From HPC-as-a-Service to Deep Learning-as-a-Service
SessionData Analytics
Presenter
Event Type
Exhibitor Forum
TimeTuesday, November 14th4pm -
4:30pm
Location501-502
DescriptionHigh Performance Computing (HPC) and Deep Learning (DL)
share many characteristics: intensive computing, large
datasets, and need for easy access to clustered
resources by end-users. We predict that DL usage
generalization will boost HPC market growth. Conversely,
HPC community experience and the large installed base of
HPC infrastructures will boost DL growth. Both HPC and
DL are evolving together to the as-a-service model by
reusing and adapting matured HPC and Cloud concepts such
as massive scalability, resource managers, containers,
orchestration mechanisms, GPU Computing, batch
schedulers, HPC-as-a-Service software, HTTP
RESTful-APIs, and web user interfaces.
Data scientists need a high-level DL API and a web user interface that both hide HPC systems’ complexity. We will illustrate our move from HPCaaS to DLaaS by showing how we manage DL training tasks and frameworks on standard HPC clusters through a web user interface. Pros and cons of possible architecture choices will be discussed.
Data scientists need a high-level DL API and a web user interface that both hide HPC systems’ complexity. We will illustrate our move from HPCaaS to DLaaS by showing how we manage DL training tasks and frameworks on standard HPC clusters through a web user interface. Pros and cons of possible architecture choices will be discussed.
Presenter
