[MinnowBoard] MinnowboardMAX performance on compute

[David Byte]

Your results align with what I saw from linpack on the board. 

David Byte
Sr. Technology Strategist, OEM & Integrated Systems
dbyte at suse.com

Sent from my iPhone. Typos are Apple's fault. 

> On Nov 24, 2014, at 9:21 AM, Vinayak A Gokhale <vgokhale at purdue.edu> wrote:
> 
> Hello,
> 
> I am a research assistant at a lab that does embedded machine learning. We purchased the Minnowboard MAX to evaluate its performance and compare with other embedded devices (Jetson, Odroid XU3 devboards) and my own custom arch (on a FPGA). While the algorithms are specific to machine learning, they comprise of SGEMMs (for the most part). I thought someone with similar workloads might be interested in performance numbers.
> 
> I profiled the board to consume *4.9W* at the wall when running both cores at 100% and having 1.3GB of memory used. I got a performance of *2.64 GFLOPS/s*. While the raw performance is almost half that of the other devboards I mentioned, the perf/watt is higher than all others. The passive cooling also gives the Atom chip an advantage in the embedded applications section in my opinion.
> 
> I think a quad core version has the potential to beat the other boards in both categories (raw perf and perf/watt). Can't wait!
> 
> On a side note, this makes me really happy because I was a bit disappointed seeing Intel take a back seat in this market segment but it looks like the new Atom arch is doing pretty good. Thank you to the Minnowboard team for this effort and any Intel guys reading this!
> 
> Regards,
> Vinayak A. Gokhale
> Graduate Research Assistant,
> Department of Electrical and Computer Engineering,
> Purdue University,
> West Lafayette, IN 47906
> <Part.002>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.elinux.org/pipermail/elinux-minnowboard/attachments/20141124/f4ac026a/attachment.html>