A GPU performance estimation model based on micro-benchmarks and black-box kernel profiling

Abstract

Over the last decade GPUs have been established in the High Performance Computing sector as compute accelerators. The primary characteristics that justify this modern trend are the exceptionally high compute throughput and the remarkable power efficiency of GPUs. However, GPU performance is highly sensitive to many factors, e.g. the type of memory access patterns, branch divergence, the degree of parallelism and potential latencies. Consequently, the execution time of a kernel on a GPU is a difficult to predict measure. Unless the kernel is latency bound, a rough estimate of the execution time on a particular GPU could be provided by applying the roofline model, which is used to map the program’s operation intensity to the peak expected performance on a particular processor.Though this approach is straightforward, it cannot not provide accurate prediction results. In this thesis, after validating the roofline principle on GPUs by employing a micro-benchmark, an analytical throughput or ...
show more

All items in National Archive of Phd theses are protected by copyright.

DOI
10.12681/eadd/41390
Handle URL
http://hdl.handle.net/10442/hedi/41390
ND
41390
Alternative title
Ένα μοντέλο εκτίμησης απόδοσης επεξεργαστή γραφικών (GPU) βασισμένο σε μετροπρογράμματα και καταγραφή μετρικών με προσέγγιση «μαύρο-κουτί»
Author
Konstantinidis, Elias (Father's name: Nikolaos)
Date
2017
Degree Grantor
National and Kapodistrian University of Athens
Committee members
Κοτρώνης Ιωάννης
Μανωλάκος Ηλίας
Κοζύρης Νεκτάριος
Μισυρλής Νικόλαος
Γκιζόπουλος Δημήτριος
Σούντρης Δημήτριος
Τζαφέρης Φίλιππος
Discipline
Natural SciencesComputer and Information Sciences
Keywords
Performance model; Graphics Processing Unit; Roofline model
Country
Greece
Language
English
Description
152 σ., im., tbls., fig., ch.
Rights and terms of use
Το έργο παρέχεται υπό τους όρους της δημόσιας άδειας του νομικού προσώπου Creative Commons Corporation:
Usage statistics
VIEWS
Concern the unique Ph.D. Thesis' views for the period 07/2018 - 07/2023.
Source: Google Analytics.
ONLINE READER
Concern the online reader's opening for the period 07/2018 - 07/2023.
Source: Google Analytics.
DOWNLOADS
Concern all downloads of this Ph.D. Thesis' digital file.
Source: National Archive of Ph.D. Theses.
USERS
Concern all registered users of National Archive of Ph.D. Theses who have interacted with this Ph.D. Thesis. Mostly, it concerns downloads.
Source: National Archive of Ph.D. Theses.
Related items (based on users' visits)