H2O Machine Learning on Parallella

Posted:
Sun Dec 07, 2014 10:46 pm
by wdraham
I am looking at the Parallella to run H2O for machine learning. It's written in Java. More info at
Wondering if anyone has done this before and knows the performance of it compared to a Xeon server?
Many thanks in advance
W
Re: H2O Machine Learning on Parallella

Posted:
Mon Dec 08, 2014 12:32 pm
by 9600
I'm not aware of anyone having yet extended any existing Java apps in order to be able to take advantage of the Epiphany accelerator.
In order to benefit from Parallella's floating-point accelerator you would need to look into extending H20 accordingly (see other posts in this forum for discussions relevant to device access from Java). If you don't do this, the performance will be that of a dual-core ARM A9 clocked at 800Mhz, i.e. only the host part of the Parallella, and not the accelerator.
Regards,
Andrew