通过 Grace-Blackwell 架构上支持 GPU 的查询引擎追踪行的路径.pdf

1、1Tracing the Path of a Row through a GPU-enabled Query Engine on the Grace-Blackwell ArchitectureClemens Lutz,Senior Developer Technology EngineerThomas Graves,Principal Systems Software EngineerDatabricks Data+AI Summit 20252Forward-looking StatementThis presentation has been prepared for informati

2、onal purposes only.The information set forth herein does not purport to be complete or contain all relevant information.Statements contained herein are made as of the date of this presentation unless stated otherwise.This presentation and the accompanying oral commentary may contain forward-looking

3、statements.In some cases,forward-looking statements can be identified by terms such as“may”,“will”,“should”,“expects”,“plans”,“anticipates”,“could”,“intends”,“projects”,“believes”,“estimates”,“predicts”,or“continue”,or the negative of these words or other similar terms or expressions that concern Da

4、tabricks expectations,strategy,plans,or intentions.Forward-looking statements are based on information available at the time those statements are made and are inherently subject to risks and uncertainties that could cause actual results to differ materially from those expressed in or suggested by th

5、e forward-looking statements.Forward-looking statements should not be read as a guarantee of future performance or outcomes.Except as required by law,Databricks does not undertake any obligation to publicly update or revise any forward-looking statement,whether as a result of new information,future

6、developments or otherwise.3Data Processing with GPUsPremiseGPU MemGPU MemGPUData=TBsHow can GPUs scale to large data?Data parallel TB/s memory bandwidth TFLOPS compute throughput4GPU MemGPU MemGPUData Processing with GPUsTypical GPU-enabled DBMS ArchitectureCPU MemCPU MemCPUstoread hoc transfer proc