evaluate EasyRider on a 400VDC-rated prototype system against published workload traces and our own GPU testbed, demonstrating its effectiveness across heterogeneous power levels and workload power profiles.

Grant Wilkins, Fiodar Kazhamiaka, Ram Rajagopal

Datacenter operators and electrical utilities rely on power traces at different spatiotemporal scales. Operators use fine-grained traces for provisioning, facility management, and scheduling, while utilities use site-level load profiles for capacity and interconnection planning. Existing datacenter power models do not capture LLM inference workloads, in which GPUs shift rapidly among compute-intensive prefill, lower-power decode, and idle states, and facility demand depends on how these states evolve and synchronize across many devices. We show that LLM inference power can be represented compositionally through two components: workload-driven transitions among operating states and configuration-specific power distributions within those states. Building on this observation, we develop a trace-generation framework that learns from measured traces and synthesizes power profiles for new traffic conditions and serving configurations. These traces aggregate from GPU servers to rack-, row-, and facility-scale load profiles at the temporal granularity required by the study. Across multiple LLMs, tensor-parallel settings, and GPU generations, our framework achieves median absolute energy error below 5% for most configurations while preserving temporal autocorrelation structure. The resulting traces support downstream analyses including oversubscription, power modulation, and utility-facing load characterization, enabling infrastructure evaluations that flat nameplate assumptions and static trace replay cannot support.

Grant Wilkins, Fiodar Kazhamiaka, Alok Gautam Kumbhare et al.

Demand for AI accelerators is rapidly increasing rack power density, with projections approaching 1MW per deployment by 2027. This poses a major challenge for datacenter power delivery designers. As power densities increase, a datacenter designed for a different target density may strand power, i.e., may be unable to use all the power that its delivery hierarchy has provisioned. Designs must remain efficient over long datacenter lifetimes and multiple hardware generations. Power utilization is particularly important as grid power capacity is a scarce resource in the AI era. Designing an efficient power delivery hierarchy for the long run is difficult because rack placement feasibility, workload impact, and cost depend jointly on electrical topology, deployment granularity, placement policy, power oversubscription, and workload mix. Moreover, each of these factors evolve over time, have inter-dependencies across multiple resource dimensions, and generally do not lend themselves to closed-form analysis. To address this challenge, we develop a framework for evaluating datacenter power delivery designs using throughput, power, and cost metrics over realistic arrival, oversubscription, and decommissioning sequences. The framework combines projection models for GPU, compute, and storage deployments with operational factors grounded in production data from Microsoft Azure. Our results show that multi-resource stranding materially changes deployable capacity, effective capital expenditure, and delivered performance, and quantify how rising density from rack- and pod-scale AI systems shapes these outcomes. For AI datacenter design, the relevant planning objective is not installed megawatts, but deployable capacity over time.

Grant Wilkins, Srinivasan Keshav, Richard Mortier

Both the training and use of Large Language Models (LLMs) require large amounts of energy. Their increasing popularity, therefore, raises critical concerns regarding the energy efficiency and sustainability of data centers that host them. This paper addresses the challenge of reducing energy consumption in data centers running LLMs. We propose a hybrid data center model that uses a cost-based scheduling framework to dynamically allocate LLM tasks across hardware accelerators that differ in their energy efficiencies and computational capabilities. Specifically, our workload-aware strategy determines whether tasks are processed on energy-efficient processors or high-performance GPUs based on the number of input and output tokens in a query. Our analysis of a representative LLM dataset, finds that this hybrid strategy can reduce CPU+GPU energy consumption by 7.5% compared to a workload-unaware baseline.