Collective Knowledge Aggregator proof-of-concept
Crowd results Raw CK browser Graphs Reports Datasets Models Home

This page is outdated! New version is available here.


Aggregated execution times of slambench across multiple platforms (crowd-benchmarking)
Grigori Fursin 1,2, Anton Lokhmotov 2
1 cTuning Foundation, France,   2 dividiti, UK
Share
Demo of aggregating results of all executions of slambench across all platforms (different CPU, GPU, frequency, compiler, etc). It can help build a realistic training set for further machine learning based autotuning and run-time adaptation as described in our papers [1, 2, 3]. We can use it to find an optimal platform for a given data set (balancing execution time, energy/frequency, accuracy, price, cost, etc - particularly useful for cloud computing or mobile devices).
CK report at cknowledge.org/repo with ranking of all devices/kernel vs FPS, energy, accuracy, etc.


Reproduce/reuse/replay/discuss via CK (interactive graphs)


   Auto-replot graph: , seconds:
View entry in raw format

Developed by Grigori Fursin           
Implemented as a CK workflow 		                         
   
                      Hosted at