A New Kind of Data Centric Performance Portability Challenge Item
release_ppdp4wlqarhpdkdux3dpnwksle
by
Tim Germann
2021
Abstract
Data science applications, including machine learning, optimization, graph analytics, and other large-scale data-driven computations, present a unique set of challenges to performance portability. Some aspects of the very heterogeneous architectures that have recently begun to emerge, such as tensor cores, specifically target use cases such as deep learning. In that case, frameworks built upon abstractions that have emerged allow for low-level architecture-specific optimizations to be implemented and enable performance-portable codes. For other data-oriented applications, portability is a greater challenge, and widely applicable abstractions not as simple. For these, lower-level programming models and languages such as Kokkos or UPC++ have been adopted. In this talk, we will provide some exemplar application drivers, approaches, and lessons learned from 4 Data Analytics & Optimizations applications and 3 Co-Design centers within the DOE Exascale Computing Project. This is a talk presented at the SIAM CSE21 conference, MS162/192: Exascale Computing Project Performance Portability Analysis.
In text/plain
format
Archived Files and Locations
application/pdf 1.9 MB
file_3gjqsdq3hvb6zp3yxf4a6po6ne
|
s3-eu-west-1.amazonaws.com (publisher) web.archive.org (webarchive) |
article
Stage
published
Date 2021-03-03
Version
v3
access all versions, variants, and formats of this works (eg, pre-prints)
Datacite Metadata (via API)
Worldcat
wikidata.org
CORE.ac.uk
Semantic Scholar
Google Scholar