DocumentCode :
652584
Title :
High-Level GPU Multi-purpose Profiler
Author :
Rotariu, Marian-Cristian ; Apostol, Elena
Author_Institution :
Fac. of Autom. Control & Comput., Univ. Politeh. of Bucharest, Bucharest, Romania
fYear :
2013
fDate :
28-30 Oct. 2013
Firstpage :
549
Lastpage :
553
Abstract :
The graphics processing units (GPUs) have become an integral part of today´s computing systems. They have risen and evolved over the last years, becoming a platform for parallel computation with a large number of scalar processors and abundant memory bandwidth. They deliver high standard computation performance, but also require a lot of power supplies and cooling systems. Its rapid increase in both programmability and capability has allowed a powerful research community to arise - they have managed to successfully map out a broad range of computationally demanding, complex problems to the GPU. This paper presents a high-level GPU multi-purpose profiler using several general methods which can analyze not only the given source code, but also the run-time binary application. These models obtain the GPU running time, memory space used and the general flow of the application and, more importantly, it provides a platform on which further new metrics can be implemented.
Keywords :
computer graphics; graphics processing units; parallel architectures; CUDA; abundant memory bandwidth; call-graph; graphics processing units; high-level GPU multi-purpose profiler; high-level GPU multipurpose profiler; run-time binary application; scalar processors; Graphics processing units; Hardware; Instruments; Kernel; Measurement; Radiation detectors; Visualization; CUDA; GPU; call-graph; profiler; threads blocks;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
P2P, Parallel, Grid, Cloud and Internet Computing (3PGCIC), 2013 Eighth International Conference on
Conference_Location :
Compiegne
Type :
conf
DOI :
10.1109/3PGCIC.2013.94
Filename :
6681288
Link To Document :
بازگشت