GTC ON-DEMAND

 
SEARCH SESSIONS
SEARCH SESSIONS

Search All
 
Refine Results:
 
Year(s)

SOCIAL MEDIA

EMAIL SUBSCRIPTION

 
 

GTC ON-DEMAND

Seismic & Geosciences
Presentation
Media
Power and Energy Prediction of Multi-GPU Kernels
Abstract:
GPUs are currently receiving great attention in the HPC community as are known to provide better performance to power ratio as compared to CPUs, for certain applications. It is not possible to measure power/energy accurately in all the cases; an alternative way would be to estimate power/energy using statistics. In this study we employ non-linear regression to estimate power and energy consumption of some common optimized high performance kernels (DGEMM, FFT, PRNG and FD stencils) on a multi-GPU platform. Using only 3 variables, we found the average error between measured and predicted values of power and energy to be ~ 5%.
 
Topics:
Seismic & Geosciences
Type:
Poster
Event:
GTC Silicon Valley
Year:
2013
Session ID:
P3264
Download:
Share: