DumpsterCluster:用 60 美元 GPU 服务 LLaMA-70B 的二手硬件集群研究
原标题:DumpsterCluster: From Dumpster Diving to Serving LLaMA-70B on $60 GPUs
AI 摘要
本文介绍了一个名为 DumpsterCluster 的 128-GPU 集群,完全由二手组件构建,用于服务 LLaMA-70B 等现代 LLM 推理。该集群成本仅为 2.2 万美元,远低于 8-GPU B200 系统的 60 万美元,并通过流水线并行优化实现了有竞争力的吞吐量。然而,研究发现二手 GPU 的能效较低,总拥有成本仅在电价低廉的地区有利,且碳排放强度较高,因此硬件再利用必须与清洁能源结合才能实现可持续性。
正文节选
Abstract As AI datacenters retire functional GPUs, vast quantities of still capable accelerators enter secondary markets. This paper investigates whether these “retired” GPUs can find a productive afterlife to form a DumpsterCluster that can serve modern LLM inference, and under what conditions such repurposing is economically viable and environmentally sustainable. We physically built a 128-GPU DumpsterCluster from scratch using only second-hand components and ran it for one year. At current ma