百万兆次级之后HPC面临的挑战.pdf

《百万兆次级之后HPC面临的挑战.pdf》由会员分享，可在线阅读，更多相关《百万兆次级之后HPC面临的挑战.pdf（6页珍藏版）》请在三个皮匠报告上搜索。

1、Estela Surez-(Jlich Supercomputing Centre)Erik Lindahl-(KTH Royal Institute of Technology)Jean-Yves Berthou-(INRIA Saclay)Fabrizio Del Maffeo-(Axelera AI)Sergi Girona-(Barcelona Supercomputing Centre)Leonardo Flores Aover(EC)-ModeratorPLENARY:Challenges in HPC after ExascaleChallengesChallengesin HP

2、C after Exascalein HPC after Exascalehttps:/globalenergymonitor.org/projects/global-coal-plant-tracker/Justify use of energy andnatural resourcesPackageInterposerEven more HW heterogeneityLack of performanceportabilityEstela Suarez18.03.2024,AntwerpenTechnology dependencyhttps:/ vs.HyperscalarsLogos

3、 owned by respective companiesGraphic owned by Disney/ImgbinHPCImpact:Data&Software is the key frontierChallenges in HPC Erik Lindahl,KTH Royal Institute of TechnologyClimate modeling&weather predictionMaterial science:H2 production,solar cellsLife science:Pharmaceutical design&whole-cell modelingWe

4、 need Manhattan-scale projects focused on solving societal problemsEngineering,CFD,Data analytics5/28/2024Some post-exascale challengesFrom edge to HPC systemsThe digital continuumAI4Science Science4AISoftware/application co-designSoftware,the new frontierInternational collaboration is key-The Inter

5、national Post-Exascale project(InPEx)Need for a proper European agenda InPEx-EUAxelera AI-Confidential1)SCALING COMPUTING POWERClassical computing doesnt allow to scale easily to Zettascale,we need an hybrid approach with classical computing and AI2)WORKING AT LOWER COMPUTING PRECISIONLower precisio

6、n computing(FP32/16/8)and machine learning to solve classical AI workload3)EU SOVEREIGNTY IN COMPUTING(HPC/AI)It requires a very ambitious plan and a strong sense of urgencyLisa Su,Plenary Session,ISSCC 2023,FebCHALLENGES TO BE ADDRESSED FO