2134 - MavenBlue 如何使用 IBM Cloud 上的 GPU 无服务器集群运行大规模金融模拟.pdf

1、Orlando,FLOctober 69IBM TechXchange 2025Paul HaggMavenBlue|ArchitectHow MavenBlue runs large-scale financial simulations using Serverless Fleets with GPUs on IBM CloudJeremias WernerIBM|ArchitectSession:2134Agenda010203040506MavenBlueIBM Code Engine Live DemoHow MavenBlue uses Serverless FleetsHow t

2、o get startedQ/AIBM TechXchange|2025 IBM CorporationMavenBlueHigh performance computing in the cloudClassification:L1-Public|L2-Internal use only|L3-Confidential|L4-Restricted|L5-SensitiveMavenBlue-HistoryMAVENBLUEFounded 2016Finance and risk&Strong focus on ITWorking from a barn own GPU pizza-boxes

3、Mathematical backboneVertical scaling using GPU cardsAs of 2020:Partner of IBMInvolved in testing L4/L40S GPU cardsInvolved in fleetsMavenBlue-nowadaysMAVENBLUEFully cloud based and cloud native Deployed in IBM Cloud,hosting in EU,GB,USAdvanced parallel computing algorithms(GPUs)Modern internet arch

4、itecture stack Security by design in infrastructure and solutionsAgenda010203040506MavenBlueIBM Code EngineLive DemoHow MavenBlue uses Serverless FleetsHow to get startedQ/AIBM TechXchange|2025 IBM CorporationIBM Cloud Code EngineRun any code.Easily.At scale7IBM Cloud Code Engine is a fully-managed

5、runtime where developers can:Go live in secondsEasily scale any code even down to zeroPay only for what you useIBM TechXchange|2025 IBM CorporationWeb apps,microservices,REST APIs,HTTP servers,Single-Page-AppsRun-to-completion,bash scripts,ongoing tasksChat bot integrations,“Glue code”,Webhook targe

6、ts8FunctionsFunctionsJobsJobsAppsAppsLarge scale,computecompute-intensive workloadsintensive workloads like Monte Carlo simulations,financial risk modelling,chemical molecule calculations,AI training&fine-tuning(serverless GPUs)FleetsFleetsIBM TechXchange|2025 IBM CorporationTaskCode3.3 Decommission