Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.33300

How AI Load Swings Reduce Datacenter Power Conversion Efficiency

Heetarth Joshi·Ali Alenezi·Fang Chen Lin·Prasad Enjeti·Xiaoyang Wang·Xin Chen
View original source ↗Read source document ↗Export metadata ↓

Abstract

AI training and inference workloads make GPU power swing repeatedly between compute and communication phases. As a result, the cascaded converters in a datacenter rack keep moving between operating points with different efficiencies. This paper uses an energy based, cycle averaged method to quantify that effect along a 240 V AC to 400 V DC to 51 V DC rack power path, using measured efficiency curves of commercially available GaN based PFC and LLC converters. For a 1 Hz, 50% duty-cycle load swinging between 5 kW and 510 W, conversion efficiency falls to 95.97%, compared with 96.82% at the equivalent constant average load, a loss of 0.88%. When 51/48V to 1V point of load conversion to the chip rail is included, it is shown that the loss grows to 1.66% (89.45% vs. 90.96%). These results show that efficiency rated at average load overstates real world performance, and that the penalty grows with each load dependent conversion stage between the grid and the GPU.

Explore related subjects

cs.ARcs.SYeess.SY

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Heetarth Joshi, Ali Alenezi, Fang Chen Lin, Prasad Enjeti, Xiaoyang Wang, Xin Chen. 2026-10-23. How AI Load Swings Reduce Datacenter Power Conversion Efficiency. https://arxiv.org/abs/2610.33300

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

FADiff: Fusion-Aware Differentiable Optimization for DNN Scheduling on Tensor Accelerators

Efficiently deploying Deep Neural Networks (DNNs) - particularly Large Language Models (LLMs) - on tensor accelerators is essential for achieving high throughput and energy efficiency at the edge. However, the combinatorial explosion of the joint design space spanning intra-layer mapping and inter-layer fusion renders existing approaches inadequate. In this work, we present FADiff, a gradient-based optimization framework that automatically discovers optimal mapping and fusion strategies to accelerate DNN inference. We first formulate a unified, differentiable analytical cost model that accurately estimates the energy and latency of individual layer mappings as well as diverse fusion configurations. We then combine CP-SAT-based pre-enumeration of feasible discrete candidates with augmented-Lagrangian penalties over continuous relaxations, enabling a constrained gradient-based search that efficiently navigates the vast design space to identify optimal joint strategies. Experimental results show that FADiff consistently outperforms prior methods, delivering superior optimization quality with significantly faster convergence in both energy and latency.

cs.AR↗
arXiv2026

MCF: A Memory- and Communication-Aware Co-Designed Framework for Long-Context LLM Inference

This paper presents Memory- and Communication-Aware Framework for Long-Context LLM Inference (MCF) that optimizes long-context inference through communication-aware and stage-adaptive KV-cache layout transitions. At the core of MCF is {Memory-Aware Replication and Scheduling} (MARS), a dynamic orchestration mechanism for chunked prefill. MARS exploits early-stage memory headroom by selectively replicating highly reused KV blocks, improving locality and hiding KV propagation latency behind attention computation. As prefill progresses and memory pressure increases, MARS progressively drains redundant replicas and converges the KV cache into a sharded decode-ready layout, avoiding expensive layout conversion at the prefill--decode boundary. For decoding, MCF adopts a query-centric execution model in which KV shards remain stationary while lightweight query tensors are routed across workers. This changes communication from KV-cache-dependent transfer to query-dependent transfer, reducing communication growth with context length. MCF further integrates a hardware-software co-designed execution layer based on an FPGA compute-NIC datapath, enabling metadata-driven KV routing and collective operations directly in the network path. Experiments show that MCF improves TTFT, throughput, and scalability over state-of-the-art distributed inference systems, with larger benefits as context length and cluster size increase.

cs.AR↗
arXiv2026

SPARTA: A Sparse Ternary LLM Accelerator with Adaptive Attention and Unified Datapath for Long-Context Edge Inference

LLMs on edge devices are constrained by limited memory and bandwidth. Although 1.58-bit ternary weights greatly reduce model storage, long-context inference is increasingly bottlenecked by KV-cache traffic and attention. Existing accelerators also use separate datapaths for ternary linear layers and attention, incurring hardware and execution overhead. We present SPARTA, a sparse ternary LLM accelerator for efficient long-context edge inference. SPARTA integrates: (1) a phase-aware sparse dataflow exploiting heterogeneous sparsity across compute-bound prefill and memory-bound decode; (2) a unified compute fabric mapping ternary SpMM/SpMV and bit-serial attention onto shared processing elements to exploit zero-weight and zero-bit sparsity; and (3) an adaptive two-stage attention engine combining page-level Top-k selection to reduce K-cache accesses with token-level Top-p pruning to reduce V-cache traffic. SPARTA reduces KV traffic by 4.1-4.8x over full attention and 1.3-1.5x over a state-of-the-art sparse-attention scheme, with task-dependent quality loss. Compared with NVIDIA Jetson AGX Orin, SPARTA achieves 3.9x higher generation throughput and 8.3x higher end-to-end energy efficiency on average; versus AMD Ryzen 7 9700X, the gains are 5.0x and 81.7x. It also delivers 1.6-12.0x higher throughput, 2.0-4.2x higher energy efficiency, and 1.3-51.4x higher area efficiency than FIGLUT, TeLLMe, and MCBP.

cs.AR↗