不依赖GPU也能低成本部署深度学习模型，适合资源有限的团队。
Deep Learning Model Deployment in Multiple Cloud Providers: an Exploratory Study Using Low Computing Power Environments
- 用CPU缓存大小优化无GPU部署，降低计算成本
- 无GPU方案平均成本比带GPU低50%
- 适合初创公司和小规模研究团队使用
深度学习模型在云端部署日益普及，但硬件要求高且云服务成本可能成为障碍。本研究以语法纠错模型GECToR为例，在Amazon Web Services、Google Cloud Platform和Microsoft Azure三大云平台的7种运行环境中共进行10次实验，评估实时延迟、硬件使用率和成本。结果表明，虽然GPU性能优越，但其平均成本是无GPU方案的300%。分析发现，仅使用CPU时，处理器缓存容量是关键因素，缓存充足的配置可使成本比GPU方案降低50%。该研究证实了无GPU的云上深度学习推理在可行性和经济性上的优势，尤其适用于资源受限的初创企业和小型研究团队。
原文摘要 · Abstract (English)
The deployment of Machine Learning models in the cloud has grown among tech companies. Hardware requirements are higher when these models involve Deep Learning techniques, and the cloud providers' costs may be a barrier. We explore deploying Deep Learning models, using for experiments the GECToR model, a Deep Learning solution for Grammatical Error Correction, across three of the major cloud providers (Amazon Web Services, Google Cloud Platform, and Microsoft Azure). We evaluate real-time latency, hardware usage, and cost at each cloud provider in 7 execution environments with 10 experiments reproduced. We found that while Graphics Processing Units (GPUs) excel in performance, they had an average cost 300% higher than solutions without a GPU. Our analysis also suggests that processor cache memory size is a key variable for CPU-only deployments, and setups with sufficient cache achieved a 50% cost reduction compared to GPU-based deployments. This study indicates the feasibility and affordability of cloud-based Deep Learning inference solutions without a GPU, benefiting resource-constrained users such as startups and small research groups.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。